On August 26, OpenAI announced that 55 additional U.S. school systems across 20 states are adopting ChatGPT for Teachers, bringing the tool to more than 100 K–12 organizations across 30 states and more than 300,000 educators and staff receiving free access and training.

OpenAI states that “1 in 5 of America’s 20 largest public school districts” are now working with them. As far as we can tell, this is referring to one of the new partner districts and three that were already in the OpenAI network:

• Orange County Public Schools (Florida) (New!)

• Houston Independent School District (Texas)

• Fairfax County Public Schools (Virginia)

• Dallas Independent School District (Texas)

Arguably, the more strategically relevant part of the launch is around procurement and privacy: OpenAI simultaneously announced a 16-state National Data Privacy Agreement, which gives districts a common and accepted pathway for districts to evaluate ChatGPT’s privacy lens, a hot button issue for LLMs in educational settings, rather than negotiating privacy terms independently.

The agreement covers Illinois, Iowa, Maine, Massachusetts, Missouri, Nebraska, New Hampshire, New Jersey, New York, Ohio, Rhode Island, Tennessee, Texas, Vermont, Virginia, and Washington; California, which is its own massive education system, has a separate agreement.

Why It Matters

• By our calculations, the states listed in the Privacy Agreement represent 51.4% of all U.S. public-school students. Standardized multi-state privacy agreements could become an important distribution advantage for OpenAI, and they are clearly establishing beachhead districts in each state

• At a moment of intense scrutiny and pushback on the role of “Big Tech” in schools, OpenAI’s education team has not shied away from K-12 sector. In fact, they’ve been heads down, pounding the pavement doing “AI Skills Jams” all over the country with teachers, developing privacy strategies to shorten procurement cycles, and expanding to states with a variety of political backgrounds.

Worth Noting

OpenAI’s release post features a new video featuring the Lynwood, CA, district, which is run by superintendent Patrick Gittisriboongul, Ed.D, a well-known AI advocate who has spoken out about AI usage for years, often through Amanda Bickerstaff’s “AI for Education”. Expect him to continue to be a superintendent “face of AI” for the foreseeable future.

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Google Classroom’s Approach to Educator-Led AI

Sponsored by Google

By: Brit Mennuti

Brit Mennuti is a Product Lead on Google for Education, where she looks after the team that develops and maintains Google Classroom.

I began my career in EdTech nearly a decade ago as a product manager on Google’s Augmented and Virtual Reality team. I was responsible for developing a tool that allowed teachers to take their classes on virtual reality field trips. Imagine a room of third graders all exploring the moon, or a high school class in an agricultural region of the US visiting historic Machu Picchu from their desks. I remember joining a real class in person for the first time to see the “field trips” in action, and my mind was blown— not just by how incredibly engaged the class was with the lesson, but by how our technology empowered the teacher to ground the student experience. While students felt like they were a part of the Milky Way Galaxy, their teacher was at the front of the room acting as a guide on their virtual tour, directing their attention to details and comprehension questions, and bringing them literally and figuratively back down to Earth. Both the teacher and the students were enamored with the tech because it actually accomplished what we were there to do: drive real student learning and engagement. It was at that moment that I knew I wanted to be in EdTech for the rest of my career—to have this profound impact on learning outcomes and empower educators with technology that is a force multiplier on the impact they already have today. And it was also at this moment that I learned the power of having the teacher in the lead, regardless of the technology.

Now, as the product lead for Google Classroom and a new mother, I find myself in a similar moment. There has been no shortage of AI tools released for educators and students, including from the product teams at Google. Many people ask me, “aren’t you worried about AI in education?” My answer always brings me back to those days working on virtual reality for schools. “Yes, I am worried, as any parent would be, but I know we will do it the right way as long as we empower educators to be in the driver’s seat.” With the rapid pace of change and the overwhelm that educators, schools, students, and families are feeling, this philosophy is more important than ever. Putting educators in the lead from design to implementation is critical to ensuring that the impact of AI in education is positive. At Google, we take this responsibility seriously. As we build AI features into Google Classroom and Workspace for Education more broadly, our approach stays centered on educators as the AI landscape continues to change and shift rapidly.

What does this mean in practice? It means tasking our best researchers with ensuring education use cases in our AI products truly shine. It means co-designing solutions alongside educators by getting regular feedback and iterating before our solutions go to market. It means vetting our ideas with experts, and holding ourselves to a very high bar for safety, bias, and quality when any AI experience goes in front of a student. It means running scientific research on our solutions with real classrooms globally to understand and measure their impact on learning outcomes. And it means giving educators and education leaders control over and visibility into the AI engagement of their students.

Our work on integrating Gemini’s Guided Learning mode into an educator-led experience inside Google Classroom is perhaps the best window into this collaborative, educator-first process of integrating AI tools into Google Classroom, and therefore classrooms around the world.

What is Guided Learning?

Guided Learning is an experience within Gemini that is designed to act as a supportive learning partner—think “encouraging the productive struggle of learning while helping you get to the answer,” rather than “giving you the answer.” The goal of Guided Learning is to encourage active participation, meeting students where they are by breaking down complex problems step-by-step and guiding them with probing, open-ended questions.

As an example, let’s say a student copies a homework problem into Gemini while in Guided Learning mode. Instead of solving the problem for the student, Gemini engages them in a conversation to break down the problem step by step until they learn how to solve it themselves and develop the metacognition to solve similar future problems on their own.

In June of this year, we announced that we are bringing this Guided Learning experience directly into Google Classroom for the upcoming Back to School season. This feature will be available this fall only for schools where administrators have enabled Gemini for their educators and students. Through this integration, educators will have the ability to assign a conversation in Gemini to their students that is locked into Guided Learning mode and grounded in their specified course materials, all while ensuring data is private and secure. Think of it as a learning support alongside their assigned work that is well versed in the student’s curriculum. As students engage with Gemini, their educators will get individual and class level insights into that engagement, including areas of strength and development, misconceptions, and access to direct chat transcripts for further detail in a secure environment. In this way we ensure the student experience with Gemini is one that encourages critical thinking, while the educator remains in the lead, not only when assigning the experience, but also after the fact as we equip them with rich insights to inform the next steps in their instruction.

Grounding in Learning Science and Measuring Impact in Real Classrooms

So what about our process of bringing Guided Learning to Classroom is particularly relevant and novel for this moment in technology? Well, it starts with learning science. Before developing Guided Learning, it was important to us, first and foremost, that the Gemini model itself was the best model for learning. Google DeepMind established a cross-disciplinary team of AI experts, neuroscientists, and cognitive scientists to answer questions like: How can we design an experience that prioritizes deep understanding and exploration of topics? How can we build experiences that honor the effort of learning while encouraging active understanding? With this group of experts we developed LearnLM, a version of the Gemini model trained specifically to behave in a pedagogical manner (i.e., encourage metacognition and stimulate curiosity). We then tested LearnLM against Gemini and other models in the market to evaluate how effectively the model exhibited pedagogical behavior. The result, after blind testing by experienced educators, was that LearnLM was the best option across all five learning science principles, leading us to merge the LearnLM model into all Gemini models starting in May 2025.

But proving the Gemini model was better at pedagogy was not where we stopped. Ultimately, we hold ourselves responsible for ensuring our tools developed on these models (like Guided Learning) lead to measurable learning impact. We recently partnered with Fab AI to conduct a pre-registered randomized controlled trial (RCT) with 1,763 students from junior secondary school (Grades 7 and 8) in Sierra Leone. The results demonstrate that students assigned to study with Gemini’s Guided Learning mode over the 8-week trial showed significantly improved learning outcomes, receiving a large boost—equivalent to moving a student from the 50th to the 59th percentile in mathematics. By standard conversion, that gain reflects roughly 1.2 to 1.7 years of extra learning progress in just 8 weeks.

Most important to note here is that in this study educators were given license to incorporate Guided Learning how they saw fit. They were advised to incorporate Guided Learning mode into their curriculum but had full control over how and what they did, and the amount of time students spent with it. Educators were in the lead.

The Power of Keeping the Educator in the Lead

When we looked at the data from Sierra Leone, a clear pattern emerged: AI in education cannot be successful without deep classroom integration in partnership with the educator. While Guided Learning is a strong learning tool alone, its impact is unlocked when an educator is actively involved. We realized that integrating Guided Learning into Google Classroom could help educators provide personalized, 1:1 guidance to students at scale—something that many educators have told us they wish they could do, but that remained a hurdle with class sizes, resource constraints, and diverse student needs. By keeping educators in the lead, they can do what they do best: determine how to support their students, including enabling technology that is relevant, helpful, and grounded in their course materials. And most importantly, they will know when using Guided Learning that there is real scientific evidence to prove its impact on learning.

With this evidence in hand, we started our work bringing Guided Learning into Classroom by sharing early concepts and designs with real educators. They provided feedback on their expectations for such a feature, the levers of control they wanted to have, and the types of insights they’d need to effectively incorporate Guided Learning into their educational practice. We used that early feedback to shape what we built in close collaboration with the Gemini team and Google DeepMind. One of the key insights we learned through this process was that the socratic nature of Guided Learning had to be different in the school setting compared to the regular consumer context. We updated how Guided Learning works specifically in Classroom to ensure it met the socratic expectations of real educators. Once we had a prototype built, we started piloting it with real schools all over the globe, from the UK to Australia. In these pilots, educators have access to the feature in Classroom and can assign Guided Learning to their students. We regularly stay in touch with them and get their input on what is working and what needs to change. We are using this feedback to fine tune the experience before it is available for all schools this fall and will continue to iterate as we get feedback from educators. Our entire process has been educator-led from day one.

An Educator-Led AI Future

As I think back to my early days helping educators incorporate nascent technology into the classroom, I am reminded of the fact that the recipe for success for all student learning has always been to let educators call the shots. This approach is not new because of the “AI wave”—it has been our mode of operation since Google started building technology for schools 20 years ago (when we quite literally traveled on a school bus from school to school understanding educator needs to inform what we built). By bringing Workspace, like Google Docs, into Classroom in ways that amplified educator impact, rather than taking away from it, we freed up educators to do what they do best. And now we are applying this same playbook to how we safely and responsibly build AI for schools.

As the mother to a one-and-a-half-year-old, I think daily about what my daughter’s future will be like. It is not lost on me that my influence, and the influence of the incredibly talented team of Googlers around me, will have a direct impact on how she learns. And that is a serious responsibility! But it is the community of educators we have the privilege of working with daily that gives me hope for her future, and the future of children all over the globe. While lately conversations about the future focus on technology, I look forward to keeping the conversation firmly focused on the people at the forefront of leveraging that technology for their students.

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Insider Announcements

Hey Insiders! We’ve partnered with the team at StartEd to offer you all 10% off your ticket to EdTech Week 2026! We’ll be there, and we know many of you will be too.

This code is active for all ticket types, including 10% off the already discounted early bird pricing through August 31. Grab your ticket now before the price goes up!

Get Your Ticket!

Upcoming Events

What Is the Future of Higher Education in America?

Sponsored by Masterclass and Studiosity

Friday, August 28
9:00-10:00 AM PT / Noon-1:00 PM ET
Webinar on Zoom

AI is reshaping learning, employers are rethinking credentials, and universities are facing unprecedented change. Join us for a live conversation with an incredible panel of experts exploring how higher education must evolve to remain relevant, accessible, and valuable in the decade ahead.

Join us live or register to receive the recording the following Wednesday.

RSVP Here

Top Edtech Headlines

1. Bringing ChatGPT for Teachers to more U.S. school districts

OpenAI is expanding ChatGPT for Teachers to 55 additional school systems across 20 states, bringing the program to more than 300,000 educators and staff across 100+ K–12 organizations. The expansion includes free access through June 2028, new privacy agreements covering 16 states, and hands-on training for educators and district leaders, as OpenAI positions teacher-led implementation, data privacy, and responsible adoption as central to bringing AI into schools at scale.

Learn more here.

2. Harvard Is Selling a $699 Course Taught by A.I. Clones of Its Faculty

Harvard Business School is using AI-generated video avatars of its professors to teach a new $699 entrepreneurship bootcamp. The experiment raises broader questions for higher education about whether AI avatars can extend the reach of expert instructors, what students are actually paying for when they enroll in courses, and how universities will balance AI-generated instruction with the value of human faculty.

Learn more here.

3. Meta to Pay Up to $17.1 Billion in Landmark Settlement Over Social Media Addiction Claims

Meta has agreed to pay up to $17.1 billion and make sweeping changes to how children and teens use Facebook and Instagram as part of a landmark settlement with states over allegations that the platforms were designed to addict young users. The agreement includes daily time limits, nighttime blocks, stronger age-assurance measures, and other protections for minors, marking a major development in the growing push to hold social media companies accountable for their impact on young people.

Learn more here.

4. Report: MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training

MIT’s new report is a cumulative look at how they are thinking about and approaching AI within their Institute, and a great case study for how AI is impacting Higher Ed and what needs to be done to move forward with intention and efficacy.

Learn more here.

Why Edtech Needs Less Hype and More Honesty with Tony Wan & Betsy Corcoran

Tony Wan is Head of Platform at Reach Capital and co-founder and former managing editor of EdSurge. Betsy Corcoran is a writer, speaker, serial entrepreneur, and co-founder of EdSurge who explores AI and learning through her work and podcast, Future Fluent.

5 Things You’ll Learn in This Episode

1. How EdSurge helped shape the early edtech ecosystem.

2. Why edtech needs more honest and nuanced storytelling.

3. What the tech backlash gets right and where it misses the bigger picture.

4. Why teacher-turned-founders and school leaders are critical voices in edtech.

5. How AI can support education without becoming the focus itself.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

This Friday, we’re bringing together leaders from across higher education and industry for a candid conversation about where higher education is headed.

AI is changing how students learn.
Expectations around the value of a degree are shifting.
Universities are navigating new enrollment, workforce, and economic realities.

So what does the next chapter of higher education look like?

We’ll explore:

• How AI is changing teaching, learning, and the university experience

• Whether traditional degrees will remain the dominant pathway to opportunity

• What institutions must do to adapt to changing student, workforce, and economic realities

Join us live, or register to receive the recording the Wednesday following the event.

Register Now

See you there!

Best,
Edtech Insiders

^{A quick note on email preferences: We’re actively working to reduce email fatigue for our audience. We’ve recently created a dedicated Events Announcements section on Substack so you can control what you receive. If you prefer not to get webinar/event updates, you can unsubscribe from just the “Announcements” section, without leaving the Edtech Insiders newsletter itself. (You’ll still receive our regular news, analysis, and podcast updates.)}

Homework Helper or AI Tutor? The Difference Determines What Students Learn.

Sponsored by Overdeck Family Foundation

By: Lin Ler

Lin Ler is a Stanford MBA and MA in Education candidate working at the intersection of business strategy, learning science, and frontier AI. With a background spanning management consulting, EdTech startups, and the social sector, Lin translates complex AI capabilities into practical insight for the future of learning.

Of all the topics raised since Generative AI hit education, the concept of “AI tutoring” may be what has most thoroughly grabbed the public imagination. It’s also one of the hardest to define. The “AI tutoring” label can plausibly be applied to a very diverse set of learning experiences:

• Dedicated edtech “AI tutoring” products (e.g., Synthesis Tutor, SigIQ, Thinkverse, Tutor.ai)

• “AI tutoring” features within larger learning suites (e.g., Ace by Numerade, Khanmigo by Khan Academy, StudyFetch’s AI Tutor)

• Mobile applications designed for on-demand homework help (e.g., Gauth, Photomath, Brainly Homework Helper, Zuoyebang, QANDA)

• “AI companions” designed for learning conversations, often in language learning (e.g., TalkPal, Superhuman, Tutor Lily, Praktika, ELSA Speak)

• AI systems designed to improve human tutoring and/or match students to tutors (e.g., Hilink, Sidekick from Step Up Tutoring, TutorOcean)

• Frontier large language models placed in “study mode” (e.g., Google Guided Learning, ChatGPT Study Mode, Claude’s Learning Mode)

When the industry asks whether “AI tutoring” works, it assumes researchers are talking about one thing when referring to “AI tutoring.” When Edtech Insiders spoke with Susanna Loeb, the Kissick Family Professor and faculty director of Stanford’s SCALE Initiative, about the potential impact of AI tutoring, she clarified an important distinction: “Are we talking about homework help, or a high-impact tutor?”

That difference matters. A LOT. Homework helpers (and off-the-shelf LLMs) are designed to resolve the task a user brings in real time: to get to an answer. Great tutors don’t provide answers. They build trusted relationships, decide what each learner needs, preserve the student’s responsibility to think and practice, and help learners make progress over time through a larger course of learning.

AI Tutoring Is Not a Monolith

A new SCALE brief from Stanford offers a useful route through this definitional problem. Instead of sorting tutoring models by how sophisticated their AI appears, AI Tutoring Is Not a Monolith places them on a spectrum of relational intensity: the depth and consistency of the human connection surrounding the student.

At the high-intensity end, a human tutor leads the instruction and holds the relationship, while AI may assist with lesson preparation, data analysis, or suggestions. In the middle, the student works directly with an AI tutor while a person oversees the process and intervenes. At the lowest-intensity, the student works with an AI-only tutor without direct human oversight.

This framework does not assume that more human contact always makes a tool better (ex: A well-designed hybrid can use AI to remove administrative work and give a human tutor more time for the relationship). It also doesn’t assume that every student needs the same degree of support.

It does show where the evidence is strongest: High-impact tutoring is still defined by live, human-led instruction. The evidence becomes thinner as direct human involvement decreases. At present, SCALE concludes that AI-led tutors may offer valuable supplemental practice, but do not yet inherit the definition or evidence base of high-impact tutoring simply because they provide one-to-one responses.

That evidence base is unusually deep. In a 2026 synthesis, Loeb and Carly Robinson report that meta-analyses estimate average effects of 0.29 to 0.37 standard deviations on student achievement, making tutoring one of the most effective interventions in education research. The strongest programs also tend to share a recognizable architecture:

1. Students and tutors meet frequently, 1:1 or in small groups

2. Tutoring connects to classroom curriculum

3. Tutors use evidence of student understanding to adjust instruction

4. The same tutor works with the student over time

5. Tutors receive training, materials, and ongoing support

These features could conceivably be applied to AI-enhanced tutoring, and many of the edtech companies designing tutoring products already use several of these principles. Human tutors improve through training and coaching. AI tutors need their own form of pedagogical preparation: structured instructional goals, curriculum grounding, examples of productive tutor moves, constraints against answer-giving, and continuous evaluation of how they respond to actual student thinking. The foundation model is only one component of the tutor, and often one that must be overridden to get to deeper learning-oriented behavior.

These features explain why tutoring works, not merely that it works. They also give us a better test for any AI tutor: Which conditions does the product reproduce, which does it change, and which does it drop?

Consider consistency: an AI system is technically available at any hour and can respond with the same patience each time. But availability is not the same as learners actually showing up, persisting through difficulty, or trusting the person asking them to try again.

“Most of that kind of engagement is going to come from the context and from the humans that [students] interact with,” Loeb told us. “It’s that human connection that really inspires people to do things that are difficult.”

A new SCALE study makes the point starkly. Across two randomized trials involving 355 elementary students, researchers gave students scheduled access to an AI literacy platform, either independently or with an in-person tutor focused on engagement rather than instruction. In the independent group, nearly half never used the platform. Those who did averaged only two to five minutes a week. Human support increased engagement by 71% to 80%, but use remained low and reading achievement did not improve. The human helped make the learning routine more real, and even that was not enough to create the dosage associated with gains.

The Homework Helper Test

Relational intensity is likely only one part of the answer. Most educators and edtech designers agree that any student-facing system still needs to behave like a tutor in some key ways to be effective. But the exact definition of what makes tutoring effective, what the National Tutoring Observatory calls effective “tutor moves”, are still very much being defined.

In a randomized field experiment with nearly 1,000 high-school mathematics students, access to a general GPT-4 interface increased students’ practice performance by 48%. A guardrailed “GPT Tutor,” supplied with teacher-written solutions and instructed to give hints, increased it by 127%.

Then, the AI was removed. Students who had used the general chatbot scored 17% lower than the control group on an unaided test. Students who had used the guardrailed tutor performed about the same as the control group—not worse, but not better.

This is the cognitive-debt problem described in the first article in this series: a system can improve what a student produces now without building the capacity they need later. This is the opposite of the role of a true tutor. The guardrails prevented the harm seen with unrestricted help, but they did not automatically create a learning gain.

Loeb adds another difference. Homework help is reactive: the student must know what to ask. A tutor needs a view of the curriculum and the learner’s position within it. “Students don’t always know what they don’t know,” she said. They may also not know “what’s next in the curriculum.” A high-impact tutor therefore does not only answer the incoming question. It helps determine the next problem, explanation, or practice the student needs.

A useful working definition follows:

A student-facing AI tutor pursues a defined learning objective, elicits evidence of the student’s understanding, adapts its next move, and keeps the student responsible for the thinking.

That is a much higher bar than generating a fluent explanation. It’s also a bar that the wider AI-in-education ecosystem is beginning to make measurable.

Building the Ecosystem for Better AI Tutors

SCALE’s role is increasingly to help guide this ecosystem: translating the established evidence on tutoring into design principles, testing new implementations, and maintaining a repository where emerging AI studies can be compared. The field is not converging on a single winning product. It is building several layers of research, infrastructure, and evaluation at once.

Products Are Changing the Interaction

Edtech developers are experimenting with structured, curriculum-aligned tutors rather than open chat boxes. Rori, a WhatsApp mathematics tutor, produced an estimated 0.36-standard-deviation gain in an experiment across 11 Ghanaian schools, although the small number of schools and student attrition make the estimate less certain. In the United Kingdom, Eedi and Google tested LearnLM inside a constrained mathematics platform with human tutors supervising AI-drafted messages. Meanwhile, frontier labs have introduced ChatGPT Study Mode, Claude Learning Mode, and Gemini Guided Learning, each designed to ask questions, scaffold steps, and delay the immediate answer. These are meaningful design shifts, although a product’s stated pedagogy is not yet evidence of durable learning.

Benchmarks as Definitions of Effective Tutor Behavior…

Research labs all over the world have been working on a variety of tutoring and education benchmarks to evaluate AI models, all with their own specific theories about what makes tutoring effective.

• TutorBench gives a model 1,490 difficult high-school and AP STEM scenarios and tests three practical capabilities: can it adapt an explanation to a student’s confusion, give actionable feedback on the work shown, and offer a hint that keeps the learner active? The best of 16 frontier models scored about 56 out of 100. By TutorBench’s definition, frontier models have lots of room for improvement, which is exactly where Edtech can take the baton and run with it.

• MRBench is another benchmark that evaluates tutor responses across 8 pedagogical dimensions such as ‘coherence’, ‘tutor tone’, ‘mistake identification’ and of course, ‘revealing of the answer’.

• One interesting approach is to define effective tutor behavior by subject, like MathTutorBench and CSTutorBench for math and computer science, respectively.

• Newer benchmarks are already pushing beyond one response. EduClaw-Bench places tutor agents in a simulated 30-day learning relationship and finds that almost no model-and-agent combination sustains strong tutoring over the full period.

• In an interesting benchmark-like test by development firm Comprendo, six frontier models were asked to hold multi-turn conversations with simulated eighth-graders who were designed to hold common algebra misconceptions. When the LLMs were given generic instructions to be helpful, the models gave away the answer in 97% of conversations. A “tutoring” prompt improved performance substantially, but even then, no model was consistently strong at both identifying misconceptions and withholding enough of the reasoning for the student to do the work. This was a simulation, not evidence that 97% of classroom interactions fail. Its value is diagnostic: it shows that a model trained to be helpful will usually optimize for resolving the problem. Tutoring requires a different objective.

… and Shared Infrastructure Can Make Those Standards Reusable

• Learning Commons is building open knowledge graphs and evaluators that connect AI outputs to curriculum, standards, and learning-science rubrics.

• Digital Promise’s $26 million K–12 AI Infrastructure Program is funding open datasets, benchmarks, and models. Its first grants include a benchmark for identifying science misconceptions, simulated student models, a National Tutoring Observatory project on educational speech recognition, and KB-TutorBench for multimodal formative assessment.

These layers serve different purposes. A benchmark can show whether a model makes a plausible tutoring move. It cannot show that a child learned. A randomized trial can estimate learning under one implementation, but it may be obsolete when the model or interface changes. Shared infrastructure allows many developers to test for the same educational properties, while repeated classroom research establishes whether those properties matter in practice.

The Future of AI Tutoring

The near-term case for AI tutoring is not necessarily that an AI chatbot will reproduce or surpass a great human tutor. It is that the field is truly beginning to specify what the label tutor requires.

AI tutors should know what a student is meant to learn, identify why they are stuck, choose an appropriate next move, and most importantly, organize the student’s behavior to optimize for learning, not performance. The surrounding system must also create the consistency, trust, and motivation that get students to return. Sometimes AI may deliver the practice directly. Sometimes its highest-value role will be helping a human tutor support learners more consistently.

Loeb’s optimism rests on that wider possibility: “We’re so far from the ideal that there are lots of opportunities to do this better,” she said. AI can theoretically help schools differentiate instruction, give students more room to explore their curiosity, and support greater autonomy — but education tools and products will need to focus on “the process of learning instead of the ultimate production.”

“We know something’s not working right,” Loeb concluded, “but we’re optimistic that it could work a lot better.”

Bibliography

Anthropic. (2025, April 2). Introducing Claude for Education. https://www.anthropic.com/news/introducing-claude-for-education

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26). https://doi.org/10.1073/pnas.2422633122

Comprendo. (2026, August 10). Can AI tutors actually teach? We measured it. https://comprendo.dev/insights/can-ai-tutors-teach

Digital Promise. (2026, June 29). Digital Promise announces first grantees of the K–12 AI Infrastructure Program. https://digitalpromise.org/2026/06/29/digital-promise-announces-first-grantees-of-the-k-12-ai-infrastructure-program/

Google. (2025, August 6). Guided Learning in Gemini: From answers to understanding. https://blog.google/products-and-platforms/products/education/guided-learning/

Hashmi, S. F. A., & Rebello, N. S. (2026). A bottom-up taxonomy of student discourse with a Socratic AI physics tutor. arXiv. https://arxiv.org/abs/2608.07373

Henkel, O., Horne-Robinson, H., Kozhakhmetova, N., & Lee, A. (2024). Effective and scalable math support: Experimental evidence on the impact of an AI-math tutor in Ghana. arXiv. https://arxiv.org/abs/2402.09809

Lane, H. C., & Kageler, B. (2026). CSTutorBench: Benchmarking small language models as tutors for block-based programming. arXiv. https://arxiv.org/abs/2607.05571

LearnLM Team & Eedi. (2025). AI tutoring can safely and effectively support students: An exploratory RCT in UK classrooms. arXiv. https://arxiv.org/abs/2512.23633

Learning Commons. (2026). AI developer tools for education.

https://docs.learningcommons.org/

Lee, U., Lee, S., Jeong, Y., Lee, E., Shin, M., & Kwon, H. (2026). EduClaw-Bench: A long-horizon benchmark for pedagogical LLM agents with simulated learners. arXiv. https://arxiv.org/abs/2608.03206

Ler, R. L. (2026, May 8). The cognitive debt problem. EdTech Insiders.

Loeb, S. (2026, August 14). Interview with EdTech Insiders on AI tutoring [Unpublished transcript].

Loeb, S., & Robinson, C. D. (2026). Tutoring. AEFP Live Handbook. https://scale.stanford.edu/publications/tutoring

Macina, J., Daheim, N., Hakimi, I., Kapur, M., Gurevych, I., & Sachan, M. (2025). MathTutorBench: A benchmark for measuring open-ended pedagogical capabilities of LLM tutors. arXiv. https://arxiv.org/abs/2502.18940

Maurya, K. K., Srivatsa, K. V. A., Petukhova, K., & Kochmar, E. (2024). Unifying AI tutor evaluation: An evaluation taxonomy for pedagogical ability assessment of LLM-powered AI tutors. arXiv. https://arxiv.org/abs/2412.09416

National Tutoring Observatory. (n.d.). National Tutoring Observatory.

https://nationaltutoringobservatory.org/

OpenAI. (2025, July 29). Introducing study mode. https://openai.com/index/chatgpt-study-mode/

Robinson, C. D., Gormley, D., Ribeiro, A. T., & Loeb, S. (2026). Access is not enough: Human support improves engagement with AI tutoring. EdWorkingPapers. https://doi.org/10.26300/pz7p-p388

SCALE Initiative. (2026). Research study repository. https://scale.stanford.edu/ai/repository

Srinivasa, R. S., Che, Z., Zhang, C. B. C., et al. (2025). TutorBench: A benchmark to assess tutoring capabilities of large language models. arXiv. https://arxiv.org/abs/2510.02663

Turano, C., Pihl, V., Agnew, C., Ziegler, L., & Loeb, S. (2026). AI tutoring is not a monolith: What we actually know. SCALE Initiative at Stanford University. https://scale.stanford.edu/research-in-action/ai-tutoring-not-monolith

Special thanks to Overdeck Family Foundation for sponsoring this FINAL article in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative).

You can support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Insider Announcements

Hey Insiders! We’ve partnered with the team at StartEd to offer you all 10% off your ticket to EdTech Week 2026! We’ll be there, and we know lot’s of you will be too.

This code is active for all ticket types, including 10% off the already discounted earlybird pricing through August 31. Grab your ticket now before the price goes up!

Get Your Ticket!

Upcoming Events

What Is the Future of Higher Education in America?

Sponsored by Masterclass

Friday, August 28
9:00-10:00 AM PT / Noon-1:00 PM ET
Webinar on Zoom

AI is reshaping learning, employers are rethinking credentials, and universities are facing unprecedented change. Join us for a live conversation with an incredible panel of experts exploring how higher education must evolve to remain relevant, accessible, and valuable in the decade ahead.

Join us live or register to receive the recording the following Wednesday.

RSVP Here

Top Edtech Headlines

1. U.S. Department of Education Releases Guidance on Responsible Use of Education Technology in the Classroom

The U.S. Department of Education has released new guidance urging states, districts, educators, and edtech providers to prioritize instructional value, evidence, educator judgment, transparency, and student outcomes when selecting and using technology in classrooms. The guidance specifically calls on schools to evaluate what learning problem a tool solves, who should use it, for how long, and what evidence demonstrates that it improves learning, while encouraging districts to remove technologies that repeatedly fail to deliver results.

Learn more here.

2. Does AI Stop Children From Learning?

A new analysis from The Economist examines research tracking 27,000 students ages 12–18 in China and finds a striking tradeoff in how students use AI for schoolwork. Students using AI saw homework scores increase by 18% and spent less time completing assignments, but performed 20% worse on exams than students who did not use AI—raising questions about whether AI can improve short-term performance while undermining deeper learning when students rely on it to complete work rather than understand it.

Learn more here.

3. Introducing ChatGPT for Teens: Built for Learning, Backed by Protections

OpenAI has launched ChatGPT for Teens, a version of ChatGPT designed specifically for users ages 13–17 with stronger safety protections and features aimed at supporting learning rather than simply completing assignments. The experience includes Study Mode, responsible homework reminders, quizzes, learning visualizations, and Study Hours, while parents can manage selected settings and receive notifications in certain high-risk situations. OpenAI is also partnering with CodeAI to help students and educators build AI literacy and learn to question, direct, and create with AI.

Learn more here.

4. Reach Capital Raises $265M Fund V to Back AI Founders Building to ‘Expand Human Potential’

Reach Capital has raised a $265 million fifth fund to invest in roughly 50 early-stage companies working across learning, health, and work. Their new thesis is focused on AI applications that “expand human potential,” with investments ranging from $1 million to $10 million from pre-seed through Series A. Listen to our interview with Reach Partner Jomayra Herrera for more!

Lear more here.

5. The Prohibition Fallacy: Why Banning Tech Won’t Protect Our Kids

A new Forbes Council essay by Erin Mote argues that banning technology outright is unlikely to protect children from its risks, and instead calls for a more nuanced approach to how young people engage with digital tools. This comes at a moment when schools and policymakers are weighing restrictions on phones, social media, and AI, raising the question of whether effective technology policy should focus less on prohibition and more on teaching young people how to use technology safely and responsibly.

Learn more here.

Inside Reach Capital’s $265M Fund V

Jomayra Herrera is a Partner at Reach Capital, an early-stage venture fund investing across learning, health, and work. Previously, she worked at Emerson Collective and Cowboy Ventures, investing in companies including Handshake, Guild, Contra, and Career Karma.

5 Things You’ll Learn in This Episode

1. Why Reach Capital is doubling down on pre-seed and early-stage investing with its new $265M Fund V.

2. How learning, health, and work intersect to create new opportunities for innovation.

3. What Jomayra looks for in exceptional founders, especially in the age of AI.

4. Why the strongest AI companies go beyond AI as a feature to create new capabilities and durable advantages.

5. How B2C-to-B2B models and school choice are creating new opportunities across education.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Students would enroll.
Degrees would lead to careers.
Technology would support learning, not redefine it.

That world is changing.

AI is reshaping how people learn.

Employers are asking new questions about skills and credentials.

And institutions are being challenged to deliver more value than ever before.

Which is why we’re excited to bring together three leaders who aren’t just talking about the future of higher education...

They’re helping shape it.

• Joseph has served the University of Oklahoma for more than 30 years, including as dean of the College of Law, general counsel, and now its 15th president.

• Has led record enrollment growth, expanded research by more than 70% in six years, and launched OU’s ambitious Lead On, University strategic plan.

• Lisa leads Southern New Hampshire University, the largest nonprofit provider of higher education in the United States, serving more than 200,000 learners worldwide.

• Previously served as SNHU’s provost and spent nearly nine years as president of the AARP Foundation.

• David founded MasterClass, the streaming platform that brings world-renowned experts directly to millions of learners around the globe.

• Has grown MasterClass to feature more than 200 instructors spanning business, sports, entertainment, food, leadership, and the arts.

Join us live or register to receive the recording Wednesday of the following week.

RSVP Here

See you there!

Best,
Edtech Insiders

^{A quick note on email preferences: We’re actively working to reduce email fatigue for our audience. We’ve recently created a dedicated Events Announcements section on Substack so you can control what you receive. If you prefer not to get webinar/event updates, you can unsubscribe from just the “Announcements” section—without leaving the EdTech Insiders newsletter itself. (You’ll still receive our regular news, analysis, and podcast updates.)}

1. EduClaw-Bench evaluates AI-agent configurations over a 30-day learning horizon using simulated learners based on real knowledge-tracing data. Their study revealed that almost none of the model/agent combinations they studied (including GPT5.5 and Alibaba’s Qwen models) maintained strong tutoring performance over time; sustained tutoring demands ongoing capabilities like remembering learner progress, spacing practice, and maintaining strategy, suggesting the need for learner modeling, orchestration, and pedagogical memory. (arXiv)

2. The ELBench research benchmark evaluated nine models (seven general-purpose frontier systems and two education-specialized models) across general capability, safety/trustworthiness, basic education, and higher-order educational development. Specialized Models Underperformed: The two education-specialized models studied failed to lead in either of the education-specific modules. (arXiv)

Why It Matters

Edtech still lacks solid, universally accepted benchmarks to define what actually makes an AI product or tutor “good” at teaching and learning… but it’s not for lack of trying.

1. Learning Commons has been publicly building a suite of ‘evaluators’ of Edtech tool outputs

2. Google has developed a thoughtful internal rubric to evaluate the pedagogical value of its LLM outputs

3. Digital Promise recently announced the recipients of the first grants from its K-12 Infrastructure program which include funding for benchmarks

4. …and last, but not at all least, research teams around the world have been suggesting lots of clever ways to benchmark the outputs of LLM models and their agents/harnesses/product delivery systems.

Our Take

As always, the gap between research and practice is vast and most Edtech companies are not even attempting to benchmark their own products against any of these research-lab-released benchmarks.

That said, maybe they should… at least one lead VC in the field is paying close attention. Here’s a quote from Jennifer Carolan of Reach Capital on Alison Dulin Salisbury’s excellent Humanist Substack:

ALLISON: What have you changed your mind about in the past year?

JENNIFER: I changed my mind about how useful AI tutors are becoming. The underlying technology is improving more quickly than I had ever imagined which is rapidly improving the quality of the AI tutors. Combined with the changes happening in society writ large, these AI tutors may become one of the primary drivers of change of our current formal education system.

I teach a class at Stanford with Steve Blank called Lean Launchpad, and we’ve been so blown away by how quickly students built working prototypes this year—like, Week Two. We’ve had a couple of education teams in this cohort, and it’s been fascinating to see how fast they created compelling AI tutors. In just a few weeks, one team made the leaderboard for Scale AI’s “TutorBench” evaluation—a benchmark for assessing AI tutor quality—so we’ve watched the speed at which these tutors improve in almost real time. Tutors are meeting a pivotal moment, as higher ed is getting pushed hard on ROI and K-12 is under tremendous pressure.

Read the rest of the article here.

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

EdTech Isn’t One Industry. It’s Nine.

By: Ian McCullough

Ian McCullough is an EdTech marketing executive who lives in the San Francisco Bay Area. You can learn more about him on LinkedIn and his website.

The post-COVID EdTech industry remains in trouble. Rather than being the driver of new industry growth, AI models have undercut revenue potential. The contrast between boundless AI-powered possibilities for learning and the near-term likelihood of scalable new ventures is stark. While it is unequivocally good that the lockdowns which triggered the 2020-21 investment surge are behind us, 2025 was down 68% from 2018’s pre-pandemic peak. Owl Ventures is looking at long-term opportunities in Africa and exploring industry intersections with HealthTech and FinTech. GSV Ventures, which has historically focused on the United States, is now shining a spotlight on Indian EdTech. Reach Capital has formally expanded its thesis and Brighteye Ventures now refers to their strategy as HumanOS.

There is more going on here than a pendulum swing from a world-shaking black swan event. The economic models and market narratives EdTech has relied on for 30 years have collapsed. We must rebuild both. If we can get the latter right, the former becomes way easier to solve.

Wider discussions about the impacts of social media, artificial intelligence, screen-based devices, and corporate responsibility across the broader tech industry have engulfed EdTech. Shrinking budgets have led schools to scrutinize student outcomes. A series of cybersecurity incidents has put IT teams on high alert. These are serious challenges and it’s understandable why pessimism presently pervades. EdTech, however, has always been a magnet for optimists devoted to learning and growth. Readiness to learn and grow is exactly what we need now to reestablish a shared understanding of who we are and what we do.

EdTech’s Identity Crisis

Questions of identity may seem abstract or irrelevant, but we can’t sell what we can’t name and explain. That’s as true of our industry as it is of all our products and services. In the days when textbooks and worksheets ruled the school, “education technology” was understood in contrast to “educational publishing.” EdTech represented solutions delivered via silicon and software rather than paper and ink. As that distinction is no longer meaningful, even pillars of the community now wonder if the name “EdTech” fits us anymore.

EdTech Insiders’ own Ben Kornell has repeatedly contemplated the matter on the Edtech Insiders’ podcast, reflecting on how he and Alex Sarlin have debated calling the show “Education Insiders” or “Learning Insiders”. The International Society for Technology in Education (ISTE) just renamed itself the International Society for Transforming Education. Computer-Using Educators (CUE) is now the California Association for Leading Innovation in Education (CALIE). The common thread in these approaches to the blurring of publishing and technology is that they erase technology from the conversation. While I understand the thinking, that approach comes at a cost. When we try to maintain relevance by invoking abstractions like “transformation” and “innovation,” we instead erode it by diffusing the value proposition of all industry participants.

There is a better way forward. By sharpening definitions and drawing clearer boundaries, we can focus our resource investments and set the stage for new entrepreneurial activity..

An Original Approach to Evaluating EdTech Opportunities

Formal definitions of EdTech already exist and so do excellent resources that categorize, organize, and track our output. The breadth of the institutional market is cataloged in detail by HolonIQ’s Global Education Market Taxonomy. Whiteboard Advisors and Phil Hill and Associates are standouts for analysis of market trends. The landscape of generative AI tools for K-12 was thoroughly mapped by EdTech Insiders. While each excels within its scope, it is difficult to assemble them into a larger whole. Fractures and gaps are evident.

In considering the existing definitions of EdTech, one thing stands out: they all have a very detached tone. A key element that has anchored me to the industry is a desire to have a positive impact on people’s lives. Any definition that I embrace has to incorporate a sense of mission and purpose. After some reflection, I came to a definition that I personally find meaningful:

EdTech is a set of tools and systems that expand access to effective instruction so that every learner can acquire a life-changing education.

In systemically constructing a framework that encompasses the whole industry, that definition became the foundation. The framework which ultimately emerged is this 3x3 grid.

Intersections of market type and solution category create nine unique areas. As a starting point for understanding, I’ve provided examples in each area. When we think about companies that offer a range of solutions, some like Coursera do business in multiple areas. Others like LEGO have their core in a different industry entirely but work within EdTech at one or more junctions. Google arguably has solutions in all nine areas.

The basic structure is simple yet specific. When we unpack the dimensions, the framework enables much deeper exploration of business fundamentals without sacrificing cohesion.

The Solution Categories

Learn-to-read software that parents choose to download to a personal device for their four year old is EdTech. A student information system (SIS) used by a university with thousands of students is also EdTech. How can those two things be obviously different and yet the same? My interest in resolving that is what set this whole exercise in motion. I wanted to compare financial performance and couldn’t do that without categories. (I found that typical IRR is < 10% for Instruction and 15-20% for Infrastructure, although that comes with a lot of nuance.) Managed Services became its own category after a conversation about online program managers (OPMs) with Workshop Ventures’ Mike Berlin.

Assessment, a big topic in its own right, merits special mention as different assessment solutions belong in different categories. A question bank would be an instruction solution while lockdown software for high stakes exams is infrastructure. A standardized testing provider like ETS could be considered a managed service.

The Market Types

If we’re strictly applying textbook principles, it could be argued that making B2C / B2B / B2G a top-level split puts too much focus on the transaction when the priority should be the customer and their job to be done. The choice, however, wasn’t motivated by the number three’s aesthetic appeal. It’s born of an issue that I frequently encounter in my own practice.

I have spoken with many founders over the years who want to take the product they built around a consumer paradigm and apply corporate enterprise go-to-market motions — most commonly product-led growth (PLG) — in pursuit of large institutional deals in K-12 or Higher Ed. The problem is that what works in one setting doesn’t automatically translate into the others. The buying cycles and processes are completely different. When you look past the transaction, B2C / B2B / B2G is the crispest way I’ve found to characterize internal relationships and organizational context in a way that relates segments.

A Restarting Point

After 25 years in the industry, what I like most about this is the fresh questions I suddenly find myself asking. I never felt like I could fit all the pieces of the EdTech puzzle together and now I can. As an example using only reasoning from definitions, we can analyze the threat of new entrants in each of the nine areas.

We now have an organized way for EdTech teams to discuss the potential risks and rewards of different market moves — and this is only the beginning. As mentioned, I’ve been collecting publicly available business data. I’m going to see what cross-area comparisons are possible and use that as a basis for potential case studies and white papers. Since my goal in sharing is to help us all find new ways forward as an industry, my hope is that the 3x3 framework sparks new questions for you too.

In the meantime, I’d appreciate your feedback in the comments section. How does this approach give you some new insight into EdTech? How does your novel product break the model? As fellow members of the EdTech Insiders Plus WhatsApp group will attest, I value rich conversations.

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Upcoming Events

Bay Area Edtech Happy Hour

Thursday, August 20
5:00-6:30 PM PT
The GreyhoundSF

Join Edtech Insiders to celebrate back-to-school outside at The GreyhoundSF beer garden! Join us for drinks, discussion, and time to catch up with edtech community.

RSVP Here

What Is the Future of Higher Education in America?

Sponsored by Masterclass

Friday, August 28
9:00-10:00 AM PT / Noon-1:00 PM ET
Webinar on Zoom

AI is reshaping learning, employers are rethinking credentials, and universities are facing unprecedented change. Join us for a live conversation with an incredible panel of experts exploring how higher education must evolve to remain relevant, accessible, and valuable in the decade ahead.

Join us live or register to receive the recording the following Wednesday.

RSVP Here

Most Interesting Thing In Edtech

Hey Insiders! We are excited to be launching a new segment on our Substack that we’re calling “Most Interesting Thing In Edtech”. This will be a quick read covering a timely headline, why it matters, and what about it caught our eye.

Here’s a sample of what this will look like, and if you want to subscribe to this offering you can do so here, just toggle on the notifications for the publication “Most Interesting Thing in Edtech.” Reach out if you have any questions about this!

New Tutoring Benchmarks Continue to Reveal Gaps in AI Models for Teaching and Learning

There are two new interesting papers about education benchmarks for LLMs:

1. EduClaw-Bench evaluates AI-agent configurations over a 30-day learning horizon using simulated learners based on real knowledge-tracing data. Their study revealed that almost none of the model/agent combinations they studied (including GPT5.5 and Alibaba’s Qwen models) maintained strong tutoring performance over time; sustained tutoring demands ongoing capabilities like remembering learner progress, spacing practice, and maintaining strategy, suggesting the need for learner modeling, orchestration, and pedagogical memory. (arXiv)

2. The ELBench research benchmark evaluated nine models (seven general-purpose frontier systems and two education-specialized models) across general capability, safety/trustworthiness, basic education, and higher-order educational development. Specialized Models Underperformed: The two education-specialized models studied failed to lead in either of the education-specific modules. (arXiv)

Why It Matters

Edtech still lacks solid, universally accepted benchmarks to define what actually makes an AI product or tutor “good” at teaching and learning… but it’s not for lack of trying.

1. Learning Commons has been publicly building a suite of ‘evaluators’ of Edtech tool outputs

2. Google has developed a thoughtful internal rubric to evaluate the pedagogical value of its LLM outputs

3. Digital Promise recently announced the recipients of the first grants from its K-12 Infrastructure program which include funding for benchmarks

4. …and last, but not at all least, research teams around the world have been suggesting lots of clever ways to benchmark the outputs of LLM models and their agents/harnesses/product delivery systems.

Our Take

As always, the gap between research and practice is vast and most Edtech companies are not even attempting to benchmark their own products against any of these research-lab-released benchmarks.

That said, maybe they should… at least one lead VC in the field is paying close attention. Here’s a quote from Jennifer Carolan of Reach Capital on Alison Dulin Salisbury’s excellent Humanist Substack:

ALLISON: What have you changed your mind about in the past year?

JENNIFER: I changed my mind about how useful AI tutors are becoming. The underlying technology is improving more quickly than I had ever imagined which is rapidly improving the quality of the AI tutors. Combined with the changes happening in society writ large, these AI tutors may become one of the primary drivers of change of our current formal education system.

I teach a class at Stanford with Steve Blank called Lean Launchpad, and we’ve been so blown away by how quickly students built working prototypes this year—like, Week Two. We’ve had a couple of education teams in this cohort, and it’s been fascinating to see how fast they created compelling AI tutors. In just a few weeks, one team made the leaderboard for Scale AI’s “TutorBench” evaluation—a benchmark for assessing AI tutor quality—so we’ve watched the speed at which these tutors improve in almost real time. Tutors are meeting a pivotal moment, as higher ed is getting pushed hard on ROI and K-12 is under tremendous pressure.

Read the rest of the article here.

How Boddle Turned Classroom Math into a Game Kids Choose to Play

Edna Martinson is the Co-Founder and COO of Boddle Learning, a gamified K–6 learning platform used by millions of students to build confidence in math, literacy, and science. She is passionate about combining great game design with research-backed learning experiences that teachers and students love.

5 Things You’ll Learn in This Episode

1. Why Boddle was built to compete with video games.

2. How gamification drives learning and confidence.

3. What teachers want from classroom technology.

4. How AI is shaping Boddle’s future.

5. The next evolution of personalized learning.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

AI is transforming how students learn.
Employers are rethinking the value of traditional degrees.
Universities are facing new enrollment, financial, and workforce challenges.

So what comes next?

Join us for a live webinar with leaders from higher education and industry as we explore the future of colleges and universities, and what institutions must do to remain relevant in the decade ahead.

We’ll discuss:

• How AI is changing teaching, learning, and the university experience

• Whether traditional degrees will remain the dominant pathway to opportunity

• What institutions must do to adapt to changing student, workforce, and economic realities

Join us live or register to receive the recording Wednesday of the following week.

RSVP Here

See you there!

Best,
Edtech Insiders

A quick note on email preferences: We’re actively working to reduce email fatigue for our audience. We’ve recently created a dedicated Events Announcments section on Substack so you can control what you receive. If you prefer not to get webinar/event updates, you can unsubscribe from just the “Announcements” section—without leaving the EdTech Insiders newsletter itself. (You’ll still receive our regular news, analysis, and podcast updates.)

Teacher Optimism on AI Is Dropping. The Reason Isn’t the Technology.

Sponsored by Overdeck Family Foundation

By: Lin Ler

Lin Ler is a Stanford MBA and MA in Education candidate working at the intersection of business strategy, learning science, and frontier AI. With a background spanning management consulting, EdTech startups, and the social sector, Lin translates complex AI capabilities into practical insight for the future of learning.

Spend some time on Linkedin these days and you’re likely to see polarized views about AI… and perspectives from educators are no exception. For every educator bursting with excitement about the innovative AI tools and simulations they’re building for their students, there is another voicing concern that AI has been eroding their students’ ability to think critically and wrestling with how to change their assessment strategy in the face of AI. Just this week, Mississippi college instructor Jason Gibson became an educator folk hero for embedding a clever trap into an midterm assessment that caught more than 90% of his students using AI.

So what does the research actually say about educator’s responses to AI? One research team from the University of Washington analyzed more than 13,000 unscripted conversations between K-12 educators and AI on an open-access platform. What they saw is teachers treating AI primarily as a “thought partner” or assistant that executes teacher-defined goals, rather than as a coach meant to improve their own instructional practices. Teachers leaned on it to reduce workload and increase efficiency — lesson planning, generating assessments, curating teaching materials, and creating differentiated or bilingual content tailored to diverse student needs.

But as you might expect, teachers are also critical about how AI is used in the classroom, by both educators and students. A Stanford team that convened teachers for a Practitioner Voices Summit found that educators evaluate AI tools through what the researchers call “deliberative sensemaking” — weighing productive tensions, like balancing the desire for personalized learning against concerns over fairness and bias, or weighing the adaptability of a tool against the time it takes to use it efficiently. Their primary concerns revolve around inaccuracies, generic outputs, and the risk of academic dishonesty. A study of 167 preservice teachers across four teacher education institutions in Ghana surfaced a different worry: that AI lacks the capacity for emotional connection and personalized human support, potentially weakening the student-teacher relationship. And a nationwide survey of Brazilian teachers raises the concern that, much like previous educational technologies, AI might exacerbate existing inequalities if introduced without careful consideration of diverse classroom contexts and resources.

The Souring of Educator Attitudes

While there is still diverse sentiments among teachers, polling has indicated a measurable cooling trend in teachers’ enthusiasm for AI. A 2026 spring survey revealed that a majority (55%) of teachers now oppose using AI in the classroom, representing an 8-point drop in support since the beginning of the school year. Furthermore, 65% of teachers oppose allowing their students to use AI for schoolwork, which is also an 8-point increase in opposition. This trend is mirrored in higher education, where faculty intent to use AI has declined significantly in the U.S. and Canada. And the Brazilian teachers surveyed earlier voiced a similar mix of worries: generic outputs, overreliance on technology, and students misusing AI for an unfair advantage.

Importantly, this cooling is not indifference — it is coming from a workforce that takes AI seriously. A nationally representative NPR/Ipsos poll of 545 K-12 teachers found that nearly 3-in-4 believe AI has bigger implications for education than past innovations like the internet or computers, and 60% are already using it themselves to save time and improve their teaching materials. Those same teachers are the ones sounding the alarm: 54% say AI makes it harder for students to develop critical thinking skills, 55% see it as mostly a shortcut for students to avoid doing the work, and nearly 6-in-10 say it is eroding trust between students and teachers. The skepticism is a considered judgment from heavy users, not a reflex from the sidelines.

It is tempting to read this resistance as generational, assuming that veteran teachers are rejecting a cutting edge technology they didn’t grow up with. The data says otherwise. When a research team surveyed 508 K-12 teachers across six countries — Brazil, Israel, Japan, Norway, Sweden, and the United States — they found that age and education level had negligible effect on whether teachers trust AI. What predicted trust was self-efficacy: a teacher’s confidence in their own ability to understand and use these systems. Teachers with higher self-efficacy perceived more benefits, fewer concerns, and reported more trust. The change in attitudes is not only a verdict on the technology, but also partially a signal about the lack of support teachers are receiving.

The Policy Vacuum and the Burden of Ambiguity

It’s important to note that teacher skepticism is often exacerbated by the fact that schools are failing to provide a roadmap for implementing AI. Only 18% of U.S. K-12 teachers report receiving formal, written guidance on how to use AI tools at work, according to a 2026 Walton Family Foundation–Gallup survey of 2,069 public school teachers. For highly consequential tasks, the void is even larger: 69% of teachers receive absolutely no guidance on using AI for one-on-one student tutoring, and 58% receive no guidance on using it for grading and providing feedback. The pattern holds across nearly every use case — fewer than one in ten teachers receive formal guidelines for any specific AI-assisted work activity, including just 5% for tutoring students, 6% for analyzing patterns in student achievement data, and 7% for getting AI coaching on their own teaching.

When formal policies are absent, the burden of navigating this rapidly changing and ethically complex technology falls entirely on individual educators. The same Walton–Gallup research links this lack of role clarity and the weight of unrealistic expectations directly to higher burnout rates and lower job engagement. The policy gap also worsens inequities, as teachers in higher-need schools are even less likely to receive guidance on AI use than those in wealthier schools.

In that vacuum, each teacher’s AI policy is assembled alone, from trial and error or through unofficial teacher channels. This is a burden on teachers and students alike; how can students successfully navigate an educational environment where every class has a different policy and a strategy that is allowed, even required, in one class might mean expulsion from another?

Global Evidence for AI Teacher Training

Educators across the globe acutely feel this lack of preparation. Returning to that Brazilian national survey, over 80% of teachers reported having only basic or limited knowledge of AI, citing a lack of training and technical support as the primary barriers to implementation. Despite these frustrations, teachers have a strong appetite for training: the same share, over 80%, said they wanted continuous, online professional development tailored to their knowledge levels.

A separate team running a national survey of teachers in Indonesia heard the same concerns. Even when teachers do receive some professional learning on educational technology, it often misses the mark on critical evaluation; for instance, while many pre-K teachers receive tech training, only about a third are taught how to actually judge the quality of those tools.

Many teachers emphasize that they do not just need technical tutorials; they need spaces where they are given protected time to collaboratively explore tools, push back against bad ideas, and align AI with their pedagogical values.

The Transformative Impact of Formal Support

The data shows that when schools do provide clear policies and training, teacher attitudes shift dramatically:

• Formal guidance encourages adoption: 69% of teachers who receive formal, official guidance say it actively encourages them to use AI, compared to only 51% of those who receive informal or verbal guidance.

• Training builds confidence and reduces fear: A team at the University of Cologne put nearly 300 German teachers through a four-week online AI certificate course and measured them before and after — perceived preparedness improved markedly, and the training demystified the technology, shifting perceptions from science-fiction notions of AI toward a realistic grasp of its everyday applications and limitations. A smaller pilot training science teachers in prompt engineeringfound the same direction of travel.

• Diagnosis should precede training: Because self-assessment is unreliable, a team presenting at the LAK26 learning analytics conference argues for objective diagnostics first — profile who overestimates, who underestimates, and who is starting from zero, then match the intervention to the profile. The rubric for this already exists in UNESCO’s AI Competency Framework for Teachers; what most districts lack is the diagnostic step.

There is a big catch to this background context: for the most part, when teachers are asked about their own AI literacy, results are self-reported. That LAK26 team measured teachers’ AI literacy both ways — asking them to rate themselves, and testing them objectively… and the correlation between the two was low. Teachers sorted into six distinct profiles, including overestimators whose confidence outran their tested ability and underestimators whose tested ability outran their confidence. Underestimators distrust tools they could use well, while overestimators trust tools they cannot critically evaluate.

Ultimately, researchers who interviewed 22 teachers in an early-adopting U.S. district came away with a clear read: teachers are not inherently anti-technology, but they are understandably resisting being forced to implement complex, high-stakes tools in a vacuum. Schools that are willing to invest in clear policies and comprehensive training can alleviate this burnout, transforming AI from a source of anxiety into a supported, effective classroom asset.

How Schools Might Support Teachers

What schools are actually doing, then, falls short in two connected ways: they are leaving teachers without a map, and they are measuring readiness in self reporting that may be disconnected from reality.

1. Put the guidance in writing, because written policy measurably encourages use where verbal suggestion does not.

2. Protect time for teachers to make sense of these tools together, because collaborative testing is how miscalibrated confidence gets corrected.

3. Assess AI literacy the way schools assess every other competency they take seriously: objectively, before the training dollars are spent.

We would never accept students self-reporting their own reading levels as an assessment system. For teacher AI literacy, that is currently the assessment system, and the souring poll numbers are what it produces.

Bibliography

Aisyah, N., Al Kautsar, M. D., Hidayat, A., & Koto, F. (2026). Grounding AI-in-education development in teachers’ voices: Findings from a national survey in Indonesia. arXiv. https://arxiv.org/abs/2604.01630

Aldis, A. (2026, May 7). Surveying teachers on artificial intelligence. EdChoice. https://www.edchoice.org/2026-surveying-teachers-on-artificial-intelligence/

Berne, J., Doss, C. J., & Shapiro, A. (2025). Pre-K teachers are optimistic about educational technology, though current use varies widely: Findings from the American Public School Pre-K Teacher Survey. RAND Corporation. https://www.rand.org/pubs/research_reports/RRA4412-2.html

Dangol, A., Kotiyal, S., Wolfe, R., Bowers, A. J., Vigil, A., Yip, J., Kientz, J. A., Shahid, S., Yeh, T., Cho, V., & Davis, K. (2025). Relief or displacement? How teachers are negotiating generative AI’s role in their professional practice. arXiv. https://arxiv.org/abs/2510.18296

Demszky, D., Mah, C., & Higgins, H. (2026). Practitioner Voices Summit: How teachers evaluate AI tools through deliberative sensemaking. Stanford University / EdWorkingPapers. https://edworkingpapers.com/sites/default/files/ai26-1466.pdf

Di Lullo, A., Bielik, D., Rong, H., Mavlonova, M., Chun, C., & Subhan, F. (2026). AI in higher education global survey 2026. Digital Education Council. https://www.digitaleducationcouncil.com/resource-library-items/ai-in-higher-education-global-survey-2026

Doss, C. J., Bozick, R., Schwartz, H. L., Chu, L., Rainey, L. R., Woo, A., Reich, J., & Dukes, J. (2025). AI use in schools is quickly increasing but guidance lags behind: Findings from the RAND survey panels. RAND Corporation. https://www.rand.org/pubs/research_reports/RRA4180-1.html

Fernandes, D., et al. (2025). AI makes you smarter but none the wiser: The disconnect between performance and metacognition. Computers in Human Behavior. https://www.sciencedirect.com/science/article/pii/S0747563225002262

Florentino, B., Sestito, C., de Carvalho, A., Cruz, W., & Bonidia, R. (2026). Artificial intelligence for all? Brazilian teachers on ethics, equity, and the everyday challenges of AI in education. arXiv. https://arxiv.org/abs/2512.23834

Gaines, L. V. (2026, June 5). Most K-12 teachers say AI’s impact on education will eclipse the internet or computers. NPR. https://www.npr.org/2026/06/05/nx-s1-5779757/school-ai-education-students-teachers-poll-critical-thinking

Gousopoulos, D. (2025). Exploring teachers’ perceptions of ChatGPT through prompt engineering. arXiv. https://arxiv.org/abs/2510.08634

Lademann, J., Henze, J., Honke, N., Wollny, C., & Becker-Genschow, S. (2025). Teacher training in the age of AI: Impact on AI literacy and teachers’ attitudes. arXiv. https://arxiv.org/abs/2507.03011

Liu, A., Esbenshade, L., Sarkar, S., Tian, Z., Zhang, Z., He, K., & Sun, M. (2025). How K-12 educators use AI: LLM-assisted qualitative analysis at scale. arXiv. https://arxiv.org/abs/2507.17985

Malek Ash, A., & Senseman, K. (2026, May 26). Most teachers receive no formal guidance on AI use. Gallup / Walton Family Foundation. https://news.gallup.com/poll/710534/teachers-receive-no-formal-guidance.aspx

Nyaaba, M., Shi, L., Nabang, M., Zhai, X., Kyeremeh, P., Ayoberd, S. A., & Akanzire, B. N. (2024). Generative AI as a learning buddy and teaching assistant: Pre-service teachers’ uses and attitudes. arXiv. https://arxiv.org/abs/2407.11983

UNESCO. (2024). AI competency framework for teachers. https://www.unesco.org/en/articles/ai-competency-framework-teachers

Viberg, O., Cukurova, M., Kizilcec, R. F., et al. (2024). What explains teachers’ trust of AI in education across six countries? arXiv. https://arxiv.org/abs/2312.01627

Zhang, X., et al. (2026). How to assess AI literacy: Misalignment between self-reported and objective-based measures. Proceedings of LAK26 / arXiv. https://arxiv.org/abs/2601.06101

Special thanks to Overdeck Family Foundation for sponsoring this article in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative).

You can support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Top Edtech Headlines

1. New Ways to Learn and Teach With ChatGPT Work and Codex
OpenAI has launched a new learning hub featuring free courses, tutorials, and live workshops designed to help people use ChatGPT and Codex effectively in education and the workplace. The resources include dedicated content for educators on integrating AI into teaching, alongside practical guidance for students, professionals, and developers looking to build AI fluency.

Learn more here.

2. High School Students Draft Their Own AI Rules for Schools
A new initiative from the Transparency Coalition highlights AI governance recommendations written by high school students themselves, offering a student-centered perspective on how AI should be used in classrooms. Their proposals emphasize transparency, privacy, academic integrity, and meaningful human oversight, demonstrating how young people can help shape responsible AI policies in education.

Learn more here.

3. Some Professors Say AI Is Pushing Them Into Early Retirement
A new Chronicle of Higher Education feature explores how the rapid rise of generative AI is contributing to early retirement decisions among some faculty members. Professors describe growing frustration with AI’s impact on teaching, assessment, and academic integrity, raising broader questions about how higher education can adapt while preserving meaningful learning and faculty engagement.

Learn more here.

4. OpenAI Expands Research Access With New ChatGPT Initiative for Academics
OpenAI has launched a new program to give researchers expanded access to ChatGPT and advanced AI tools, aiming to accelerate academic research across disciplines. The initiative provides eligible scholars with enhanced capabilities and resources to support research projects, signaling OpenAI’s continued investment in advancing scientific discovery through AI.

Learn more here.

Claude for Teachers: How Anthropic Is Reimagining AI for Educators

Drew Bent leads Education at Anthropic, where he oversees education initiatives including Claude for Teachers. A former math teacher, Khan Academy engineer, and co-founder of Schoolhouse.world, Drew focuses on applying AI to improve teaching and learning.

Sandra Liu Huang is President of Learning Commons, where she leads the development of open AI datasets and learning science infrastructure for education. She previously held product leadership roles at Google, Facebook, and Quora.

5 Things You’ll Learn in This Episode

1. Why Anthropic built Claude for Teachers exclusively for educators.

2. How Learning Commons grounds AI in standards, curriculum, and learning science.

3. The role of open datasets, AI skills, and shared infrastructure in improving educational AI.

4. How Claude integrates with leading edtech tools through new connectors.

5. What’s next for Claude for Teachers, including district partnerships and future AI capabilities.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

America Doesn’t Just Have a Skills Gap. It Has a Hiring Model Problem.

By: Pierre Dubuc

Pierre Dubuc is the Co-Founder and CEO of OpenClassrooms, a global education-to-employment platform that partners with organizations worldwide to provide online learning programs and DOL-approved apprenticeships. A recognized leader in workforce development, he was nominated by the French-American Foundation among the 2020 Young Leaders and decorated by the French government for his commitment to expanding professional opportunity.

America does not have a shortage of talent. It has a shortage of imagination in how talent is recognized, developed, and hired.

For years, the workforce conversation has been framed in familiar terms: employers cannot find the people they need, workers do not have the right skills, and education providers must do a better job preparing people for the jobs of the future. There is truth in that framing, but it misses the larger issue.

The deeper problem is not simply that workers lack skills. It is that too many employers still rely on outdated ways of identifying readiness. They say they value skills, but often hire for proxies. They say they want adaptability, but screen for pedigree. They say they need talent, but expect the market to deliver polished candidates with minimal training, minimal investment, and minimal risk.

That is not just a skills gap. It is a hiring model problem.

If we are serious about solving it, we need to stop treating education, hiring, and workforce development as separate conversations. We need stronger bridges between learning and earning. We need systems that help employers hire for potential, develop talent in context, and create more credible pathways into good jobs. That is why apprenticeship deserves a much more central place in the American workforce conversation.

The Labor Market Is Full of Contradictions

The macroeconomic picture is pulling in two directions at once. On one hand, nearly seven out of ten (69%) American employers report difficulty filling open roles (source), especially in fast-changing technical sectors. On the other hand, the labor market remains stubbornly hard to enter for non-traditional candidates.

The contradiction is everywhere. Employers say they cannot find talent, yet candidates are screened out for lacking experience they could only have gained by already being hired. Companies say they want durable skills and agility, yet many still rely on filters that reward conventional credentials, polished résumés, and prior access more than actual capability. Leaders talk about mobility and opportunity, yet the path from education to employment remains fragmented, expensive, and uneven.

The result is a labor market that fails both sides. Employers spend months searching for people who look “ready,” while capable workers remain locked out because they have not yet had the chance to prove themselves. In sector after sector, the problem is not that potential is missing. The problem is that our systems are bad at turning potential into proof.

That is especially visible in fast-changing fields, where roles evolve quickly and yesterday’s ideal candidate profile is already outdated. In those environments, waiting for the flawless hire is not rigor. It is avoidance dressed up as rigor.

Skills-First Hiring Is a Good Idea. It Is Not Yet an Operating System.

One of the most encouraging shifts in recent years has been the rise of skills-first hiring. Major employers and state governments have removed bachelor’s degree requirements from many job descriptions. At its best, that reflects an overdue recognition: talent is more broadly distributed than opportunity, and capability cannot be reduced to a diploma, a brand-name institution, or a narrow career history.

But skills-first hiring has a credibility problem when it remains more branding than practice. Research from Harvard Business School and the Burning Glass Institute shows that the actual percentage of skills-first hires has barely budged. Removing one barrier from a job posting is not the same as redesigning how people are sourced, assessed, onboarded, supported, and promoted.

That is where the conversation often falls apart. We ask education providers to produce fully job-ready candidates for roles that are constantly changing. We ask employers to broaden access without giving them better tools to reduce hiring risk. And we ask learners to navigate a maze of tuition, credentials, unpaid signaling, and disconnected systems before they can access meaningful opportunity.

The issue is not that skills-first hiring is the wrong idea. The issue is that skills-first hiring needs infrastructure.

Apprenticeship is that infrastructure.

Apprenticeship Is Not Old-Fashioned. Our Hiring Model Is.

Apprenticeship is still too often treated as if it belongs to another era, associated mainly with the trades or with workforce systems sitting adjacent to the “real” higher education and talent economy. That is exactly backwards.

In many ways, apprenticeship is one of the most modern responses we have to the modern hiring problem because it addresses the failures that define today’s labor market:

• Hiring for potential over polish: it gives employers a structured way to evaluate baseline aptitude, work habits, and behavioral alignment rather than relying on pedigree.

• Contextual competence: it lets people build capability through real, production-level work rather than simulated classroom readiness.

• Objective readiness signals: it replaces vague résumé claims with demonstrated performance in context.

In other words, apprenticeship narrows the gap between education and employment. It gives learners paid access to opportunity instead of asking them to take on more cost and risk before they can begin. It gives employers a practical way to build talent instead of endlessly searching for perfect-fit candidates who may not exist. And it produces a much stronger signal of readiness because people are learning by doing.

This is why apprenticeship should not be discussed only as a social good, though it can absolutely expand mobility and access. It should also be understood as a business solution. When done well, apprenticeship is not charity. It is talent strategy.

The Model Is Getting Stronger — and More Credible

What makes this moment especially important is that apprenticeship is no longer confined to an “alternative pathway” category. The boundaries between workforce training, employment, and recognized higher education are beginning to shift.

In California, OpenClassrooms is now authorized to grant degrees, with Texas still pending. That matters not only as a regulatory milestone, but as evidence that work-based pathways can also carry recognized academic legitimacy.

For too long, workforce systems and higher education systems have been treated as if they occupy separate lanes: one for employability, one for legitimacy; one for practical skills, one for recognized credentials. But if apprenticeship pathways can combine paid work, structured learning, and degree outcomes, that old divide starts to break down.

And it should. The future of talent development should not force people to choose between earning and learning, practical experience and credible credentials, or economic mobility and educational quality. The strongest models increasingly combine those things instead of separating them.

Apprenticeship Matters Even More in an AI Economy

This becomes even more urgent in an economy shaped by AI. As roles evolve more quickly, the shelf life of technical skills gets shorter. PwC’s Global AI Study suggests that AI is reconfiguring job roles across knowledge sectors. Employers need people who can adapt, learn continuously, and build competence in motion. That makes static models of preparation less useful.

If work is changing in real time, then preparation for work cannot remain detached from work itself. We need systems that help people learn while contributing, build capability while solving real problems, and receive feedback in environments that resemble the jobs they are preparing for.

That is where apprenticeship has a structural advantage. It is not built on the assumption that education happens first and productive work happens later. It is built on the idea that learning and contribution can reinforce one another. That is a much better fit for an economy in which no one is ever really “finished” learning.

What Modern Apprenticeship Actually Requires

Of course, apprenticeship is not automatically effective just because the label sounds good. Like any model, its value depends on execution.

If apprenticeship is going to scale meaningfully in the United States, several things have to be true. First, it must be rooted in real employer demand. Programs work best when they are anchored in actual roles, actual business needs, and real employer commitment to participate in talent development.

Second, the learning model must be rigorous. Apprenticeship is not about throwing people into jobs and hoping they figure it out. It requires clear skill milestones, intentional progression, strong support, and measurable outcomes.

Third, management matters. Apprenticeship is not only a curriculum challenge; it is a leadership challenge. Learners need feedback, coaching, and clear expectations. Managers need practical ways to develop people, not just evaluate them.

Fourth, incentives matter. Many employers need help de-risking the initial investment. The best long-term models are sustainable, but early adoption often depends on public, philanthropic, or workforce funding that makes participation more realistic.

These are not side issues. They are the point. If we want employers to stop complaining about talent shortages and start building talent more intentionally, we need systems that make that shift practical.

The Real Question Is Not Whether We Need More Talent

We do. But that is not the hardest problem.

The harder problem is whether we are willing to change the systems that decide who gets seen as talented in the first place. For too long, we have defined readiness as something people must prove before opportunity is extended to them. In reality, readiness is often built through opportunity, not before it. People become more capable by doing real work, under real expectations, with real support.

That is what apprenticeship gets right. It does not assume talent arrives fully formed. It assumes talent can be developed.

And at a moment when employers say they need adaptability, workers need clearer pathways, and the economy demands continuous learning, that is not a marginal insight. It is a central one.

America does have a skills challenge. But more fundamentally, it has a systems challenge, a transition challenge, and a bridge challenge. Bridges do not appear on their own. They are built because people decide they matter.

If we want a labor market that is more dynamic, more inclusive, and more honest about how talent actually develops, then we need to stop asking whether people are perfectly job-ready before they begin. We need to build models that let people earn while they learn, prove themselves while they grow, and access opportunity through demonstrated capability rather than inherited advantage.

The future of work will not be shaped by who talks most about skills. It will be shaped by who finally builds better ways to turn potential into opportunity.

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

BRIDGES Summit Rundown

By: Ben Kornell

This summer, New York City has been THE city of champions. In June, the Knicks won their first NBA championship in 53 years, and then in July FIFA crowned Spain as the World Cup champions. So what about edtech champions? Last week, the inaugural BRIDGES AI Summit kicked off at Chelsea Pier 60 with a room filled with leaders across K12, Higher Education, and GovTech more broadly.

Edtech Insiders had a front row seat as this conference flipped the normal ratio of founders-to-system leaders from 4:1 to 1:4. With Jenn Womble and Jacob Kantor helping to organize the Starbridge-backed event, the room was filled with Superintendents, CTOs, and system leaders.

Key takeaways from both panels and informal sessions included:

The Tech-Lash Is Real, but So Is Coherence

Education leaders are culling the tools used across schools, universities, and systems in hopes of creating streamlined, coherent, and integrated teaching & learning experiences. While screen time is getting a lot of the airtime, the biggest concern amongst attendees is that the proliferation of edtech tools in COVID and then with AI undermined instructional alignment and rigor. Going forward, education leaders hope to go deeper with fewer learning tools that align to system-wide goals.

Partnership Versus “Just Another Vendor”

With this quest for depth, systems are much more oriented to organizations that can invest in implementation and customization aligned to the unique needs of staff and students. One participant shared “It’s ironic that edtech companies are pitching personalization but their product roll-out is one-size fits all.” The other push for partnership is that fact that AI is leading to product changes at a much faster clip. System leaders want to be a part of shaping the roadmap and also the roll-out.

Go Slow to Go Fast With AI

Overall, leaders expressed caution about AI, especially in student facing contexts. A simple frame would be:

• Red light: student facing AI. Leaders are concerned about both learning and safety.

• Yellow light: teacher-facing AI. Leaders want to ensure that new tools are aligned with system-wide best practices and instructional strategies.

• Green light: administrative AI. Leaders are genuinely excited to generate bureaucratic lift and greater efficiency on the back-office side.

• Don’t lead with AI: Leaders also emphasized that if companies lead with AI as their value-prop, then they are turning off their end customers.

Include CTOs in the Sales Motion

A big point of contention was the prevalence of bottoms-up, freemium adoption of AI tools. CTOs warned that when teachers are encouraged to adopt tools directly, companies are alienating district leaders who have to deal with the consequences — whether that is cognitive offloading, cybersecurity risk, or data privacy violations. CTOs also lamented to-down superintendent sales, as often IT is the last to know. A frank panel of CTOs warned that the IT leadership can be an implementation partner or a real roadblock, so companies should consider carefully how they enter school systems.

It’s All About Outcomes

The final takeaways is a theme we’ve been covering for quite some time. It’s not about the tech, it’s about outcomes for students. Generally speaking, system leaders expressed skepticism about company-produced research. They favor pilots that show real impact in their own context, followed by referrals from education leaders that they trust.

We’re excited for next summer, when the champions of edtech reconvene for Bridges 2.0!!

Top Edtech Headlines

1. OpenAI Launches Campus Leads Program for Students to Build AI Communities

OpenAI is inviting university students to apply for its first Campus Leads program, a global initiative focused on building student communities around ChatGPT and Codex. Selected students will help organize campus events, share AI resources, and connect peers with OpenAI’s tools, expanding the company’s efforts to make universities a key part of its AI ecosystem.

Learn more here.

2. Coursera Invests $100M in Andrew Ng’s New AI Education Startup

Coursera is investing $100 million in LearnVector, a new AI education company founded by Coursera co-founder and chairman Andrew Ng. The startup aims to use AI agents to provide personalized, one-on-one learning and help workers develop new skills as AI reshapes the workforce—marking a major bet on AI expanding rather than replacing the market for online learning.

Learn more here.

3. Washington Post Opinion: EdTech Backlash Shouldn’t Mean Walking Away From Technology

A new Washington Post opinion piece argues that the growing backlash against education technology risks oversimplifying a much more complicated question: how technology should be used in schools. Rather than abandoning edtech altogether, the piece calls for a more thoughtful approach centered on students and focused on distinguishing between tools that genuinely support learning and those that create distraction or harm.

Learn more here.

4. TIME Names the World’s Top EdTech Companies of 2026

TIME and Statista have released their 2026 ranking of the World’s Top EdTech Companies, highlighting 500 organizations across the global education technology landscape.

Learn more here.

Student-Centered Learning, AI, and School Redesign with Julie Young of Florida Virtual School

Julie Young is the Founding CEO of Florida Virtual School (FLVS), where she helped pioneer online public education and expand access to millions of learners. She later founded ASU Prep Digital and is the author of Say Yes: How Virtual Became Reality, continuing to shape the future of student-centered learning.

5 Things You’ll Learn in This Episode:

1. How Florida Virtual School grew from a small pilot into a global model for virtual education.

2. Why putting students and not systems at the center changes everything about school design.

3. Lessons for driving innovation within public education while earning trust and support.

4. How ASU Prep Digital is blurring the lines between K-12 and higher education.

5. Why AI should be a catalyst for redesigning education, not simply automating existing practices.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Exclusive: T4 Education Reveals Top 25 Shortlists for Second Annual Global EdTech Prize

Hey Insiders! We’re excited to exclusively break this story with T4 Education. We’ll be covering updates on the Global Edtech Prize as it progresses, through the winners being selected in January 2027.

T4 Education revealed the Top 25 shortlists today for the second annual Global EdTech Prize. Founded in 2025 by T4 Education, with the support of Owl Ventures and Digital Promise, the Global EdTech Prize recognizes trailblazing solutions that are driving change and grappling with the most crucial challenges in today’s classrooms.

Judged by educators themselves, the prize provides a vital platform to help companies and organizations scale their products, tools, or services and their impact worldwide. It celebrates EdTechs that are making a real difference to the next generation and champions innovative tech solutions that actually work for the people who matter most – educators, their students, and their communities.

The prize is awarded in three categories: Majors, Start-Ups, and Non-Profits, with winners announced live at the World Schools Summit in London on January 17, 2027.

Top 25: Majors Category

1. 5E Labs: Human-AI Inquiry Ecosystem for Immersive STEM Learning – The Education University of Hong Kong (China, Hong Kong)

2. AI Talent Shift – Globant (Argentina)

3. Amira Reading Suite – Amira Learning (United States of America)

4. Aprimora Educacional – Positivo Tecnologia S.A. (Brazil)

5. Benesse AI Tutor – Benesse Corporation (Japan)

6. Challenge Touch, Shinken Zemi Elementary School Course – Benesse Corporation (Japan)

7. ClearMath Elementary – Carnegie Learning, Inc. (United States of America)

8. Complete Literacy Solution – Renaissance Learning UK Limited (United Kingdom)

9. D2L Lumi – D2L Inc. (Canada)

10. Doctor of Education (Ed.D.) – upGrad Education Private Limited (India)

11. Duolingo English Test (DET) – Duolingo, Inc. (United States of America)

12. Efekta Agentic Teaching Platform – Efekta Education (United Kingdom)

13. HOCMAI – Galaxy Education Joint Stock Company (Vietnam)

14. Interplay Achieve – Interplay Learning (United States of America)

15. International Schools – Ray Education Group (China)

16. Kahoot! – Kahoot! (Norway)

17. Leverage Edu – Leverage Edu (India)

18. Lenses on Literature – Carnegie Learning, Inc. (United States of America)

19. LingoAce – LingoAce (Singapore)

20. Mesa Educacional – Positivo Tecnologia S.A. (Brazil)

21. OpenApply – Faria Education Group (United Kingdom)

22. PLUS Tutoring – PLUS Tutoring, part of Carnegie Mellon University (United States of America)

23. Stepful: Re-imagining Healthcare Education – Stepful Inc. (United States of America)

24. SwiftPal – ConveGenius Edu Solutions Private Limited (India)

25. Vectare – Faria Education Group (United Kingdom)

Top 25: Start-Ups Category

1. 21st Century Competencies Assessment – Beyond Education (France)

2. ¡A Leer en Vivo! – Ticmas (Argentina)

3. Area9 Rhapsode – Area9 Lyceum ApS (Denmark)

4. Class Saathi – TagHive Inc. (Republic of Korea)

5. Curriculum Linked – Virtual Work Experience – Developing Experts Ltd. (United Kingdom)

6. EDGE Tutor International – EDGE Tutor International (Singapore)

7. Edify Play – Edify Education (Brazil)

8. Ei NEEV – Educational Initiatives Private Limited (India)

9. eKidz – eKidz.eu GmbH (Germany)

10. ELSA School – ELSA Corp (United States of America)

11. EYouth – eYouth (Egypt)

12. Frizzle – Frizzle, Inc. (United States of America)

13. International GCSE Global Skills Projects – 8billionideas Ltd (United Kingdom)

14. International Internship Program – The Intern Group (United Kingdom)

15. Kinderpedia – Kinderpedia SRL (Romania)

16. kinship – Build Kinship Inc. (Canada)

17. Kollegio – Kollegio, Inc. (United States of America)

18. LifeJourney by Territorium – Territorium, Inc. (Mexico)

19. Mindjoy – Mindjoy Learning, Inc. (United States of America)

20. Peekapak – Peekapak Inc. (Canada)

21. Shaia – MentuLabs (Colombia)

22. Snorkl – Snorkl, Inc. (United States of America)

23. STEM Innovator’s Workbook Series – SKOOL21 PTE. LTD. (Singapore)

24. Tutoring and Training Services – Sparkl Edventure Pvt Ltd (India)

25. Wild Zebra – Wild Zebra (United States of America)

Top 25: Non-Profits Category

1. AlfaDigital Programme – Chamos – In Aid of Children of Venezuela (United Kingdom)

2. AprendoLab – Fundación para el Impulso de la Educación y la Cultura Reimagina (Chile)

3. Biofeedback Breathing Program – Rural Opportunity Institute (ROI) (United States of America)

4. Bridginality – Bridginality (Nigeria)

5. Darsel – Darsel (United States of America)

6. DECILE – Fundación Universidad Torcuato Di Tella (Argentina)

7. DIL AI English Tutor – Developments in Literacy (United States of America)

8. Escuela Blynk – Reaprende (Peru)

9. HackersGen – Fondazione Gianfranco Marilli ETS (Italy)

10. Her Path Initiative – Refugee Youth Education Hub (Kenya)

11. JA Boost – Pitch Master – JA Worldwide, Inc. (United States of America)

12. Lazuli – Learning Design Alliance, Inc. (United States of America)

13. MathVoice – PowerMyLearning, Inc. (United States of America)

14. Meghshala E Lessons Platform – Meghshala Trust (India)

15. Oral Reading Fluency (ORF) / Bhasha Saarthi – Wadhwani AI (AI unit of Lords Education and Health Society) (India)

16. Oxford Digital – Oxford University Press Australia and New Zealand (Australia)

17. PadhaiPal – DEVI Sansthan (Dignity Education Vision International) (India)

18. Pathfinders Institute on Infosys Springboard – Infosys Foundation USA (United States of America)

19. POVIR Educational Platform – EdCamp Ukraine NGO (Ukraine)

20. SchoolBox – Peddle iLabs Innovation and Impact Foundation (Nigeria)

21. Scratch Tactile Ecosystem – Sistema THEAD, SCCL (Spain)

22. She is Digital – Junior Achievement Americas (United States of America)

23. Top Parent – NudgED Trust (India)

24. UWEZO Academy – Great Hope Foundation (United Republic of Tanzania)

25. VL2 Storybook Apps and Creator Program – Motion Light Lab (Gallaudet University) (United States of America)

Next Steps

The Top 10 finalists for each of these three categories will be announced in September, with finalists invited to the World Schools Summit where they will showcase their work before an Expert Panel comprising leading figures in the technology and education sectors, including investors and policymakers from around the world.

The Expert Panel will narrow down the Final 3 per category who will present their products and tools before educators at the World Schools Summit. Educators will vote by secret ballot to determine the winner of each category, and they will be announced live on stage at the culmination of the summit.

For more information about T4 Education or the Global EdTech Prize, please contact Salman Shaheen at Carter Fleet – salman@carterfleet.com

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Why Kids Should Help Design EdTech: Inside the Joan Ganz Cooney Center’s Co-Design Movement

Michael Preston, Executive Director, Medha Tare, Senior Director of Research, and Liza Gak, Postdoctoral Fellow at the Joan Ganz Cooney Center, are leading researchers focused on child-centered educational technology. Together, they are helping edtech organizations design better products by combining learning science, participatory co-design, and thoughtful approaches to AI.

5 Things You’ll Learn in This Episode:

1. Why involving children early in product design leads to better edtech.

2. How the Joan Ganz Cooney Center’s co-design methodology differs from traditional user testing.

3. What children’s feedback reveals about AI tutors, reading buddies, and learning experiences.

4. Why learning science and child-centered design should guide AI-powered products.

5. Practical ways edtech teams can adopt co-design to build more engaging and effective learning tools.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

The World Cup: Can America Afford to Play?

By: Ben Kornell

The final matchup is set: on Sunday, the world’s biggest sporting tournament and multi-cultural experience will wrap up with the FIFA World Cup Final in New Jersey between Spain and Argentina. As a world soccer fanatic, I have been following every moment of the tournament and can’t wait for these final, glorious 90 minutes (or, perhaps, slightly more).

Throughout the tournament there has been another nagging story that goes beyond soccer fandom and connects with anyone who cares about children and opportunity. The United States is hosting this year’s World Cup, but in the US, the path into soccer, and all of youth sports, has become increasingly inaccessible and gated by income. This story has parallels that resonate across education as we balance governmental, nonprofit, and for-profit models to scale opportunity. In many ways, youth soccer in America has lost its way, and it offers a broader warning for education as states are increasingly advancing education marketplaces.

The Price of Play

Soccer is supposed to be simple. A ball, a patch of grass or concrete, a few kids, two makeshift goals. Around the world, that simplicity is part of the sport’s genius. It is why the game belongs as much to working-class neighborhoods as to national stadiums. But in the U.S., the youth soccer pathway often looks very different: club fees, tournament fees, travel fees, hotel stays, uniforms, private coaching, camps, showcases, recruiting services, and subscriptions.

And soccer is not alone. Basketball, baseball, volleyball, hockey, lacrosse, cheer, and other sports have all been pulled into the same gravitational field. Youth sports in America has become big business. The question is whether the business is still serving kids.

The rising cost of youth sports is showing up in national data. According to the Aspen Institute’s Project Play, the average U.S. sports family spent $1,016 on a child’s primary sport in 2024. That was a 46% increase since 2019, roughly twice the rate of inflation over the same period. When families counted a child’s other sports, Aspen found another $475 in spending, bringing the average annual sports experience to nearly $1,500 for one child.

Those averages hide a wide spread. Some families reported spending nothing. Others reported spending almost $25,000 on a child’s sports. Aspen also found that households earning $100,000 or more spent $1,471 more per year on a child’s primary sport than households earning under $50,000.

The burden is especially visible in soccer and basketball. Aspen’s youth sports cost data lists average annual costs of $1,188 for soccer and $1,002 for basketball, higher than baseball and tackle football among the major sports it compares. Aspen also notes that soccer, basketball, and baseball all have pay-to-play models that shape access.

Travel is one of the biggest culprits. Across sports, Aspen found that parents spent more annually on travel than on equipment, private lessons, registration fees, or camps. That average includes all children who play sports, not just the subset on travel teams, meaning the actual burden on competitive families can be far higher.

This is where the middle-class trap begins. A family starts with a local team. Then comes the invitation to a higher-level team. Then the regional league. Then the out-of-state tournament. Then private training because everyone else is doing it. Then the showcase because college coaches might be there. Then another season, another roster, another fee.

The language is always aspirational: development, exposure, pathway, elite. The bill is always real.

Access Is Becoming a Class Marker

The most troubling part of this story is not that some families spend a lot of money on sports. Families spend money on all kinds of activities, and sports can be profoundly valuable. Kids build friendships, confidence, resilience, physical health, and identity through teams. The problem is that access to those benefits is increasingly stratified by income.

Project Play cites Sports & Fitness Industry Association data showing that in 2021, only 24% of children ages 6 to 12 from households earning $25,000 or less played sports on a regular basis, compared with 40% of children from households earning $100,000 or more.

As educators, we know that sports are part of how children build social capital. They are where kids meet mentors, learn how to fail, practice discipline, and find belonging. For some students, sports are also a reason to stay engaged in school.

But when the best opportunities are attached to family purchasing power, youth sports start to resemble the rest of the American opportunity market: tutoring, test prep, enrichment camps, college counseling, private lessons. The families with more money can buy more development, more visibility, and more second chances.

Low-income families are often asked to make extraordinary sacrifices for a dream whose odds are long. A parent may pay for a travel team instead of tutoring. A family may skip savings to afford tournament hotels. A student may spend weekends chasing exposure while schoolwork becomes secondary. For some families, the hope of an athletic scholarship becomes not just a dream, but a (risky) financial strategy.

The NCAA estimates that of nearly eight million high school athletes in the United States, only about 560,000 compete at NCAA schools. In boys’ basketball, about 3.6% of high school participants go on to play at any NCAA level, and about 1.1% reach Division I. In boys’ soccer, about 5.9% reach the NCAA level, and about 1.4% reach Division I. The numbers are somewhat higher for girls’ soccer, but still narrow: about 7.9% reach NCAA soccer, and 2.8% reach Division I. The scholarship funnel is even tighter: only 2% of high school athletes are awarded athletic scholarships to compete in college.

While the path to a college scholarship or pro contract has always been slim, even competing for the high school varsity team often requires participation in pay-to-play leagues. Even if a student does not plan to play at a college level, that varsity sport extracurricular can be an important piece of their college application.

Soccer’s American Contradiction

The World Cup and the U.S.’s disappointing exit in the round of 16 has sparked a public conversation about this dynamic. The U.S. wants to be a soccer country. We have the fans, the stadiums, the sponsors, the media, the youth participation, and the commercial energy. But we also have a youth development model that filters talent through family income.

Former U.S. men’s national team star Landon Donovan said there was “zero chance” he could have afforded today’s youth soccer system if he were growing up now. Donovan was raised by a single mother and has said his own access depended on someone helping sponsor him.

There are wonderful community programs, school-based initiatives, nonprofits, rec leagues, and volunteer coaches all over the country. The problem is that the competitive pathway — the one many families are told they must enter if their child is “serious” — often runs through private clubs and travel competition.

Once a sport’s development ladder is privatized, the ladder starts to acquire toll booths.

Former USMNT defender and TV commentator Alexi Lalas has spoken out on the issue, to much controversy. He wrote on X that “youth sports is a competitive market with businesses selling a product that obviously customers are willing to pay for. I’d love it if soccer were free to all. But who is going to pay for all this free soccer? So they should be taxpayer-funded government entities or charities?”

Private Equity Enters the Chat

Private equity did not invent America’s pay-to-play sports model. The causes are broader: public disinvestment, weak recreational infrastructure, specialization, college admissions anxiety, parental fear of missing out, inflation, and the collapse or weakening of some local volunteer systems.

But private equity is accelerating and consolidating the model.

A 2026 White & Case analysis describes youth sports as a compelling growth opportunity for private equity because the sector is moving from fragmented, community-based operations toward nationally scaled platforms. The firm estimates the U.S. youth sports industry at about $40 billion annually, spanning club teams, travel tournaments, training academies, technology platforms, and purpose-built facilities. It also describes the market as growing 8% to 10% annually, driven by rising household spending, year-round specialization, and parental investment in competitive pathways.

In other words: the things that make parents anxious make the market attractive.

The investor logic is straightforward. Youth sports have recurring demand. Families pay season after season. Parents are emotionally invested customers and therefore have low price-sensitivity. Operators are fragmented. Technology can consolidate payments, scheduling, rosters, rankings, video, ticketing, and communication. Facilities can host tournaments, rent fields, sell concessions, and drive hotel nights. Data can feed recruiting. Streaming can monetize grandparents and out-of-town parents. Uniforms and apparel refresh every season. A child’s season becomes one large revenue funnel.

Some of the biggest names in finance are now connected to pieces of this ecosystem.

• KKR acquired Varsity Brands, whose platform supports school and team sports, athletics, and spirit programs and reaches more than eight million athletes and students annually.¹¹

• EQT acquired IMG Academy in a deal valuing the sports education company at $1.25 billion; IMG supports more than 100,000 student-athletes across boarding school, camps, online coaching, and college recruiting.

• Accel-KKR invested in LeagueApps, a youth sports management platform helping organizers manage clubs, tournaments, leagues, camps, and facilities.

• Genstar-backed PlayMetrics and Stack Sports combined to create a larger sports software platform, with PlayMetrics serving more than 2,700 youth sports organizations.

• 3STEP Sports, backed by Juggernaut Capital Partners, owns and manages youth sports clubs, events, and tournaments, sells apparel, and provides content, data, and analytics for colleges, media outlets, and athletic brands. In 2026, 3STEP described itself as serving nearly two million athlete participants across seven sports in more than 40 states.

A local team may still feel local. The coach may still care deeply. The parents may still volunteer. But the systems around that team — registration, payment, tournaments, facilities, uniforms, video, recruiting, and scheduling — are often part of a much larger commercial machine. Where it’s heading is private equity actually owning the club teams themselves.

The Rest of the World Offers a Different Default

No country has a perfect youth sports system. Europe has its own problems with access, facilities, elitism, and professionalization. But many peer countries start from a different premise: youth sport is more connected to community clubs, public facilities, nonprofit structures, and public or quasi-public funding.

In Norway, the national sports model channels 64% of gaming proceeds from the national lottery and sports betting mechanism into the sports ecosystem, roughly $400 million annually for projects. Much of that support goes back into communities for coaches, equipment, and facilities

In Germany, sport is organized through both public sports administration and self-governed clubs and associations. YouthWiki reports that clubs are the backbone of Germany’s sport movement, with about 86,000 sports clubs and 28.7 million members. Project Play notes that the German Olympic Sports Confederation receives hundreds of millions in federal funding and supports community sport through membership fees, lottery revenues, and marketing licenses.

The U.K. government’s Multi-Sport Grassroots Facilities Programme invested £125 million in 2024/25 to support facilities across the country, with an emphasis on deprived areas and underrepresented groups.

America has made a choice, whether intentionally or by neglect. We have allowed youth sports to become a largely privatized marketplace and then acted surprised when the marketplace behaved like a marketplace. Whether it’s through increased school funding for sports, governmental subsidies for families, or a National Youth Sports program akin to the IOC, we have many options to literally level the playing field.

A World Cup Legacy Worth Building

The 1994 World Cup helped grow soccer in America. The 2026 World Cup should do something harder: force America to ask what kind of sports culture it wants. Do we want a system where the next great player or even a recreational player has to be able to afford a club or do we want a system where every child has a real place to play?

In May 2026, Congressional lawmakers introduced the Let Kids Play Act, a bill aimed at banning private equity firms from youth sports and targeting practices such as junk fees and costly lock-in arrangements. Reasonable people can debate whether a ban is the right tool, but at the local, state, and federal level, this is becoming a bigger public policy issue.

The World Cup Final will be spectacular. While the outcome will cement global legacies, the real legacy of the World Cup itself will not be decided under the lights in New Jersey.

It will be decided by whether a child can join a team without her parents taking on debt. It will be decided by whether a talented player from a low-income family is seen early enough to be supported, not priced out. It will be decided by whether we treat youth sports as a luxury product or a public good.

Over the past month-a-half, the world’s game came to America. Now America has to decide whether the game belongs to all of our kids.

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Building an AI Tutor That Gets Students to Think with Sue Khim of Brilliant

Sponsored by Brilliant

Sue Khim is the Co-Founder and CEO of Brilliant, the interactive learning platform that has helped millions of learners build deep understanding in math, science, and computer science. In this episode, she shares how Brilliant’s new AI tutor, Koji, is designed to develop critical thinking rather than simply provide answers.

5 Things You’ll Learn in This Episode

1. Why the best AI tutors ask questions instead of giving answers.

2. How Brilliant applies active learning and game design to STEM education.

3. The philosophy behind designing AI that builds intelligence rather than dependence.

4. What personalization should look like in the next generation of AI tutoring.

5. Why learning should feel like a “climbing wall” instead of an answer machine.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

The Three Jobs of Teaching a Child to Read, and Where AI Fits In

Sponsored by Overdeck Family Foundation

By: Lin Ler

Lin Ler is a Stanford MBA and MA in Education candidate working at the intersection of business strategy, learning science, and frontier AI. With a background spanning management consulting, EdTech startups, and the social sector, Lin translates complex AI capabilities into practical insight for the future of learning.

On the 2024 Nation’s Report Card, 40% of American fourth graders scored below “basic” in reading, and the share of eighth graders below “basic” was the largest in the assessment’s 32-year history. Into that emergency walks AI, carrying the playbook that worked in math: tight feedback loops, hints before answers, a tutor that never tires — the design that lifted math scores by a third of a standard deviation for five dollars a student in Ghana. Whether that playbook transfers to reading is a genuinely open question, and the early trials split. In 2024, a randomized trial in Beijing found a chatbot helped kindergartners understand stories as well as a trained human did. A year later, a randomized trial across seven English schools found teenagers who studied texts with an LLM learned less than classmates who took notes. Two studies can’t settle anything, but the split between them is instructive, because the two chatbots were doing entirely different jobs. Teaching a child to read involves three jobs. It’s worth auditing each one and how AI fits into it.

Three Jobs, One Question

Watch a parent or teacher work with a young reader and you’ll see three jobs performed so fluidly they look like one.

The first is talking. The adult pauses mid-page and asks, “Why do you think the bear is hiding?” The child guesses, the adult builds on the answer, asks a follow-up, connects it to the child’s own life. The child does most of the talking; the adult steers. This dialogue around the text is where comprehension and vocabulary grow.

The second is listening. Now the child reads aloud, and the adult sits beside them tracking every word: catching the stumble on “through,” noticing that words with silent letters keep causing trouble, deciding whether to correct now or let this one go. The child produces; the adult diagnoses.

The third is supplying. Before any of that can happen, someone chose this book — hard enough to stretch this child, easy enough not to defeat them — and decided which questions were worth asking about it. The child never sees this job. It happens the night before, in the teacher’s planning hour.

Every AI reading product is a bet on at least one of these jobs. Chatbot reading companions bet on talk. Speech-recognition reading tutors bet on listening. Content generators bet on supply. The results diverge sharply enough that “does AI teach reading?” has no single answer. So let’s audit, job by job, what the research shows.

Job One: Talking With a Child

In reading, the conversation is the curriculum. Four decades of research on dialogic reading — the questioning technique Grover Whitehurst formalized in 1988 — shows that children’s language and comprehension grow when an adult asks questions during shared reading: prompting the child, evaluating the answer, expanding on it, repeating the prompt. The technique works. The bottleneck has always been the adult: many parents and teachers never learn it, and the children with the least access to trained questioners are the ones who fall behind first.

The news is that a machine can now carry that conversation. In the Beijing trial, published in the Journal of Experimental Child Psychology, 148 kindergartners from less-resourced families were randomly assigned to share a picture book with either a chatbot or a trained human, using either dialogic questioning or plain narration. These children are pre-readers; the book was read to them, and what the trial measured was story comprehension — how well a child understands a story they’ve heard, the foundation that reading comprehension is later built on. The dialogic technique drove the gains in comprehension and word learning, and it worked just as well from the chatbot as from the human. On one measure, learning four-character idioms, the chatbot-led sessions actually came out ahead. The control condition is just as instructive: the chatbot that simply narrated the story, no questions asked, produced the lowest comprehension scores of all four groups. The technology without the technique is inert.^[1]

This is a different mechanism from the one our math article found. In math, the effective AI is a gatekeeper: it withholds the answer the student is trying to extract, forcing the productive struggle. In storytime there is no answer to withhold. The effective reading AI is an initiator: it asks questions the child was never going to ask, and the child does the cognitive work of answering. The principle underneath is the same — learning happens when the child does the thinking — but the design move is the opposite. Math AI succeeds by refusing to talk. Reading AI succeeds by starting the conversation.

The product implications are already visible. StoryBuddy, a CHI-published system from the same research lineage, generates dialogic questions for any storybook rather than a preset catalog, and lets parents curate which questions get asked. Parents in non-English-speaking households saw a use case the researchers hadn’t led with: a grandparent who speaks no English can now supervise English storytime.

The finding stands as the first result of its kind in this series: an AI matching a trained human at a core teaching practice. The machine didn’t replicate the adult’s warmth. It replicated the adult’s questions. That turns out to be the part that teaches.

The Impostor Version of Job One

Here is the trap in Job One: there is a second kind of “AI reading companion” that looks identical from across the room — a student and a chatbot, talking about a text — but runs the conversation in reverse. In Beijing, the machine asked and the child answered. In the products teenagers actually use, the student asks and the machine answers. That reversal flips the learning result.

The English schools trial, run by Cambridge University Press & Assessment and Microsoft Research with 344 students aged 14 and 15, tested exactly this setup: students studied history passages with an LLM chatbot available to answer their questions, against classmates who took notes. Three days later, tested without any tools, the note-takers won on every measure — literal retention, comprehension, and free recall. The students hadn’t used the LLM foolishly; 90% asked for elaboration and deeper explanation, exactly the prompts you’d hope for. They lost anyway. Answering a question about a text forces you to select, connect, and put ideas into your own words — that effort is what makes reading stick. When the machine does the answering, the machine gets the practice.

The students couldn’t feel it happening. They rated the LLM easier and more helpful than note-taking, and preferred it 42% to 27%. One in six pasted the LLM’s output into their notes nearly verbatim. Out of 4,929 prompts across the study, exactly six questioned whether the LLM was right. Readers of the first article in this series will recognize the shape: cognitive debt in a reading jacket — the tool that feels most helpful teaching the least.^[2]

So the audit of Job One comes down to one observable: which way the questions flow. When they flow toward the child, the child does the thinking, and the learning follows. When they flow toward the machine, the machine does the thinking, and the learning goes with it. After ten minutes with any reading tool, count who has said more about the text — the child or the machine.

Job Two: Listening to a Child

The less glamorous job in reading instruction is sitting beside a child, hearing them read aloud, and catching the stumbles. It is also the job AI has been learning for the longest. Project LISTEN, launched at Carnegie Mellon in 1990, built a Reading Tutor that displayed stories, listened to children read aloud through speech recognition, and intervened the way an expert tutor would. In a seven-month comparison across 178 students in grades 1–4, students using the Reading Tutor outgained peers doing sustained silent reading on word identification, comprehension, and fluency, with the largest effects in first grade — and the lowest-proficiency students often gained most. The lineage predates chatbots by a quarter century, and it has been quietly compounding ever since.

The current commercial version of the idea looks like BuddyBooks, and its design is worth dwelling on, because every element maps to something reading science already validated. The platform co-reads: the machine reads a short section aloud with the words highlighted — modeling fluent pacing — then the child reads the next section, and speech recognition scores their accuracy and fluency in real time. Text is chunked into short segments to build stamina without overwhelm. Sections the child struggled with get flagged for Review Mode, where the child hears a fluent model, hears their own attempt, and rereads — which is repeated reading, the classic fluency intervention, automated. The machine never sighs, never rushes, never looks disappointed, which matters for struggling readers who have learned to dread reading aloud. In a 2024–25 study of 107 students with dyslexia in grades 2 through 5 in a Texas district, students who engaged consistently — at least ten focused minutes in most weeks — outgrew less consistent peers on i-Ready and posted nearly double the growth on the state reading test. The study is correlational, so it suggests rather than proves. But its design lesson stands on its own: total minutes didn’t track gains, and neither did total weeks. Only regular, focused practice did.

The capability underneath is improving fastest where trained assessors are scarcest. A 2025 study in South Africa fine-tuned speech models to score early-grade oral reading in Xhosa and reached roughly 91% diagnostic accuracy with a few hundred labeled recordings per test item; the same task given to a state-of-the-art multilingual model off the shelf scored 6.9%. Child speech in noisy classrooms breaks generic AI, and modest local investment fixes it. A growing research stack — from child-speech LLMs to automated fluency scoring in Tamil and Malay — is working the same problem across languages. For countries that assess early reading through oral tests administered by scarce trained examiners, this is a genuinely new capacity.

The honest summary of Job Two: the modern studies are descriptive, and no recent randomized trial certifies that today’s AI listeners cause reading gains. But this is the job with working products, a validated mechanism, and a thirty-year pedigree. Before AI can teach a child to read, it has to be able to hear them. In English, that problem is largely solved. Everywhere else, it is solvable and cheap.

Job Three: Supply the Text, Fluent but Aimless

The job AI looks most obviously built for is generating the material: passages, comprehension questions, leveled versions of the same text for a mixed classroom. This is where the audit turns up the quiet failure. Generation is solved. Pedagogical targeting is not.

Consider what happened when researchers at ETS asked GPT-4o to generate reading comprehension questions targeting specific inference skills — the diagnostic precision a struggling reader actually needs. Expert reviewers judged 93.8% of the generated questions good enough for operational use in grades 3 through 12. Impressive, until the second number: only 42.6% tested the inference skill they were asked to test. A third of the output drifted into literal, fact-lookup questions requiring no inference at all. The items were polished. They just weren’t measuring what they were supposed to measure, and a question bank that can’t target a skill can’t diagnose a struggling reader.

Text leveling shows the same signature. A 2025 evaluation had three LLMs simplify sixty 12th-grade passages down to 8th-, 6th-, and 4th-grade reading levels, across 2,315 outputs. Simplifying to 8th and 6th grade worked. At 4th grade, not one model-and-prompt combination reliably hit the target; the texts stayed too hard. And the further down the models reached, the more meaning they shed along the way. The pattern is exactly backwards from the need: the students requiring the deepest simplification — young readers and English learners furthest behind grade level — are the ones current models serve worst.

The mechanism is worth naming because it will outlast these particular models. LLMs control the form of educational content — clean questions, simpler words, fluent prose — far better than they control its cognitive demand, which is the property that makes content teach. In both studies, performance was rescued the same way: human structure injected into the pipeline, expert-written examples in one case, expert-chosen keywords in the other. That is not an argument against AI-generated reading material. It is an argument that “the AI wrote it and it reads fine” is the wrong acceptance test, and every reading product built on auto-generated content is currently relying on a human check it may or may not be doing.

The Report Card

Score the three jobs against the evidence and the market looks upside down.

Talk earns a split grade: the strongest causal result in the set when the AI asks the questions, delivered through a designed protocol — and a causally demonstrated harm when it answers them for an unsupervised teenager. A for dialogic design; F for the open chatbot.

Listen earns a B, incomplete. Deployable products, a solved capability problem, real-world state-test correlations, a thirty-year evidence lineage — and no modern causal study to certify it.

Supply earns a C minus. Unlimited fluent output, unreliable targeting, and a mandatory human in the loop that the economics of content generation quietly encourage everyone to skip.

The Question Worth Asking

Buyers ask whether a reading product “uses AI.” The audit says that’s the wrong question, and it supplies three better ones. Who does the talking? If a session ends with the child having produced fewer words about the text than the machine, the tool is doing the impostor version of Job One. Does it listen? A tool that never hears the child read aloud is not touching the job where AI is most proven. Who checks the text? If no human verifies that generated material hits the level and the skill it claims, the product is running on fluency and hope.

For builders, the collision is the same one the math evidence exposed, transposed into a new subject: the products that demo best — instant answers, instant content — are the ones the evidence grades worst, while the designs the evidence rewards are slower to build and harder to sell. The difference in reading is that the winning design isn’t about withholding anything. A child learns to read by talking about books with someone who listens. The best AI so far is the one built to remember that.

1. The Beijing chatbot was a scripted, purpose-built system that used speech recognition to classify children’s answers — not a generative LLM. What the trial licenses is structured dialogic prompting delivered by a machine, not handing a kindergartner ChatGPT. Outcomes were measured immediately after a single session; how long the gains last is untested.↩︎

2. The trial had no reading-only control, so it shows the LLM losing to an evidence-based study strategy, not to doing nothing. And the pattern may not hold for older learners: in a German laboratory experiment with 334 university students, unrestricted AI-tutor access during textbook study improved unaided test scores, and students kept reading anyway — with gains concentrated among students with strong self-regulation. Teenagers, midway through developing exactly that capacity, may be the population most exposed.↩︎

Special thanks to Overdeck Family Foundation for sponsoring this article in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative).

You can support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Top Edtech Headlines

1. KORA: The First Independent Child Safety Benchmark for AI Apps

A new independent benchmark, KORA, evaluates how popular AI tools perform on child safety, with leaderboards currently focusing on both models and apps.

Learn more here.

2. New Edtech Job Board by Old Soul

Our friends at Old Soul have launched a new Edtech Job Board! Peruse open roles or submit your open roles to be included.

Learn more here.

3. All Education Tech Purchases on Pause as NYC Schools Wrestle With AI

New York City Public Schools has paused all new edtech purchases while it reevaluates procurement policies in response to the rapid rise of AI. The move reflects growing efforts by districts to strengthen oversight around privacy, safety, and instructional value as AI becomes increasingly embedded in classroom technology.

Learn more here.

4. Apply Now: The Educator Index Report

As outcome-based contracts become more common and every investment faces greater scrutiny, Edtech organizations are under growing pressure to demonstrate real impact, not just usage. The companies that win renewals aren’t waiting for ARR and NPS to slip, they’re reading the early signals that predict outcomes and changing the trajectory before it’s too late.

The Educator Index (EI) is a research-based report that surfaces those early, educator-grounded signals of adoption, implementation health, and long-term impact, giving organizations actionable insights before challenges show up in renewal conversations. Thanks to a generous supporter, up to five EdTech companies will receive a fully funded Educator Index Report, including insights that can inform implementation strategy, customer success efforts, QBRs, board discussions, and strategic planning. Apply for a sponsored Educator Index Report and discover implementation risks before your next renewal cycle!

Learn more here.

How AI Coaching Is Transforming STEM Classrooms

We recently had Smita Saxena and Alexia Lewis on The Edtech Insiders Podcast!

Smita Saxena is the Founder of Maestro, an AI coaching platform helping K-12 STEM teachers improve instruction and student outcomes. Alexia Lewis is an 8th Grade Algebra I teacher at KIPP Polaris and a Maestro coach whose students achieved remarkable gains using the platform.

5 Things You’ll Learn

1. How AI coaching saves teachers hours every week.

2. Why human coaching + AI outperforms either alone.

3. How AI supports differentiated instruction.

4. Using AI to turn assessment data into action.

5. Practical ways AI can reduce teacher burnout.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Balancing the Equation: Why AI in Math Teaching Cuts Both Ways

Sponsored by Overdeck Family Foundation

By: Lin Ler

Lin Ler is a Stanford MBA and MA in Education candidate working at the intersection of business strategy, learning science, and frontier AI. With a background spanning management consulting, EdTech startups, and the social sector, Lin translates complex AI capabilities into practical insight for the future of learning.

Special thanks to Overdeck Family Foundation for sponsoring this article in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative). Stay tuned for more!

Picture this: A student in the West African country of Ghana—where the average annual income is $235 USD—sits down during study hall, opens WhatsApp and starts working through math problems with a chatbot named Rori. When she answers a question incorrectly, Rori does not correct her: it offers a hint, surfaces a misconception, gently guides her in the right direction… and only if she is still stuck does it walk her through the worked solution.

In a randomized trial across eleven schools, roughly 1,000 students in grades 3-9 used Rori for two short sessions a week… and pulled ahead of their peers. In fact, Rori demonstrated an effect size of 0.36, which in education research is close to an extra year of learning at a cost of about five dollars per student. Those results are real, affordable, built on existing technology, and, as with most tech-delivered solutions, scalable.

Research suggests that the core detail that explains the results is the order in which Rori, or other math tutors designed with pedagogical intent, delivers its inferences: hints and guidance first, answer last. The strongest evidence for math and science learning consistently shows that the AI that helps is the AI that refuses to simply hand over the answer and instead engages students in metacognition and productive struggle. Here’s the catch; the thing that general- purpose AI does best when asked a math question- that is, producing an instant correct solution- turns out to be the thing that most reliably reduces the learning for students.

Why Math Is Built for This

Ask an AI to grade an essay and you are asking it to make a judgment call. Ask whether a student correctly solved for x and the answer is simply right or wrong. That difference helps explain why math, not writing or history, has been the easiest place for edtech products to show a measurable win, which is why so many products focus on the domain. Math is also the domain in which the evidence for AI in education is strongest. A math problem has objective answers, decomposable steps, and diagnosable errors, so software, and now AI models, can tell a student exactly where her reasoning broke and what to try next, instantly, on every step.

According to a 2025 systematic review of 258 K-12 STEM studies, mathematics was the single largest subject of study, and research on intelligent tutoring systems (ITSs), tools built to give this kind of step-by-step guidance, accounted for nearly half of all the work. Many researchers note one key mechanism in ITSs: feedback density: ITS’s, and AI tutors, have the ability to create a learning loop that closes after every step, in which students get immediate, consistent and iterative feedback instead of waiting for a teacher who has the responsibility to reach, and grade, the work of an entire class .

That said, the same review found that only 34% of its studies bothered to report a standardized effect size. The evidence base is wide and shallow, not deep, but offers tantalizing insights into some of the core design principles that could make AI-driven education technology a game-changer for math education .

The Active Ingredient Is What You Withhold

Picture two AI tutors working with the same student on the same hard problem. One, when she gets stuck, explains the answer. The other refuses, asks her a question, and points at the next step she has to take herself. A Carnegie Mellon study of a university discrete-math course ran almost exactly that contrast. Its system had two parts: a chatbot that would give the answer if a student pushed for it, and a proof-reviewer engineered never to release the solution, only to flag where the reasoning went wrong.

The result is unsurprising; it is the same order that made Rori work: hints, Socratic questions, carefully designed guidance, all before solution. Students who used the answer-giving chatbot more often scored worse on exams, while students who used the answer-withholding reviewer more often scored better. This finding is correlational, not a clean randomized comparison, so it is best read as evidence consistent with a pattern rather than proof of one. But the pattern keeps recurring.

It is built into the physics tutor whose designers prompted it to “never reveal the full solution,” only the next step. And it shows up when an interactive, step-by-step math method was tested against a standard worked-solution explanation: on the harder problems, students judged the method that made them answer at each step more useful by a margin of 47.5% to 26.7%, and 78% preferred working that way.

This is one of the core paradoxes of education technology; the instructional design that best helps students learn is often the design that frustrates them in the moment. The implications for education technology companies, which are often incentivized to optimize for tool usage, user satisfaction, or reduced friction, continue to be difficult to navigate.

Just as carefully designed tools (like ITSs) are often less popular than the ones that simply give answers, this tension is exacerbated in an era when students can opt out of such designs and into commercially available “homework help” tools—prevalent in the Apple and Google app stores—that reveal the answer outright. That puts the evidence on a collision course with how most of these products are built and sold.

Confident, and Wrong

A student finishes a physics problem and gets feedback that is specific, fluent, and authoritative. It is also wrong, and she has no way to tell. The same physics feedback system from the previous section—the one its designers engineered never to hand over the full solution—was grounded carefully in an expert’s map of what a correct solution should contain, and yet its feedback was still wrong in about one case in five. What’s worse—the users may often not notice that the feedback is incorrect—in this case, the students testing it were Physics Olympiad participants, among the most capable in the country, and they caught those errors in only two of thirty-eight cases.

This is the catch hidden inside the feedback-density story. A loop that closes instantly is only worth having if AI responses are actually correct, and the confident, expert-sounding register is exactly what stops a student from questioning it. Part of this gap is temporary and will narrow as models get more accurate- we’ve already seen enormous jumps in accuracy in the last few years, especially in math. But part of the issue is structural: the polished, authoritative tone that makes AI feedback feel trustworthy is intrinsic to how these models write, and it will keep masking errors even as the errors grow rarer. Pair this with the previous section and the risk compounds. An AI that hands over an answer is bad for learning, and an AI that hands over a wrong answer with full confidence is worse. The tighter the loop, the more convincing the mistake.

Performance Is Not Learning

One of the most consistent, and important, findings in AI research is the gap between performance—how students perform on assessments or homework when assisted—and genuine learning, including retention of material and transfer to other environments. Then comes the exam, with the AI switched off. In the Carnegie Mellon study, while only the treatment group had the tutor, their homework scores rose significantly; on exams, the two groups were indistinguishable. The students who leaned hardest on the answer-giving chatbot were the ones who started with the least confidence, the students who could least afford to offload the thinking in the first place.

The sharpest version of this comes from a field experiment with nearly a thousand high-school math students, which randomized them between two GPT-4 tutor experiences: the first was a plain, ChatGPT-style interface that would hand over answers, and the other was prompted to safeguard learning by withholding answers. While students had access, both lifted performance sharply—a 48% gain for the answer-giving version, 127% for the safeguarded one. But when the AI was switched off, the students who had used the answer-giving tutor scored 17% worse than classmates who had never touched it. The researchers’ explanation for what happened is a concept running through this whole series: when students had used the model as a “crutch,” their performance did not predict their deeper and sustained learning- but intelligently designed safeguards can keep the crutch from forming.

A caveat belongs here. Even the Ghana result is not a clean test of durable learning, because students were still using Rori when they were assessed, and there is no high-quality causal study of durable, unassisted learning in US K-12 classrooms anywhere in this evidence. What the research can say is narrower and still important: AI reliably improves student performance while students have access to it, but whether it improves students’ understanding, retention and transfer after the AI is removed is an area of active study.

AI reliably improves student performance while students have access to it, but whether it improves students’ understanding, retention and transfer after the AI is removed is an area of active study.

That said, there are some early encouraging signals on transfer; one UK-based RCT conducted by Google, LearnLM and Eedi in math found that “students who received support from LearnLM were 5.5 percentage points more likely to solve novel problems on subsequent topics (with a success rate of 66.2%) than those who received tutoring from human tutors alone (rate of 60.7%).” Even more noteworthy from the same study was the fact that the human tutors overseeing the AI tutor approved more than ¾ of the AI-generated messages either without changes or with only minimal edits; the implication is that AI tutors, when designed carefully, may already be generating pedagogically sound content the majority of the time.

Evening the Playing Field: The Expertise-Reversal Effect

In a study of 661 high-school physics students, students who started with weaker skills gained the most from richer, multi-format feedback from AI, while the advantage shrank for students who were already strong. Education researchers call this the expertise-reversal effect: the same scaffold that helps a novice can slow down an expert who no longer needs it. The interactive-math study found the same shape from the other direction, with step-by-step prompting helping on hard problems but acting as “a distraction” on easy ones.

The same pattern, in which support is concentrated where it is needed most, appears when the AI coaches the tutor instead of the student. In a randomized trial with more than 700 tutors and 1,000 students from underserved communities, a Stanford designed tool called Tutor CoPilot fed expert moves to tutors in real time. Students whose tutors used it were four percentage points more likely to master a topic… but for those assigned the lowest-rated tutors, the gain jumped to 9%. In other words, lower-performing tutors got more of a boost from AI than others- the expertise-reversal effect in action. An analysis of 350,000 messages showed the mechanism: the tool pushed tutors toward more probing questions and away from generic praise—nudging humans, at about twenty dollars per tutor a year, toward exactly the answer-withholding pedagogy the rest of this evidence rewards.

If the students using AI support end up being the ones it helps most, not least, this complicates the negative findings about AI for learning. But for lower-performing students, the harder problem may be getting them to use it at all: in the physics study, students left the elaborated feedback unopened on two-thirds of their attempts. So the real question is not whether AI helps “students” as a group. It is which students, on which problems, and whether those being helped are building understanding or quietly leaning on a crutch they will not have on test day.

The Question Isn’t Whether AI Can Do the Math

The question of whether AI can handle K-12 math is effectively closed; it clearly can. The real challenges are determining whether any given AI math tool actually helps the student: does it solve problems for them and deprive them of productive struggle? Is its feedback both accurate and actionable? Are students genuinely learning or merely leaning on the tool to perform? While current evidence consistently favors the role of guidance over direct answers, there is a research gap: the cleanest test we have—the GPT Base versus GPT Tutor experiment described earlier—pitted a bundle of learning safeguards against none, rather than isolating the single variable of withholding answers. That clean trial still doesn’t exist, so our conclusions remain well-supported hypotheses rather than settled facts

For developers, this creates a difficult paradox: the most marketable products (and those dominating the App Stores) provide immediate answers, yet the most effective educational design withholds them, admits uncertainty, and truly challenges the student at the right level. These are fundamentally different products, and the latter is both harder to build and harder to sell.

Ultimately, the challenge of AI in education transcends the math itself; it rests on whether the EdTech community can internalize this emerging research and design products that prioritize productive struggle over convenience, finally shattering the seductive illusion of learning to cultivate true, resilient student understanding.

Studies Cited:

Adewumi, T., Liwicki, F. S., Liwicki, M., Gardelli, V., Alkhaled, L., & Mokayed, H. (2025). Findings of mega: Maths explanation with llms using the socratic method for active learning. arXiv preprint arXiv:2507.12079. Link

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26). Link

Chen, E., Judicke, S., Beigh, K., Tang, X., Xiao, Z., Li, C., ... & Koedinger, K. (2025). Generative AI alone may not be enough: Evaluating AI Support for Learning Mathematical Proof. arXiv preprint arXiv:2509.16778. Link

Henkel, O., Horne-Robinson, H., Kozhakhmetova, N., & Lee, A. (2024). Effective and scalable math support: Evidence on the impact of an AI-tutor on math achievement in Ghana. arXiv preprint arXiv:2402.09809. Link

Maus, H., Tschisgale, P., Kieser, F., Petersen, S., & Wulff, P. (2025). Developing and Evaluating a Large Language Model-Based Automated Feedback System Grounded in Evidence-Centered Design for Supporting Physics Problem Solving. arXiv preprint arXiv:2512.10785. Link

Memari, M., & Ruggles, K. (2025). Artificial Intelligence in Elementary STEM Education: A Systematic Review of Current Applications and Future Challenges. arXiv preprint arXiv:2511.00105. Link

Revenga-Lozano, N., Avila, K. E., Steinert, S., Schweinberger, M., Gómez-Pérez, C. E., Kuhn, J., & Küchemann, S. (2026). Personalized Multimodal Feedback Using Multiple External Representations: Strategy Profiles and Learning in High School Physics. arXiv preprint arXiv:2601.09470. Link

Wang, R. E., Ribeiro, A. T., Robinson, C. D., Loeb, S., & Demszky, D. (2024). Tutor CoPilot: A Human-AI approach for scaling real-time expertise. arXiv preprint arXiv:2410.03017. Link

Special thanks to Overdeck Family Foundation for sponsoring this article in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative). Stay tuned for more!

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Top Edtech Headlines

1. Anthropic Launches 1,000-Fellowship Program to Support AI for Public Good

Anthropic has committed $150 million to launch the Claude Corps Fellowship Program, a new initiative that will support 1,000 fellows working on AI applications in education, government, research, and nonprofit sectors.

Learn more here.

2. Lawsuit Against Social Media Companies Advances as Debate Over Youth Mental Health Intensifies

A major lawsuit brought on behalf of families and school districts is moving forward against several social media companies, alleging that platform design choices contributed to harms experienced by young people. The case comes amid increasing scrutiny of technology’s impact on youth well-being and could have significant implications for how digital platforms are regulated, designed, and used by children and adolescents.

Learn more here.

3. New AI Literacy Framework Aims to Define Essential Skills for the AI Era

A coalition of education, research, and technology organizations has launched the AI Literacy Framework, a new resource designed to help educators, students, and institutions build a shared understanding of the knowledge and skills needed to engage with AI responsibly. The framework outlines key competencies ranging from technical understanding to critical thinking and ethical decision-making, providing a roadmap for integrating AI literacy into teaching and learning.

Learn more here.

4. SuperCharger Ventures Opens Applications for Malta-Based EdTech Accelerator

SuperCharger Ventures is accepting applications for Cohort 7.0 of its Malta Accelerator, a 12-week, in-person program designed to help EdTech and Future of Work startups expand into Europe. Participants gain access to mentorship, investor connections, corporate partners, and up to €1.5 million in non-dilutive funding opportunities, with select companies also eligible for direct investment from the SuperCharger Ventures Fund.

Learn more here.

Preparing Students for 2076: Ben Kornell on AI, Assessment, and the Future of Learning

In this special crossover episode, Edtech Insiders host Ben Kornell joins Allison Salisbury, author of The Humanist, for a wide-ranging conversation about AI, education, workforce development, and the future of learning. Together, they explore how AI could reshape assessment, personalized learning, school design, and the skills students will need to thrive in a rapidly changing world.

5 Things You’ll Learn in This Episode

1. Why AI should be used to redesign education and not just make existing systems more efficient.

2. How assessment could shift from annual testing to continuous, actionable feedback.

3. The durable skills students need to succeed in a world of constant technological change.

4. What the personalized learning movement got right and wrong.

5. How AI, school choice, and new funding models could reshape the future of K-12 education.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

There’s More to AI Grading Than Scoring

Sponsored by Overdeck Family Foundation

By: Lin Ler

Lin Ler is a Stanford MBA and MA in Education candidate working at the intersection of business strategy, learning science, and frontier AI. With a background spanning management consulting, EdTech startups, and the social sector, Lin translates complex AI capabilities into practical insight for the future of learning.

Special thanks to Overdeck Family Foundation for sponsoring this article, the third of seven in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative). Stay tuned for more!

When gauging the value of an AI grading tool, typically there is an appeal to scoring accuracy: the AI scores student work as reliably as a trained human grader. That claim, in many cases, is true. But, it’s not showing the whole picture.

In a pre-registered randomized trial across 178 public schools in Brazil, roughly 19,000 students preparing for the national university entrance exam were split into two groups: one had AI score their essays, the other had AI an scorer plus human graders reviewing every essay. Both groups of students improved their scores by about a tenth of a standard deviation. For structured scoring against a clear rubric, AI performs at the level of a trained human reviewer.

However, feedback on student work is different from scoring, and it’s generally lumped into the tasks a “grading” tool will complete. Feedback requires identifying what a specific student got wrong in the context of that student’s specific reasoning, and explaining what to do differently. This is a diagnostic act.

When a school adopts an “AI grading” tool, it’s generally implemented to support both of these tasks: scoring and feedback. But, most evaluations for AI grading tools only measure the effectiveness of the first task. And, these tasks are often performed most effectively by different LLMs. A 2024 study testing three AI systems on middle-school math open-ended responses found that the model that scored most reliably produced feedback teachers chose only 12% of the time. The model whose feedback teachers trusted most — GPT-4 — turned out to be the worst scorer of the three models tested. The model teachers wanted to use for explaining errors to students was not the model that scored accurately. They were optimized for different things.

This lumping together of scoring and feedback into the broad category of “grading” opens up risk for incorrect information and ineffective results. Let’s dive into it.

When AI Grading Works Best: Scoring and Feed Forward Feedback

Before examining where AI grading fails, let’s be precise about where it is working.

AI scoring is reliable when the tool is properly configured. “Properly configured” means the model is given the rubric, sample responses, or domain-specific training before any scoring occurs. Without this, performance collapses. A benchmarking study of LLM scoring across 12,978 standardized essays found that GPT-4 when given a student essay with no rubric or examples — what researchers call zero-shot prompting, meaning the model receives no scoring framework and no sample responses — produced grading accuracy near chance on some essay sets. The same model, given structured guidance and example essays, approached human-level accuracy. This is not a difference in AI capability, it is a difference in how the tool was set up. Before evaluating any AI grading product, the first question should be: how was this model configured for this specific task?

AI is also consistently well-suited to one category of feedback: what feedback researchers call Feed Forward, meaning next-step guidance. Telling a student to “cite a counterargument,” “separate your variables,” or “develop this paragraph with evidence” is scalable, consistent, and context-independent — it requires knowing what good work looks like, not what this particular student misunderstood. A 2025 study comparing LLM and teacher feedback on student science work found LLM and teacher Feed Forward feedback statistically indistinguishable. At generating next steps, AI performs at the level of an experienced teacher.

Beyond Feed Forward Feedback: What AI Catches vs. What a Teacher Catches

While AI is great at offering Feed Forward notes, supporting students beyond this gets more complicated.

The best study to elucidate this tested more expansive AI feedback directly on real student work. Researchers at Technical University of Munich gave the same student science protocols to an LLM and to experienced biology teachers, then had expert reviewers rate all the feedback blind. One of the protocols involved a pine cone humidity experiment. A common student error: proposing to study both pine cone shape and size simultaneously without isolating either variable, which means any result would be uninterpretable.

The LLM’s feedback told the student to separate their variables and test one at a time. That is the correct next step. It is the kind of rubric-aligned guidance that AI generates reliably.

A teacher reading the same protocol addressed the specific reasoning that produced the problem: “You’ve set up your study so that shape and size both vary. If you get a result, you won’t be able to say whether shape caused it, size caused it, or some combination. To fix this, pick one variable and hold the other constant.” That second version names what the student actually did and why it creates an interpretive problem.

When the expert reviewers rated all this feedback, overall quality scores were nearly identical — LLM at 3.78 versus teachers at 3.81 on a five-point scale. On the surface, equivalent. But when researchers measured content correlation — how much the LLM and teacher feedback addressed the same things in the same student work — the correlation was 0.06 to 0.21. Near zero. The language correlation, measuring how similarly they wrote, was 0.54 to 0.61. In other words, the feedback sounds alike, but they actually offer meaningfully different things.

On Feed Back support — identifying errors grounded in the student’s actual reasoning — LLM scored 3.05 versus teachers’ 3.52, a statistically significant gap. AI is good at telling students what to do next, but it is structurally weak at telling them what they specifically did wrong and retracing their steps.

The Risk in AI Feedback Without Review: The 20% Inaccuracy Gap

The surface similarity between AI and teacher feedback — fluent, confident, appropriately structured — creates a specific risk when no human reviews the AI’s output before it reaches students.

A 2025 study from the Leibniz Institute for Science and Mathematics Education built what is arguably the most carefully designed AI feedback system in the current published literature: an LLM-based system for physics problem solving, grounded in Evidence-Centered Design — a framework that anchors AI feedback to explicit knowledge types and problem-specific reasoning schemes rather than letting the model generate freely. It was built specifically to reduce the kinds of errors that unstructured prompting produces.

The researchers deployed it with 38 participants in the German Physics Olympiad and had expert reviewers check every piece of AI-generated feedback for physical correctness. In 51 of 64 feedback instances, roughly 80%, the feedback was accurate. In the remaining 13, it contained errors ranging from minor to substantial: wrong physics concepts, invalid assumptions, correct student approaches flagged as wrong.

Then came the question that matters for deployment: could students tell the difference?

Students rated the accuracy of the feedback they received at 4.4 out of 5. Whether the feedback was correct or incorrect made no statistically significant difference. Students receiving wrong feedback rated it just as accurate as students receiving right feedback.

Language models are trained to produce fluent, expert-sounding text. When an AI feedback system makes an error, it does not hedge or signal uncertainty. It produces the same confident, well-structured paragraph it would produce if correct. Students have no external signal to distinguish the two — and these participants were high-achieving physics competitors, among the most likely to catch an error if one was detectable.

Supporting, Not Replacing Teachers

None of this is an argument against AI in grading. It is an argument about what role AI plays well, and where it requires a human in the chain.

Let’s return to Brazil for a moment. The finding that human graders added nothing to AI scoring is not an argument for removing humans from the feedback loop. The Brazil RCT found that when AI handled essay scoring, teachers redirected their time: students in AI-supported schools had roughly 30% more one-on-one conversations with their teachers about their writing, with no increase in teacher workload. The AI did not replace the diagnostic role. It freed the teacher to perform it.

The Tsinghua co-grading experiment shows the same logic from a different direction. In a controlled study, novice graders — college students with no grading experience — achieved a grading accuracy score of 0.53 out of 1.0 against expert judgment. The same novices with LLM assistance reached 0.66, a level statistically indistinguishable from expert performance at 0.71. The AI’s rubric scaffolding and explanations closed a gap that training alone had not.

Expert graders with AI assistance showed a different pattern: accuracy improved but not significantly. What they found useful was the AI’s confidence scores for flagging essays that warranted closer review. The AI sorted and flagged areas for human expert decisions.

A tension in these two findings is worth naming. Adding experienced human graders on top of AI scoring produces no incremental benefit; adding AI assistance to novice graders produces significant gains. These are not contradictory: the teachers in the Brazil study were already near the ceiling of what essay scoring can achieve — the AI freed their time rather than improving their marks. The novice graders in the Tsinghua experiment started well below that ceiling; the AI’s scaffolding lifted them. AI’s amplification value is largest where the human expertise gap is largest.

Where the Loop Must Close

Teacher-in-the-loop is not a compliance requirement. It is the error-correction layer that students cannot provide for themselves. Here are three design choices for implementing AI grading tools in the classroom that evidence supports with this margin of error in mind:

• Review Before Delivery: In complex or high-stakes domains, AI-generated feedback should reach a teacher before it reaches a student. The 20% error rate in the Leibniz study came from a carefully ECD-grounded system with explicit evidentiary scaffolding. Without structured grounding, error rates are higher, not lower.

• Divide Labor by Feedback Type: AI handles Feed Forward — consistent, scalable next-step guidance. Teachers handle Feed Back — diagnosing a student’s specific reasoning error in a student’s specific work example. This is reallocation by feedback type, not just by volume.

• Match the Tool to What Your System Actually Uses: Assessment cultures differ in how much they rely on iterative feedback to drive improvement. In high-stakes exam contexts — where a score is the primary outcome and formative feedback plays a smaller role in the learning cycle — AI’s strength at accurate, consistent scoring is directly aligned with what the system most needs. Research on AI grading adoption across contexts notes the mismatch explicitly: feedback tools built for formative use show limited uptake in summative systems.

The Question Worth Asking

The debate about AI grading has been organized around the wrong question. “Is it as accurate as a human teacher?” is not wrong, it is incomplete.

Accurate at scoring structured work with proper configuration: yes, the evidence is consistent. Accurate at diagnosing what a specific student got wrong in their specific reasoning: no — not reliably, not without human review in complex domains, and not in ways students can independently verify.

Scoring accuracy and feedback accuracy are not the same thing. They are not measured by the same test, and a tool can pass one and fail the other. Most evaluations run only the first test and draw conclusions about both.

Before deploying an AI feedback tool, three questions are worth asking:

• Which job is this tool doing — scoring, feedback, or both?

• If feedback, who reviews it before it reaches students, and do they have the domain expertise to catch errors?

• What does the teacher do with the time the tool frees — and is the school designed for that reallocation to happen?

Schools have learned to ask whether the AI scores accurately. The question they haven’t started asking — whether it explains accurately — is the one that matters to the student trying to understand what went wrong.

Special thanks to Overdeck Family Foundation for sponsoring this article, the third of seven in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative). Stay tuned for more!

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

Top Edtech Headlines

1. Bett Has Launched in the U.S.

Backed by 100+ founding educator and industry leaders — including 50+ Launch Partners and 50+ Bett Builders — Bett USA is launching with momentum from across the K–12 ecosystem.

Taking place November 8–10, 2027 at the Music City Center in Nashville, Bett USA is the new national home for innovative K–12 educators — bringing together the leaders who set direction, evaluate solutions, and bring learning to life alongside the companies building what’s next.

Learn more here.

2. Request for Proposals: AI in Math Education

Applications are now open for a new Request for Proposals from the K-12 AI Infrastructure Program. This RFP focuses on developing open-source Gold Standard datasets, evaluations and benchmarks to improve AI in math education.

Up to $2M for Adaptive Learning Experiences & Feedback for Students (1 award), up to $1.5M for Lesson and Instructional Planning for Teachers (1 award), and up to $2M for Teacher Coaching (1 award).

Learn more here.

3. Apply Now: Global EdTech Prize

The Global EdTech Prize, founded with the support of Owl Ventures, has opened applications for edtech tools, products, and services online in anticipation of the next prize which will be awarded in January 2027! Applicants will be judged on purpose, quality, impact, innovativeness, and scalability.

Learn more here.

4. Edtech Insiders Co-Founder Ben Kornell at NewSchools Summit!

Our very own Ben Kornell recently spoke as part of the "Visions for What Comes Next" morning plenary at NewSchools Summit 2026, examining how AI is forcing a rethink of assessment and the purpose of school. Check out his talk below!

Lisa Gevelber on Why AI Fluency Is the New Workforce Skill

Lisa Gevelber is the founder and leader of Grow with Google, Google’s economic opportunity initiative focused on helping people build skills, careers, and businesses. Through Google Career Certificates and AI training programs, she has helped millions of learners develop workforce-ready skills and navigate an increasingly AI-driven economy.

5 Things You’ll Learn in This Episode:

1. Why AI fluency is becoming an essential workforce skill for every profession.

2. How Grow with Google is helping learners build practical AI skills at scale.

3. The growing role of AI credentials and certificates in higher education.

4. How universities are integrating AI training into student, faculty, and alumni learning.

5. Why lifelong learning and career mobility will be critical in the age of AI.

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

The MBA, Redesigned for the AI Era

Sponsored by MasterClass

Most AI training is built to help people catch up. MasterClass Executive is built to help them pull ahead.

The program covers the same core disciplines as a traditional MBA, finance, strategy, marketing, leadership, but reframed for a world where the question isn’t whether you can do the math, it’s whether you can ask the right question, evaluate the answer, and apply it. A custom AI tutor from OpenAI personalizes case studies to each learner’s industry and role.

Instructors include Ray Dalio, Mark Cuban, Paul Krugman, plus professors from Chicago Booth, Stanford, UCLA, and Oxford. There’s an in-person experience at Booth and live AI labs with operators from OpenAI, ElevenLabs, and Weights & Biases.

Apply now. It’s easier than ever to finish. It takes 5 minutes and there’s no additional preparation required.

Apply Now

What Higher Ed Now Demands From Its Vendors in 2026

By: Samvitha Ram & Jesse Victoroff

This piece appears concurrently on Mission to Margin’s Substack.

This piece was co-authored with Samvitha Ram & Jesse Victoroff (Co-Founders of Mission to Margin and seasoned advisors to higher education leaders). If you’re a college leader navigating a shifting financial and technological landscape or an edtech operator strategizing your future roadmap, we want to hear from you. Let us know how these questions resonate, which hits closest to home for your organization, and what you’re most worried about in the upcoming academic year.

Subscribe to Mission to Margin for higher ed strategy and institutional execution.

In a single month this spring, we saw three fronts squeeze American higher education:

1. Financial: By April 2026, eight more private nonprofit colleges announced they would close, on top of sixteen that closed in 2025

2. Regulatory: The same month, following a year of steep federal funding cuts that have seen NSF lose 30-40% of its staff and thousands of research grants rescinded, the Trump administration dismissed every member of the National Science Foundation’s governing board.

3. Pedagogical: Meanwhile, a Yale faculty committee tasked with restoring public trust in higher education recommended that classrooms across the university go device-free by default - no phones, no laptops, no tablets.

These are signals of a broader theme across the sector, where universities are under increasing pressure to rethink their core operating model, attract students in a moment of shrinking enrollment, and deliver outcomes that can justify (or exceed) the ever-high cost of attendance. These pressures have re-written what institutions buy, and from whom.

The procurement model that defined the last decade of edtech is gone. The office of the CIO rarely buys for the office of the Provost. Now the room hosts the CFO, the general counsel, and a skeptical faculty senate, plagued with the baggage of closing programs, OCR complaints, and generations of failed edtech initiatives.

The most useful questions for edtech founders right now are not about features, but whether the product meets the most pressing needs of the industry. Three such questions follow directly from where higher ed actually is in 2026.

Does your product directly improve the top- or bottom-line?

2026 is shaping up to be one of the worst years for college closures since the pandemic. The Hechinger Report and Higher Ed Dive are tracking eight announced 2026 closures so far (e.g., Hampshire, Anna Maria, Lourdes) on top of the sixteen nonprofit closures in 2025. Huron Consulting estimates more than 120 institutions are at the highest closure risk and roughly 442 are at meaningful risk.

Federal funding has gone from stable to shrinking and uncertain. The NSF board firing in April was most visible, but the underlying picture is bleaker: NSF staff is down 30-40% since the start of 2025, the administration has rescinded thousands of approved grants, and HHS, Title IV, and indirect cost rate negotiations are all adding downstream pressure.

Latest news from Inside Higher Ed and representative of a sector wide disruption

States and institutions are cutting spending visibly and publicly. Indiana eliminated or suspended 210 degree programs across its public universities and merged 375 more. Syracuse offered voluntary buyouts to 175 professors. HACC cut 120 positions. 86% of college presidents told ACE just 18 months ago that they were worried about their institution’s long-term financial viability. The data since has only made the case for worry stronger.

Institutional “innovation budgets” are all but gone. Vendor partnerships must keep core operations stable, including and especially those with edtech. Your platform is on the chopping block if it costs money and does not visibly defend revenue or take out cost.

How do you build trust around AI when faculty see it as a threat?

Nearly every edtech product launched in the last ~24 months has AI in its stack. AI has not only been a selling point to otherwise frazzled college administrators trying to “incorporate AI” in their tech roadmap, but in many cases, it’s been a way to upgrade users’ experience.

But there’s a huge adoption gap in AI. The Digital Education Council’s 2026 LATAM survey put student adoption at 92% versus 79% for faculty. 50% of students want AI-assisted feedback on assignments, but only 19% of faculty currently provide it. DEC’s 2025 Global Faculty Survey found that while 61% of faculty used AI in teaching, 88% of those used it minimally.

From DEC’s 2025 Global Faculty Survey - faculty remain cautious about AI usage

The numbers obscure the more important point: faculty conversations about AI are dominated by detection, guardrails, and policy. 65% of students in the HEPI 2026 survey reported that assessment has changed substantially in response to AI, and false-misconduct anxiety is rising on both sides of the syllabus. To a typical faculty member, AI threatens integrity and replacement first, and supports pedagogy somewhere distantly thereafter.

Partners who pitch AI as a feature may get blocked at the faculty senate even when they’ve already proven their value to other key decision makers (e.g., the CIO). Edtech partners will get through only if they can pitch AI as something that makes faculty look more credible, more in control, and less buried in administrative work.

Test these outcomes: time saved per week, integrity disputes reduced, prep time displaced. If you can point to cool features, but not impact for the faculty you serve, you’ve already lost your pitch.

Has the technology earned its place in a classroom?

Following a similar trend as in K-12, the public is now rejecting the assumption that more technology means better learning. In its April 10, 2026 report, Yale’s Committee on Trust in Higher Education recommended classrooms go device-free by default (i.e., no phones, laptops, or tablets). President Maurie McInnis accepted the recommendation and directed her deans to implement it. The report explicitly cited Jonathan Haidt’s Anxious Generation research and framed tech-driven distraction as a core threat to the academic mission. The same report tied classroom device culture directly to the broader collapse in public trust in higher education, from 57% in 2015 to 36% a decade later.

The screen-free trend may be seeing greater adoption by students themselves

Yale is not an outlier. A growing roster of provosts is arriving at the same conclusion from different angles: the classroom needs to recover its irreplaceable interpersonal culture. Unmanaged technology is in the way.

Faculty are tech-fatigued too. A typical instructor today has to navigate an LMS, a grading platform, an analytics dashboard, an AI tutoring backend and a content library, often with minimal integration or training. Given campus feedback, the procurement question has shifted from “what should we add?” to “how much can we cut?”

The bar has risen to add any technology to the classroom. Products that compete for student attention or add to faculty cognitive load are losing cultural permission, regardless of promised outcomes.

What this means, and what to do about it

Pull the threads together, and it’s clear that higher ed procurement in 2026 will be defensive, not aspirational. The buyer is more financially constrained, more regulatorily exposed, and more concerned about technology itself.

This doesn’t mean that edtech is out of the classroom. It means that edtech founders and operators need to take the concerns of higher ed leaders more seriously, responding to the three-fronts that have squeezed the sector this year and beyond.

For provosts and CIOs, the 3 questions above are a procurement and renewal filter. Walk every vendor and partner on your current contract list through them. The ones that hit the mark across all 3 have earned their place and are worth protecting in this budget cycle.

For edtech founders and operators, the questions are a gut-check for your product. You should be able to answer them with named institutional case studies, outcomes with dollar figures, faculty testimonials that go beyond satisfaction surveys, and evidence for learning outcomes. If you don’t have them yet, it’s time to take a serious look at the product roadmap.

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

David Rogier on the New “Masterclass Executive”

Sponsored by Masterclass

We recently had David Rogier, Founder and CEO of MasterClass, on The Edtech Insiders Podcast! We dove into the future of AI-powered learning, the unbundling of higher education, and the launch of the new MasterClass x Booth School of Business AI Accelerator developed alongside OpenAI, exploring what AI-native business education could look like for the next generation of professionals

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

How AI Is Changing Teaching Workflows

Sponsored by Overdeck Family Foundation

By: Lin Ler

Lin Ler is a Stanford MBA and MA in Education candidate working at the intersection of business strategy, learning science, and frontier AI. With a background spanning management consulting, EdTech startups, and the social sector, Lin translates complex AI capabilities into practical insight for the future of learning.

Special thanks to Overdeck Family Foundation for sponsoring this article, the second of seven in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative). Stay tuned for more!

Teachers in the U.S. spend an average of seven hours per week on lesson planning alone, plus another three for students with diverse needs. That’s before grading, parent emails, IEP documentation, and the administrative overhead that has made teaching among the highest-burnout professions in the country. When AI began to enter classrooms, it arrived carrying an unsettling question: would it replace teachers entirely? What the evidence has since made clear is that the answer is no. But a quieter, more consequential question has taken its place: can AI help in reducing teacher burnout? This question is receiving less attention than the panic around AI emptying teachers from classrooms, although it is much more pressing. Attrition and administrative overload are already emptying classrooms of experienced educators today.

The early evidence says AI can reduce teacher burnout, with surprising consistency. In a controlled trial ran by England’s Education Endowment Foundation across 68 schools and 259 science teachers, the group of teachers using ChatGPT spent 69% of the time the control group spent on lesson preparation. That’s a reduction of roughly 25 minutes every week, with no detectable difference in the quality of the materials produced based on a blind expert panel reviewing lesson resources from both groups. Crucially, teachers didn’t pocket that time. They redirected it — toward other planning, toward grading, and towards the parts of teaching that AI couldn’t touch. It is a pattern of reallocation, not just reduction.

What Teachers Are Actually Doing With AI — and Why

The most comprehensive picture of how teachers use AI in practice comes from an analysis of 13,071 multi-turn conversations and over 104,000 messages from more than 15,000 educators on a K-12 AI platform. Lesson planning dominates, but the data reveals a much broader pattern of use.

The conversation histories themselves are telling. The average educator prompt touches 1.7 categories at once — a teacher asking for a lesson plan that includes differentiation scaffolds and an embedded formative assessment, all in one message. Part of this is workflow maturing: teachers have moved past “generate me a lesson plan” toward more layered requests. But part of it is the AI itself. Researchers found that AI proactively introduced instructional elements — success criteria, differentiation strategies, feedback loops — even when teachers didn’t ask for them. Not every educator will take the suggestion, but the dynamic is shifting: AI isn’t just filling orders, it’s surfacing things teachers haven’t yet considered.

The most revealing finding, though, isn’t about what teachers ask AI to do. It’s about why they started using it at all.

In a qualitative study of 22 K-12 teachers across a large, diverse U.S. school district, the dominant driver for AI adoption wasn’t efficiency but survival. Teachers framed GenAI not as a productivity tool but as a sustainability measure. One described 80-hour work weeks and AI handling “bureaucratic stuff” so they could be present with students. Another reported that AI had decreased their stress dramatically. A third used it to rapidly adapt when taking on unfamiliar classes mid-year.

This matters because it reframes the value proposition, and demands a more honest accounting of where the profession actually stands. The conversation about AI in teaching isn’t happening against a backdrop of comfortable normalcy, where efficiency gains are welcome extras. It’s happening in a profession that is, by almost any measure, already in crisis. Teacher attrition remains elevated above pre-pandemic levels, and staffing shortages are pushing people into classrooms they were never trained for.

In that context, aspirational framings — AI unlocking personalized learning at scale, teachers evolving into learning architects, schools transforming into individualized ecosystems — overshoot the reality within classrooms. The more grounded, and more urgent, question isn’t how to get from good to great but how to get from unsustainable to functional. The evidence suggests AI can be part of the solution — modestly, in specific ways, for teachers who use it deliberately.

Where the Quality Holds and AI Wins

One of the persistent anxieties about AI in teaching is that faster means worse. The emerging evidence suggests this fear is nuanced: there are distinct areas where AI excels and where it falls short.

The EEF trial’s blind expert review suggests no significant difference in pedagogical quality between AI-assisted and traditionally prepared lesson materials. Teachers themselves largely agreed — 52% rated quality as similar, 26% said slightly higher, 22% slightly lower. And 75% reported a positive overall impact on their lesson preparation, with 78% saying they intended to continue using AI after the trial ended.

But quality isn’t uniform across lesson components. AI-generated lesson conclusions — cool-down activities, exit tickets, reflective summaries — were preferred 59.7% of the time over human-designed equivalents. This was the only lesson component where AI consistently outperformed professional curriculum designers. The structured, reflective nature of lesson wrap-ups plays to AI’s strengths: synthesizing content, generating clear summary frameworks, producing consistent quality without the variability of a human designer having an off day.

The grade-level pattern was equally striking. At the elementary level, human plans dominated — educators preferred them roughly 65% of the time, citing developmental appropriateness and engagement hooks that AI couldn’t replicate. By middle school, AI became competitive, with customized GPT-4 actually preferred 54.5% of the time. At the high school level, a fine-tuned model outperformed human designers 59.2% of the time. The more structured and content-dense the instruction, the better AI performed.

This connects to one of the more practical findings from the EEF trial: AI appears especially valuable for teachers working outside their primary expertise. The moderator analysis showed that teachers less confident in their science subject knowledge used ChatGPT more and experienced greater time savings. When a biology specialist has to teach physics, drafting a 10-question recall quiz from scratch requires 45 minutes of textbook research. AI eliminates that research friction, providing an accurate content baseline so the teacher can focus on how to teach it rather than learning what to teach. In a profession where staffing shortages routinely push teachers into unfamiliar subjects, this isn’t a convenience — it’s a structural support.

The Reallocation Effect

The evidence on time savings and quality is useful. But the most important question — does any of this actually improve student learning? — requires a different kind of study.

The cleanest answer comes from a large-scale randomized experiment across 178 public schools in Brazil, involving roughly 19,000 high school seniors. The study tested an AI writing platform that automated essay feedback. One version paired AI feedback with human graders; another let AI handle the full loop alone. A control group had neither.

The headline result: both AI groups produced identical improvements on Brazil’s national university entrance exam, independently scored. The human graders, at roughly $0.85 per essay, added zero incremental learning benefit over pure AI.

But the real story is what happened inside classrooms, not on the exam. When AI took over the mechanical grading, teachers didn’t simply work less. They reinvested the time. Students in AI-supported classrooms had roughly 35% more one-on-one conversations with their teachers about their writing. They wrote 30% more essays. Teachers in the AI-plus-human arm saw a 20% drop in at-home work hours and a dramatic shift in how they experienced their workload — the share reporting time as “very insufficient” fell from 23% to 9%. The work changed, but the workload didn’t increase.

And here’s the finding that matters most for educators: the largest learning gains weren’t on the skills AI handled well. They were on the most complex, highest-order writing task on the exam, and the one AI is least equipped to evaluate. AI can’t assess whether a student’s argument is coherent or original. But by absorbing the routine grading, it freed teachers to have exactly those conversations. The AI didn’t need to be good at evaluating arguments. It just needed to free up the person who was.

This is the reallocation effect at work. AI handles what it can — mechanical evaluation, surface-level feedback, scoring — and teachers redirect their effort toward what only they can do. The bottleneck in this system was never feedback quality, but rather teacher bandwidth.

One important caveat: the bottom quartile of students — the very lowest performers — showed no improvement under either AI condition. For the most struggling learners, freed-up teacher time alone may not be enough. This doesn’t undermine the approach, but it marks a clear boundary. AI-driven reallocation works best when students already have enough foundational skill for teacher conversations to land.

Where It Breaks Down

The evidence for AI in teaching workflows is real, but it comes with fault lines that are easy to miss if you only look at the headline numbers.

The first risk is what might be called the prompting gap. In the EEF trial, while survey data showed teachers felt confident in their prompting ability, an analysis of actual ChatGPT transcripts revealed a different picture: almost no teachers used follow-up prompts to iteratively refine AI output. They took the first result and manually edited it themselves. This is the equivalent of using a search engine for every query — one input, one output, move on. Teachers are leaving significant efficiency on the table. If the quiz is too hard, the instinct is to rewrite it by hand, not to prompt the AI to adjust the difficulty and swap in more relevant examples.

The same 13,000-conversation analysis cited earlier reinforces this pattern from a second angle— and sharpens the stakes. Across that dataset, prompt quality directly determined output quality. Educators who specified standards, student context, and pedagogical strategies got materially better results than those who wrote “generate a lesson plan for fractions.” The implication is uncomfortable: the teachers who need AI most — early career, under-resourced, teaching outside their expertise — are often the least equipped to prompt it effectively. Professional development on AI use isn’t a nice-to-have. It’s a prerequisite for the productivity gains to materialize.

The second risk — identified in that same study — is what researchers called the assessment trap. Nearly half of all educator conversations with AI involved assessment-related tasks — generating rubrics, formative assessments, student feedback. But some educators requested evaluations of student work without specifying grading criteria or rubrics. AI-generated assessments applied without human oversight risk inconsistency and bias, particularly in high-stakes contexts. AI is an excellent tool for creating a rubric. It is a dangerous tool for applying one blindly.

The third risk concerns equity, and it operates at two levels. At the student level, a controlled comparison of AI-generated and human-designed lesson plans found that AI materials were consistently rated as “neutral” regarding support for multilingual learners and students with disabilities — not actively harmful, but lacking the targeted, nuanced scaffolding that human designers build in. A 30% reduction in prep time is a net negative if it comes at the expense of vulnerable learners who need explicit accommodation. Teachers must actively direct AI to include these supports, because AI won’t generate them by default.

At the teacher level, the qualitative study of 22 K-12 teachers previously cited found that the reallocation of saved time is deeply uneven. Teachers with stronger peer networks, more professional development access, and more institutional support tend to redirect freed-up time toward higher-value instructional activities — redesigning lessons, deepening differentiation, investing in student relationships. Under-resourced teachers may simply use AI to keep pace with existing demands without upgrading their practice. The efficiency dividend risks widening the gap between well-supported and under-supported schools.

What’s Coming

Most of the evidence above is based on teachers using general-purpose chatbots through typing prompts, getting outputs, and manually editing results. That workflow is already shifting.

Agentic AI systems represent the next evolution: instead of responding to a single prompt, these systems can decompose a complex goal into sub-tasks, retrieve relevant data, execute steps autonomously, and iteratively refine their own outputs before presenting them to a user. In education, this means moving from “write me a quiz” to “design a personalized learning path for this student based on their assessment history,” with the AI handling the intermediate steps.

Early proof-of-concept work suggests the gains are real but specific. A multi-agent scoring system — where separate AI agents evaluated content, grammar, and coherence independently before a lead agent synthesized scores — outperformed standalone GPT-4o on essay grading, achieving 8.4% better accuracy and 13% greater consistency. The consistency improvement matters most: for assessment fairness, reducing variability between scores is arguably more valuable than marginally improving average accuracy.

The teacher’s role in this model shifts from prompter to orchestrator — less time crafting the perfect input, more time reviewing and approving the output of automated workflows. The 30% time reduction currently comes from delegating content generation. The next wave will come from delegating processes, and the tradeoffs are yet to be fully considered.

The Question Worth Asking

By now the time-savings question has a clear answer: yes, AI saves teachers roughly 30% of lesson preparation time, with no measurable quality loss. But what about the tougher question: can AI reduce teacher burnout?

If the time that AI saves gets absorbed by existing demands rather than redirected toward higher-value work, then any reallocation effect will only materialize in schools that intentionally design for it. For under-resourced teachers, the default outcome isn’t transformation; it’s keeping pace.

Which brings us back to where we started. Can AI reduce burnout? The honest answer is a qualified yes — not because it shrinks the job, but because it changes where a teacher’s hours go.That relief is conditional: it takes deliberate design — protected time, professional development — to turn saved minutes into the relational work that makes the profession sustainable.

Moving teachers from unsustainable to sustainable is a problem much larger than AI alone — but AI looks like it can make a dent, contingent, as always, on intention and implementation. The question worth asking isn’t whether AI can make teaching more efficient, but whether we’ll build the conditions for that efficiency to lift teaching off the floor of chronic burnout. From there, we can begin optimizing for better experience and learning, for teachers and learners alike, including the harder questions of reallocation and everything beyond.

Special thanks to Overdeck Family Foundation for sponsoring this article, the second of seven in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative). Stay tuned for more!

Top Edtech Headlines

1. Anthropic and Gates Foundation Launch $200M Partnership Focused on AI for Education and Global Impact

Anthropic and the Gates Foundation have announced a four-year, $200 million partnership aimed at expanding the use of AI across education, health, agriculture, and economic mobility initiatives. The collaboration will combine grant funding, Claude AI credits, and technical support to build tools for teachers, students, researchers, and underserved communities.

Learn more here.

2. The Seed 100 Barely Covers EdTech. Here’s the Map That Does.

Angela Chen has released a new map and essay covering the concentrated ecosystem of specialist funds, corporate investors, grants, and competitions that continue to fuel the sector.

Learn more here.

3. HireEducation 2026 Salary Report

Sponsored by HireEducation

HireEducation has released their 2026 HireEducation Salary Report: a data-driven analysis of compensation trends in the education products and services sector. This report is based on salary expectations gathered during over 1,500 interviews with candidates across the industry, offering unique insights into what it truly takes to attract and retain top performers.

Learn more here.

4. Why U.S. Test Scores Are in a ‘Generation-Long Decline’

New analysis from Stanford’s Educational Opportunity Project and The New York Times finds that while some districts are rebounding from pandemic-era learning loss, many students across the U.S. remain behind in math and reading compared to pre-pandemic levels. The findings reveal major differences in recovery between districts, underscoring how factors like attendance, staffing, and local policy decisions continue to shape academic outcomes years after school closures ended.

Learn more here.

5. Plug and Play Foundation EdTech Pitch Competition

​Have an EdTech idea? This is your chance to pitch directly to the network behind companies like Dropbox, PayPal, and Honey and get your idea investor-ready before you even step on stage.

​The Plug and Play Foundation is hosting an EdTech Pitch Competition on June 1 (1-3:30 PM PST) at Plug and Play’s Sunnyvale HQ, and they’re looking for EdTech founders building the future of education.

Learn more here.

Listen In: AI, Jobs, and What Comes Next

We recently had Kumar Garg on The Edtech Insiders Podcast!

Kumar Garg is the President of Renaissance Philanthropy, where he leads thesis-driven philanthropic funds focused on major global challenges. Previously, he worked in the Obama White House Office of Science and Technology Policy and helped build Eric Schmidt’s science and tech initiatives.

5 Things You’ll Learn in This Episode:

1. How AI is projected to impact jobs, with forecasts suggesting modest but meaningful workforce contraction

2. Why community colleges and skilled trades may see rising demand in an AI-driven economy

3. How “superforecasters” are being used to predict labor market shifts more systematically

4. What policymakers and educators should track to prepare for AI-driven disruption

5. How Renaissance Philanthropy is rethinking how capital flows into big, thesis-driven ideas

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

The Cognitive Debt Problem

Sponsored by Overdeck Family Foundation

By: Rui Lin Ler

Special thanks to Overdeck Family Foundation for sponsoring this article, the first of seven in our AI & Efficacy Editorial Research Series diving into key research findings from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative). Stay tuned for more!

The promise of AI in education is real: personalized explanations and reduced barriers for students who previously had no support. But Stanford’s SCALE Initiative recently reviewed 20 causal studies on AI in K-12 education — one of the most rigorous syntheses of the evidence to date. Their headline finding: AI tools improve student performance while students have access to them, but those gains weaken or disappear when students are assessed independently. The question of whether AI is supporting learning or quietly replacing it is no longer rhetorical. The evidence is emerging.

The Numbers Look Fine… Until They Don’t

To understand why this matters, it helps to use a framework most educators already know: Bloom’s taxonomy. Learning isn’t flat. It builds from the bottom up: remembering and understanding at the base, then applying, analyzing, evaluating, and creating at the top. Durable learning has always meant moving up the pyramid, building on incremental gains.

AI is doing something very specific to this pyramid. It’s accelerating performance at the lower tiers while leaving the upper ones largely untouched… and sometimes even making them harder for students to reach.

In one randomized study, a group of students using AI outperformed a control group that was not using AI by 22 points on immediate recall and comprehension tasks — exactly the bottom rungs of Bloom’s. That’s the kind of result that shows up in a grade book and looks like success. But three weeks later, the same gap between these groups had collapsed to 6 points and was no longer statistically significant. For higher-order tasks — synthesis, evaluation, application — the control group without AI led at every single timepoint.

Another study found that students who used ChatGPT to improve their essays showed significant score gains compared to peers who didn’t — but zero difference in knowledge gain or transfer when tested afterward. Offloading the bottom of Bloom’s pyramid did not free students to wrestle with deeper learning or perform better at the top. The essay improved. The same cannot necessarily be said about student understanding.

This is the core measurement problem facing schools right now. Most assessments capture performance at the moment of submission. They aren’t designed to catch cognitive debt, because the debt comes due later — after the unit, after the term, sometimes after graduation. When schools evaluate AI tools against grade outcomes, they’re looking at exactly the metric most likely to give a false positive.

What’s Actually Being Offloaded

One of the most striking bodies of evidence on this question doesn’t come from test scores. It comes from EEG scans.

In a four-month longitudinal study, researchers tracked brain activity across three groups of students — those writing with ChatGPT, with a search engine, or without any tools. The pattern was consistent and worsening over time. But it wasn’t just a general dimming of brain activity. The research identified specific, meaningful changes:

• Brain rhythms associated with deep memory formation (theta and alpha waves) declined across all four months — the oscillations that fire when we struggle to recall something, make a new connection, or work to integrate an idea into what we already know.

• Synchronization between executive reasoning and memory retrieval areas weakened — meaning the brain’s core architecture for durable learning was progressively less active.

• The outcome was stark. By the final session, 78% of students who had relied on ChatGPT couldn’t quote anything from essays they had written minutes earlier. In the no-AI group, that figure was 11%.

This isn’t a story about effort or motivation. It’s about what happens neurologically when the brain is relieved of certain tasks — repeatedly, over time. The brain is efficient: if a tool will do the remembering, connecting, and evaluating, it will gradually stop doing those things itself.

Process-mining research shows the behavioral counterpart. By tracking exactly which steps students took during AI-assisted writing tasks, researchers mapped where the work goes when AI is present:

• Natural Learning Loop: Write → Return to sources → Orient & plan → Evaluate own work → Revise

• The AI Loop: Write → Ask AI → Accept output → Ask AI again

The steps that disappear — re-reading, self-evaluation, planning — are precisely the ones that drive encoding and retention. What the EEG data shows at the neurological level, the process-mining data confirms at the behavioral level: the same functions, going quiet.

The Load Reduction Paradox

Reducing cognitive load isn’t inherently bad. In fact, it’s often exactly what good teaching tries to achieve.

Cognitive load is the mental effort required to process new information. Researchers distinguish between two types: extraneous load, the overhead created by poor design, unclear instructions, split attention between sources, and intrinsic load, the difficulty that comes from genuinely grappling with complex ideas. Eliminating extraneous load is unambiguously good; it’s wasted effort that gets in the way of learning. Eliminating intrinsic load is where cognitive debt accumulates — because that struggle is where learning actually happens.

Several studies have found that AI-powered explanations significantly reduce student overwhelm, improve self-efficacy, and make students more likely to engage with difficult material. That’s the real value. In one study, students using AI explanations reported significantly lower cognitive load and higher self-efficacy compared to students using traditional textbook materials, a genuine improvement in how they experienced learning!

But the same study found no meaningful improvement in actual learning outcomes. The comfort was real. The learning gain wasn’t. Reducing the struggle didn’t produce more learning, it produced more ease.

Most AI tools don’t distinguish between the two types of load. They reduce friction broadly — including the productive friction students need to form durable knowledge. The result is a tool that feels better to use and, in many cases, teaches less.

The Capability Prerequisite

There’s an equity argument that often gets made in favor of AI in schools: that it democratizes access to high-quality, personalized explanation, giving every student the kind of patient support that used to require an expensive tutor or an unusually attentive teacher. This argument is compelling but incomplete.

What the data shows is more uncomfortable. In a study of a student-facing AI chatbot deployed specifically to support struggling learners, lower self-efficacy students used the tool most intensively. But 40% of their queries were direct answer requests. More than 80% had no relationship to their enrolled coursework at all. More usage, no score improvement, shallower engagement throughout.

The pattern points to something that rarely gets said plainly: using AI well is itself a skill, and that skill requires baseline knowledge. Students with strong conceptual foundations can use AI as a multiplier — to pressure-test their thinking, explore alternative explanations, identify gaps in their understanding. Students without that foundation tend to use it as a replacement. They can’t evaluate what AI gives them, because they don’t yet know enough to know what’s missing.

This doesn’t make AI inequitable by design. But it does mean that deploying AI without first building prior knowledge structures may be widening the very learning gap it was meant to close. The students who benefit most from AI-assisted learning are, by and large, the ones who need it least.

What Good Design Actually Looks Like

None of this argues for removing AI from classrooms. It argues for being deliberate about when AI appears and what role it plays — and the research is specific enough to offer real guidance.

The most important finding is about sequence. In the four-month EEG study mentioned above, one group of students wrote without AI first, then used it later. These students didn’t just retain more — they used AI better. Their prompts were more precise, their critical integration of AI output stronger, their essays more coherent. Having first built a real knowledge structure, they could deploy AI against it productively.

The risk is not AI’s presence. It’s its persistence.

A few design principles that the research supports directly:

• Protect the First Attempt: Before AI enters, students should grapple with the task on their own — even imperfectly. This is when initial encoding happens. Temporal separation (task first, AI second) is one of the simplest and most consistently supported interventions in the literature.

• Force the Evaluation Loop: One study found that students who used a structured self-assessment rubric before accessing AI were the only group to significantly increase their own evaluative thinking during the task. A simple checklist — what do I already know, where am I stuck, what am I actually trying to figure out — before AI access keeps metacognitive processes active.

• Target the Right Kind of Load: AI explaining a concept before a student attempts a problem is different from AI completing the attempt. The former reduces extraneous load and can genuinely help. The latter reduces intrinsic load and is where debt accumulates. Good design does the first and protects the second.

• Give Teachers Visibility: Students can’t see the cognitive debt they’re accumulating — they feel more capable, not less. Teachers can’t intervene on a problem they can’t see. Student prompt logs as formative data aren’t about surveillance, they’re about giving educators the visibility to notice when AI use has drifted from support into substitution.

The Question Schools Aren’t Asking Yet

Right now, students have no signal that cognitive debt is building, and most schools have no metric that would catch it before it compounds.

Students who rely heavily on AI report higher confidence and lower perceived effort. They feel like they’re learning more. The debt is invisible precisely because supported performance feels like real performance — until the scaffold is removed and there’s nothing underneath.

The question worth asking isn’t whether to use AI in classrooms. That ship has sailed. The question is whether we’re designing for the moment of use — the essay that gets submitted, the quiz that gets aced — or for the learning that has to survive after the session ends.

Right now, most tools are optimized for the former. The research is making a clear case for the latter.

Special thanks to Overdeck Family Foundation for sponsoring this article, the first of seven in our AI & Efficacy Editorial Research Series diving into key studies from Stanford’s AI Hub for Education Research Repository (a project by Stanford’s SCALE Initiative). Stay tuned for more!

Top Edtech Headlines

1. Hackers Deface School Login Pages Following Instructure Breach Claims

Hackers have begun defacing school login pages after claiming responsibility for another breach involving Instructure, escalating concerns about the scale and visibility of recent cyberattacks targeting education systems.

Learn more here.

2. NYC Parents Demand AI Moratorium as Concerns Boil Over at School Board Meeting

More than 100 parents, students, and educators spoke out at a marathon New York City school board meeting, calling for a pause on AI use in schools amid growing concerns about transparency, safety, and oversight. The push for a moratorium reflects rising tension between rapid AI adoption and a lack of clear policies, with critics arguing that schools are moving too quickly without fully understanding the risks or impacts on students.

Learn more here.

3. Google Expands AI Training Efforts by Putting Educators at the Center

Google is expanding its AI education efforts across the Asia-Pacific region with $10 million to equip 4.7 million students, teachers and workers with AI skills. The initiative emphasizes teachers as key drivers of impact, aiming to equip them with the tools and knowledge to integrate AI thoughtfully into classrooms and scale its benefits across entire learning ecosystems.

Learn more here.

4. Education Department Prioritizes AI in Federal Grant Funding Decisions

The U.S. Department of Education is placing new emphasis on artificial intelligence in its discretionary grant programs, giving preference to proposals that expand AI literacy, support ethical use, and improve student outcomes.

Learn more here.

Kira 2.0: Real Voices from the Launch Event

Sponsored by Kira

At the Kira 2.0 launch event in New York City, Sarah Morin captured rapid-fire insights from educators, administrators, and edtech leaders reacting in real time to one of the most ambitious AI learning platforms to date. Listen in to our episode on The Edtech Insiders Podcast to hear insights from:

• Brandon Chitty, Executive Director of Innovations, Broken Arrow Public Schools

• Brett Roer, CEO & Founder, Amplify and Elevate Innovation

• Harl Roehm, CTE Teacher, Collierville High School

• Jesse Lubinsky, Principal Evangelist, Adobe Education

• Lance Key, Instructional Technology Coordinator, Putnam County Schools

• Rachelle Dene Poth, Educator, Consultant & Attorney

• Hannah Gurmankin, Special Education Teacher

• Scott Brewster, Co-Founder, Hats & Ladders

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now

The Edtech Insiders Rundown of ASU+GSV 2026

That’s a wrap! ASU+GSV is over and we’ve had the weekend to recover, reflect, and celebrate another incredible edtech community gathering. The entire Edtech Insiders team left the event with stimulating ideas, new connections, and most of all a feeling of gratitude.

Firstly, thank YOU — our Edtech Insiders community — for showing up and joining us in San Diego! It was amazing to see so many of you in person. Our Happy Hour was another amazing gathering of old and new faces. We had so many incredible conversations there, and we hope you did too. Thanks for taking the time during this busy Summit to drop by and say hello!

This Happy Hour, and everything we do at Edtech Insiders, wouldn’t be possible without the support of our amazing sponsors!

• Our Leading Event Sponsors: Learning Commons and Anthropic

• Our Supporting Event Sponsors: Magic Edtech, Lincoln Learning Solutions, Penn GSE, Kira, and EdTechnical

• Our Season Sponsors, who support everything we do at Edtech Insiders: HireEducation, Tuck Advisors, and Cooley

If you’d like to get involved as a partner or sponsor, email us at info@edtechinsiders.org.

Last but not least, we also want to say a special thank you to the ASU+GSV team for once again hosting an INCREDIBLE event. Special shout-out to our very own Alex Sarlin, who is now part of the ASU+GSV team too.

With all of this gratitude and appreciation in mind, here are our key takeaways from ASU+GSV 2026.

Team Takeaways: Notes from Alex Sarlin

Every year at ASU+GSV I try to walk the halls, listen to as many conversations as I can, and leave with a handful of themes that feel real rather than manufactured. This year the signal was louder than usual, and more contradictory. Here’s what stuck with me:

Different people are on very different parts of the hype cycle

The mood in panel rooms and the mood at the kitchen table are not the same right now, and you could feel that gap all week. Among founders, investors, and school leaders who are deep in this work, the energy was high. Among the broader public, parents, professors, K-12 teachers, and even plenty of students, something closer to a trough of disillusionment is setting in. Academic integrity anxieties, data harvesting stories, screen time fatigue, the steady creep of Big Tech into classrooms, the pull of social media and games and endless video… All of it is compounding into a pretty serious pushback on edtech in general and AI in particular.

At the same time, the companies themselves are entering what I think is a genuinely promising chapter. Enterprise-grade security, real data privacy work, stronger pedagogical design, more serious piloting before wide rollouts. The product bar is rising fast. That creates a weird kind of tension: the tools are getting better right as public patience is getting thinner. Early adopters in districts and on campuses are watching policy decisions come down from statehouses, school boards, and provost offices, quietly hoping the guardrails don’t cut through the tendons of the tools they’ve come to rely on.

The line between Big Tech and EdTech has basically dissolved

The exhibitor list this year was a who’s who. OpenAI, Anthropic, Google, Microsoft, Meta, NVIDIA, Apple, Perplexity, Lovable, Synthesia, HeyGen. Alongside them, more than 900 edtech companies, most of them AI-native or AI-infused. It’s a double-edged sword. On one hand, the frontier showing up in San Diego raises the profile of the whole industry and pulls in talent, capital, and attention. On the other, it raises the real question of whether independent edtech companies can hold their ground when the platforms themselves ship directly into classrooms and onto devices. The distinction between “AI company with an education play” and “edtech company using AI” gets thinner every year.

A welcome return to real problems and real evidence

The GSV Cup Fan Favorites were a bright spot, and a useful tell about where the field is heading. The companies breaking through were the ones solving concrete, stubborn problems with actual data behind them.

Journify Learning is tackling the special education crunch, automating the paperwork and progress tracking that drains IEP teams, and generating personalized, evidence-based materials aligned to each student’s plan. It’s already in districts across more than ten states.

Nectir is building course-aware, FERPA-compliant AI assistants that sit on top of a university’s own content. Their system wide rollout across the California Community Colleges is the largest deployment of its kind in U.S. education, and a peer-reviewed study at Los Angeles Pacific University showed a 7.5% campuswide GPA lift, one of the only published outcomes studies of its kind.

DeweyLearn is turning classroom video and audio into structured insight about engagement, conceptual mastery, and collaboration, giving teachers feedback that used to require a coach in the room.

There was plenty of agentic AI talk too, but most of it is still slideware. The products aren’t quite there yet. Give it a year.

The skills revolution finally feels real

The Force, the new workforce track at ASU+GSV, was one of the most energetic corners of the Summit. The companies in that space are doing something genuinely new: taking a company’s policies, data, and internal expertise and turning them into skill progressions, delivery models, and tools that live where people actually work. Not LMS nostalgia. Not generic training libraries. Real in-the-flow-of-work learning, finally.

The money conversation is harder than the stage suggests

For all the energy on the main stages, the sidebar conversations about capital were sober. Five years past the pandemic, the investment picture still hasn’t recovered. Generalist investors remain wary of edtech. Several edtech-focused VCs have not delivered the returns they promised their LPs. Few people expect a healthy M&A market or a credible IPO window in the next twelve months.

Add in the structural stuff, long sales cycles, usage concerns, the post-ESSER budget cliff, higher ed enrollment softness, Big Tech bearing down from above, and a cultural moment that is at best ambivalent about both tech and AI, and the financial viability question is as live as it has ever been. The Summit felt lively. The cap tables feel tight. Both things are true.

Team Takeaways: Notes from Ben Kornell

What a week! ASU+GSV is always a moment where the community comes together, and that’s what makes it so special. At the same time, the event amplifies themes that have been brewing over the past twelve months and previews themes that are yet to come. This year, it was clear that companies, institutions, and investors are all hurting with declining student outcomes, funding cuts and pullbacks, and societal headwinds. And yet there remains a persistent optimism that AI may be the answer over, through, or around these challenges.

I always approach ASU+GSV as a learning opportunity. I’m listening to new themes, ideas, or connection points that might shape where things are going and provide insight into unlocks for edtech and its impact. The three words that caught my attention this year were:

Jagged

The term “jagged” has been thrown around quite a bit with regards to the intelligence of AI systems, but it is an apt term to describe this entire era. There are places where AI superperforms; there are places where human beings superperform; there are areas where AI + human beings superperform. And yet, there remain huge crevasses where no combination is meeting the needs of learners, educators, and systems.

We live in a jagged era where the peak of what’s possible contrasts supremely with the abyss of… also what’s possible. I was consistently reminded of Ethan Mollick’s book Co-Intelligence because the love affair with AI as a silver bullet has faded and the focus has moved towards humans + agents + AI. This is much more complex and, frankly, much more interesting than early fantasies about the use cases for this technology. Understanding where the jagged peaks are and where the jagged valleys are will be critical to constructing coherent learning and teaching experiences for every single edtech platform.

Harness

Every year I host a small lunch that I call the “Interesting People Lunch.” This year’s provocation came from Lawrence Holt, who asked, “Who will build the education harness for AI?” Harness is such an interesting word to use for this prompt.

In one context, it’s how do we harness the power of something, but in this case it’s more like a horse’s harness. How do we steer technology to provide enough learning friction and enough safety to excel in the teaching and learning environment?

This also raises questions of agency: Who should define the constraints and direction of AI learning models? Is it centrally controlled? Is it controlled by educators? Is it controlled by the learners themselves? If we as a sector could align on a framework for the AI harness, we could ensure not only a floor in terms of safety and ability, but also a higher ceiling for what’s possible with educational AI tools.

Dislocation

Less of a concept and more of a deep-held fear, many are focused on the oncoming tsunami of white collar AI automation and concerned that it will catch society, institutions, and education completely unprepared. There is a growing sense of urgency around fixing AI resilience problems within the edtech industry that doesn’t seem to resonate with institutional leaders themselves. There is great appreciation that school principals, university presidents, chief HR officers, etc. have their hands full in the current state. Are they and their organizations ready for what’s to come? And how can we help them prepare?

Join the Conversation

The Summit has prompted some amazing conversation in our Edtech Insiders+ members-only group chat. We’ve been debriefing our time together, and jumping down rabbit holes instigated by industry conversations, themes, and noticings. Here’s a few notes from some of our community members:

• The power of relationships will help weather all storms. Lots of money in the system or a lack of dollars, relationships remain important above all else.
  Jacob Kantor, Chief DODO

• I was surprised that “coherence” became a buzzword at this year’s ASU+GSV. Yet, it seems everyone defines coherence differently.
  Vincent Zhang, Co-founder & CEO, Thinkverse

• EdTech is officially way ahead of their average user.
  Christopher Kelly, Director of Growth

• Let’s move students and educators from AI consumers to AI creators!
  Sari Hurme-Mehtälä, imagi CGO

• While the forward-looking picture across education is a jumble these days, this year’s ASU+GSV Summit was a unique opportunity to examine all of the puzzle pieces.
  Ian McCullough, Marketing Leader

To support our continued work at Edtech Insiders, and to join the deep dive conversations in our members only WhatsApp group, join Edtech Insiders+ as a paid member today.

Thanks for another amazing year together, and see you at ASU+GSV 2027!

Subscribe now

Top Edtech Headlines

1. Blanket EdTech Bans Miss the Mark: Better Policies for Student Tech Use

A new analysis from New America warns that sweeping efforts to restrict classroom technology may backfire, limiting access, widening inequities, and overlooking how digital tools can support learning when used effectively. Drawing on student and expert perspectives, the piece argues for more nuanced, evidence-based policies that prioritize intentional use, digital literacy, and stronger implementation over one-size-fits-all bans.

Learn more here.

2. HireEducation Releases New Salary Report

HireEducation’s latest Salary Report sheds light on compensation trends across the education sector.

Learn more here.

3. Introducing Learn Mode: Your Personal Coding Tutor in Google Colab

Google has introduced a new wave of updates to Colab, transforming the platform into a more AI-native coding environment with agentic assistance, real-time code generation, and integrated data analysis tools. The changes aim to make coding more accessible and collaborative, lowering barriers for students and developers while accelerating workflows across data science and machine learning tasks.

Learn more here.

4. ‘Vibe Coding’ Helped a Washington District Save $250K in EdTech Costs

A Washington school district has saved over $250,000 in edtech spending by adopting “vibe coding”, allowing staff to quickly build and customize internal tools instead of purchasing software.

Learn more here.

Why Graduation Is Higher Ed’s Most Overlooked Growth Lever with Chase Rigby of Tassel

Sponsored by Tassel

Chase Rigby is the CEO of Tassel, a platform modernizing graduation experiences for universities worldwide. A former Google PM and Chief Product Officer at Canopy, his career sits at the intersection of education, technology, and product innovation.

5 Things You’ll Learn in This Episode:

1. Why graduation is a key (and overlooked) growth moment

2. How Tassel simplifies complex commencement logistics

3. Where AI improves the student and admin experience

4. How graduation drives alumni engagement and giving

5. Why Tassel helps students earn money at graduation

Listen Here

Thanks for reading! Support our work by becoming a paid subscriber, or email info@edtechinsiders.org to learn more about partnerships and sponsorships.

Subscribe now