In this issue, I’m reading the curve to examine:

• Where physical AI is poised to take hold

• Why data will shape how quickly physical AI advances

• What the next phase of human-machine collaboration looks like

Reading the Curve

I’ve always enjoyed finding ways to get a glimpse of the future. A few years ago while visiting Los Angeles, I felt the future when getting my first ride from an autonomous ride hailing service and seeing an expanding fleet of self driving food delivery vehicles. At first, these experiences were novel enough that I made a point of seeking them out.

Today, however, these once-futuristic experiences are easier to find. Autonomous vehicles and delivery robots are moving into more cities and everyday settings, bringing physical AI out of controlled environments and into the world around us.

In fact, Morgan Stanley Research forecasts that the market for robotaxis could reach $1 trillion by 2040. That’s just one application. Robots are working alongside people on factory floors, moving through hospital hallways and warehouses, and increasingly venturing into less predictable environments like our sidewalks.

As their capabilities and applications have expanded, so has our idea of what constitutes a robot. My colleague Jim Wilson, Accenture’s global managing director of IT research, has been following this evolution for years. In a 2015 Harvard Business Review article, he noted that robots could already take radically different forms, from industrial arms to autonomous vacuums to humanoids. Rather than defining them by how they look, Jim offered a broader definition based on their relationship with us: “an artificially created system designed, built and implemented to perform tasks or services for people.”

AI is accelerating this evolution. Our research shows that interest in physical AI is rising, as measured by activity in academic papers. This trend suggests a coming wave of AI that moves intelligence beyond screens and into the physical world around us.

In fact, companies are already changing how they plan for the future. My colleague Max Blanchet, who leads Accenture’s global supply chain and operations strategy practice, wrote in December that nearly all operational activities for manufacturers will be either AI-augmented with human oversight or fully autonomous by 2030. Today, 63% of these companies have integrated autonomy into their strategic plans and 25% are actively implementing it.

Subscribe now

Investors appear to be betting on this future, too. Benjamin Lawrence, who leads CB Insights’ emerging tech and industrial research teams, said their research shows that physical AI has become the most active area of the AI market, by deal volume.

“Essentially we have a massive amount of funding from investors going to physical technologies because the application layer, the software layer, got a lot of funding in the last few years and you don’t have as much first mover advantage,” Benjamin said. This includes investments in companies that make AI-powered robotics, drones and autonomous vehicles, as well as the software systems that control them, the model layer and the adjacent supporting hardware.

The value proposition of physical AI solutions is anchored in a combination of multiple factors including productivity enhancement, operational efficiency and/or increased safety.

To help me understand what comes next and what it will take to get us there, I spoke with Benjamin and my colleagues at Accenture Research, including Jim, and emerging tech researchers Prashant Shukla and Shruti Shalini.

They told me how technology is getting smarter, but also that making tech useful in the physical world means solving for everything from data scarcity and safety to how people and machines work together.

Intelligence takes on a physical form

Up until this latest issue of my newsletter, we’ve mostly addressed how intelligence helps us think, decide and act in the digital universe. Physical AI takes that intelligence out into the world around us.

“We have all the intelligence in the world now. We have AI systems that can improvise and learn,” said Prashant, referring to the most advanced LLMs that can perform intelligent feats in seconds that no human could match. “What if we put this level of intelligence in things? All of a sudden you’ve got an embodiment of a very intelligent being.”

But putting intelligence into things introduces a new challenge, because those things have to work in the environments we’ve built for people. Intelligence alone isn’t enough.

The physical shape and fit of a technology must match the environment it’s meant to serve. The robots that fold laundry or vacuum our floors need to look different from the bots making cars and assembling equipment on the factory floor.

The environment itself matters, too. Robots have worked in factories for decades, where tasks and surroundings can be tightly controlled. What’s changing now is their ability to move beyond those settings into places like roads and sidewalks, where the world is far less predictable.

Autonomous vehicles navigating around San Francisco, Los Angeles and various other cities in the south of the US may be the strongest proof of concept for physical AI systems leaving the factory floor for the mean streets.

Operating in the physical world raises the stakes considerably. A mistake made by digital AI can often be corrected, but a machine acting in the physical world can cause physical harm. As Prashant put it, “the calculus of risk is much higher.” In order to safely operate, these systems must learn how the physical world behaves.

That brings us to one of the biggest constraints on physical AI today: the data to teach them.

Data data everywhere, but not the kind we need

Our digital lives generate an enormous amount of information about what we search, watch, buy, say and do online. That abundance helped fuel the rise of large language models. But physical AI needs a different kind of data.

“The difference is you need to train these systems on atoms and objects,” said Jim.

And unlike the text, images and video already available online, much of that data still has to be created or captured.

Benjamin of CB Insights described three ways that companies are going about it. One is synthetic data, which can be generated through simulation, generative models or other methods to generate experiences a physical system can learn from. “But it’s not a replacement. It’s a complement. You can use that to help fill gaps, but you can’t rely on it.”

Another is human demonstration data. Workers can wear sensor-equipped gloves or other devices while performing physical tasks over and over, capturing the movements a robot eventually needs to learn. This is labor-intensive, making it difficult to generate the sheer volume of real-world data that these systems require.

Then there’s the data the physical world is already producing. Cameras in warehouses can capture how people and goods move through a facility, while sensors can record existing machines and systems as they operate. Benjamin pointed to autonomous driving as perhaps the clearest example, relying on human drivers navigating real roads, including the hiccups every driver has experienced: “The squirrel running in the road, the human walking across the street or the car that turned when it shouldn’t,” Benjamin said. “And you do that over millions and millions of miles and eventually you get a system that is actually capable. That’s where the training data comes into play.”

That last example gets at the scale of the challenge. Physical AI doesn’t just need to learn how to perform a task. It needs enough experience of the physical world to know what to do when reality doesn’t go according to plan. Which helps explain why some environments are much closer to physical AI readiness than others.

A question of readiness

So where is physical AI showing up first? The answer has a lot to do with how predictable the environment is. A factory floor or warehouse, where movements and processes can be controlled, presents a very different challenge from a construction site or city sidewalk.

“Most of the early stage robot systems were in controlled environments like factories: single task robots, flexible robots, collaborative robots,” said Jim. “Where the robotic systems leave the factory and go into less controlled environments—like sidewalks or drone deliveries—they’re doing things that are industrial or consumer applications.”

Collaborative robots have been working alongside people for years and can often be reprogrammed by the people working with them. Jim points to auto manufacturers as examples of companies that have used collaborative robots for at least a decade.

But as physical AI moves into more complex environments, working alongside people becomes a bigger part of the equation. Opportunities range from retail and e-commerce fulfillment to hospitals, where robots can take on routine activities such as retrieving medical supplies, freeing nurses to spend more time with patients.

The closer these systems get to people, the more trust matters. In a hospital or home, for example, physical AI may encounter sensitive information while collecting the real-world data it needs to operate. And, as noted, unlike with digital AI, mistakes can have physical consequences.

Still, there’s a longer view worth keeping in mind. People have been adapting how they work alongside machines for generations. Physical AI is another step in that progression, but one with an important new capability: “A robot in a factory might learn a new way to grasp something in a way that is transferable to other contexts like delivery or even surgical robotics,” Jim said. “You can actually have robotic systems learning from each other.” (Jim and his Human + Machine co-author Paul Daugherty wrote about this recently for Harvard Business Review, detailing how machines learn through trial and error and human feedback.)

Our own ability to learn alongside machines is changing, too. Simulations now allow managers to test many different approaches before putting a physical system into production.

“Now management can see what works best and maximize and innovate in ways they haven’t before. Because you can run simulations or 50 different simulations and choose the ones that you like the most,” Jim said. “Before, when you were putting in a new assembly line in your new plant, you had to get it right at the start.”

In that context, my colleagues Tracey Countryman, Inge Oosterhuis, Jeff Wheless and Rushda Afzal recently published an article (referenced below) in which they explain how “the best manufacturers build AI with workers, not for them”. Leaders who include people in redesigning roles, train them in the flow of work and make it clear who is accountable for outcomes when work includes AI will not merely help their workforces keep pace with the learning curve. They will help them to define it.

Worth Your Attention

• From MIT Technology Review, Prashant offered this piece: Why humanoid robots need their own safety rules. It’s a look at the unique risks these robots introduce and the standards needed before they can work closely alongside people.

• For Harvard Business Review, my colleagues Tracey Countryman, Inge Oosterhuis, Jeff Wheless and Rushda Afzal wrote this guide to building trust as AI changes the factory floor: The best manufacturers build AI with workers, not for them.

• Also from HBR, colleagues Adam Burden, Christian Souche, Shruti Shalini and Alpana Dubey wrote: 5 questions to ask before investing in humanoid robots.

• Read more about the shift toward autonomous operations and how manufacturers can prepare for it in Max Blanchet’s article for Ivey Business Journal.

• For a deeper look into emerging technologies, explore the Accenture Research Journal, where you can access our full reports on everything from physical AI to cyber resilience.

Overheard

— Dr. Fei-Fei Li, AI pioneer and developer of ImageNet, on what world models can learn that language models cannot

The insights above are made possible by 350 researchers, editors and AI agents across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

In this issue, I’m reading the curve to examine:

• Why confidence matters when it comes to who gets ahead with AI

• How co-learning turns AI use into a two-way skill

• What schools can teach every organization about readiness

Reading the Curve

I’ve been teaching graduate-level business courses for over two decades now, and I’ve seen how each generation of students arrives ready to make sense of whatever technological shift defines their moment. Back then, we were exploring the possibilities of the internet. Today we’re doing the same with AI.

The technology and tools might change, but what’s constant is how eager people are to make the most of them. I saw this with my most recent class, where I ran a small experiment with students—mostly with international backgrounds—at Sciences Po, a French university where I teach corporate strategy. For the midterm exam, I allowed AI resources in line with the school’s guidelines and on the condition that students explain how they used them. For the final, they were on their own.

Most students naturally gravitated toward LLMs, which isn’t surprising since it’s something that’s already part of their daily life. What struck me in the results was there was no consistent gap in overall grades between the two exams. But there were differences: The AI-assisted essays tended to be better structured but more generic, while the final exams had richer, more idiosyncratic content, even if some students ran out of time to finish.

An interesting observation was how the rankings shifted. Except for a small group of consistently strong students, the order changed completely between the two exams. It’s as if the technology reshuffled the deck, and those who used it particularly well gained a real advantage, all else being equal.

AI didn’t make the class better or worse. It changed who came out ahead, rewarding those who used it well rather than lifting everyone equally.

That advantage won’t disappear after graduation. Employers are already signaling that AI fluency will shape what entry-level jobs look like. In a recent survey of C-suite leaders across the world’s largest organizations, we found that 74% say their companies are likely, or very likely, to create new entry-level roles centered on working with AI over the next one to two years.

Now, my classroom story is just one tiny example, but it points to something larger, which is why this issue looks at a new report from my colleagues Mike Moore and Babak Moussavi, alongside the British nonprofit Teach First, on how school education leaders in England are approaching AI. Their central finding is that confidence, not technology, is the real constraint. Schools have the tools. What they’re missing is the shared understanding and training to use them well.

Subscribe now

I also spoke with two researchers behind Learning, reinvented: Accelerating human–AI collaboration, Apurva Soumya and Deeksha Khare Patnaik. Between the classroom and the workplace, which this report examines, the same question kept coming up: how do we get the most out of this technology without losing the human judgment that makes us good at what we do?

This topic is one I care about deeply. I started teaching because I wanted the chance to pay back what my French public education gave me. Some of my former students are now colleagues, which is the quiet reward that keeps me at it after all these years.

What I hope they carry forward, more than any particular skill, is a sense that using AI responsibly is a personal and social choice. With the advancement of the internet, my generation opened up access to knowledge. We didn’t always get responsible use right. This is a real chance to do better, and for the next generation of leaders and creators to take these tools further than we ever imagined.

Being graded along the AI curve

The reshuffling that I saw in my classroom, where the rankings shifted depending on how well students used AI, begs the question: What determines who ends up ahead? It also underscores the importance of AI readiness for our future workforce.

The two reports I’ll reference here, AI in schools in England and Learning, reinvented, suggest that access to AI alone doesn’t determine the outcome. What matters just as much is how people engage with the technology. At the institutional level, it’s leadership modeling and confidence-building; at the individual level, it’s how actively and critically someone engages with the tool, rather than deferring to it.

These examinations matter because no matter how you may feel about AI, students and employees are going to use it whether or not they get the proper training. One English school’s own internal survey found that more than 80% of its sixth-formers (this would be 11th and 12th grade in the US) were already using AI, regardless of whether the school had a policy for it. So banning the technology outright doesn’t stop that. It just means the use goes ungoverned—what’s known as shadow AI—where some students will figure out how to use it well and others won’t.

So let’s start where it often begins, with the teachers and leaders whose own comfort with AI ends up shaping everyone else’s.

Confidence in two parts

Babak helped me break down what confidence with AI looks like on the ground, among teachers and school leaders. “One is being unsure about how to use it and also the lack of trust in the outputs,” he told me, as some teachers question if what’s coming out of AI is trustworthy.

In other words, confidence isn’t a single switch. Someone can feel comfortable typing a prompt and still not trust a word of what comes back. Someone else might trust the tool completely, yet not know how to get good results from it. Either gap is enough to stall adoption.

That split shows up clearly in the interviews behind the report. School leaders fell into roughly three groups: eager adopters, who trusted both their own ability and the technology’s output; a cautious middle group, broadly optimistic but aware of the risks, and still working out what to think; and skeptics, who doubted both. Babak described the eager adopters as “confident at trying new things, but also confident that what they were getting back was useful, helpful, efficient.” The skeptics, by contrast, “did not trust the output” and “often admitted that they weren’t sure how to get the best out of it either.”

Sixty-three percent of school leaders cite a lack of staff confidence or skills as the top barrier to AI adoption in their schools, far outweighing the 16% who point to cost. The tools are there. What’s missing is the shared understanding of how to use them.

The report mentions several ways teachers can learn AI skills and support one another that don’t require a large budget or formal curriculum. Some schools have found success naming “AI champions” within departments (teachers who experiment early and share what they learn with colleagues). Others have built small peer groups where staff review AI use together.

On the more formal end, one multi-academy network surveyed its staff to understand where confidence stood, then ran an all-school training day exploring the “good, bad and ugly” of AI.

Co-learning and taking AI “off days”

Business shows a version of the same pattern, from a different angle. Eighty-four percent of executives expect AI-human collaboration to be standard within three years, but only 26% of employees say they’ve actually been trained to do it.

That’s a readiness gap rather than a direct confidence measure, but it’s hard not to see the connection. People rarely feel confident doing something they were never taught to do in the first place.

That’s where co-learning comes in.

Co-learning, as Apurva described it to me, is less like training and more like a working relationship. “The AI teaches the human, the human teaches the AI, and they both learn from each other in the flow of work,” he said. It’s not a course you complete once. It’s a continuous adjustment between a person and the technology they’re using, each one shaping the other over time.

Deeksha broke the idea down into three parts:

• The first is co-creation. “The employees are actively shaping the work,” she explained. “They write the prompt, challenge the output.” Rather than accepting whatever the AI produces, the person stays in the loop, questioning and refining it.

• The second is holding on to enough independent skill to catch AI when it’s wrong. As Deeksha put it, this means an employee is “being able to explain why a decision was made, defend it and spot where AI went wrong.” That’s a different bar than simply knowing how to use the tool. It requires knowing enough, independent of the tool, to recognize when it’s failed.

• The third, and maybe the least intuitive, is deliberate reinforcement: intentionally exercising judgment so it doesn’t quietly atrophy. Some companies build this in directly. Deeksha mentioned organizations that hold regular “AI off days,” where employees complete their usual work without any AI assistance at all. “That’s deliberately taking the brain to the gym,” she told me, borrowing a phrase from one of our colleagues, Accenture Chief Learning and Research Officer Majd Sakr. It’s a small, almost old-fashioned practice, but it’s a sign that a company takes this seriously. “This is where long-term competitive advantage comes from,” she said.

Taken together, these three habits describe something narrower and more demanding than simply “using AI.” They describe staying as an active participant in your own thinking, even while a very capable tool is standing by, ready to think for you.

A signal worth watching

There’s a reason a study of English secondary schools is useful well beyond the classroom. Schools are at the forefront of how the next generation is being most immediately impacted by AI, in terms of how students and teachers are using the technology.

If AI can reshape how one institution operates, it’s a meaningful signal for the rest of us. “Schools provide a bit of a test bed,” Mike told me, “in understanding how small organizations can use the technology more effectively.”

Worth Your Attention

• AI-ready graduates don’t emerge by chance, research from Pearson and AWS: This is a global study finding that AI readiness among graduates doesn’t fail for lack of trying, it falls apart at the handoff between what schools teach and what employers need.

• The smartest kids in the world, by Amanda Ripley: An American journalist looks at what countries like South Korea, Finland and Poland are doing differently to consistently outperform on global education benchmarks. Babak told me the book offers an interesting way to see just how different education systems are around the world, with different norms and priorities. “It makes me think that the way AI will be handled could differ quite significantly based on things like prevailing educational structures and the sense of urgency about reform,” he said.

• Exam nation, by Sammy Wright: Babak also recommends this book from a UK headmaster. He said, “It’s about rethinking the education system in the UK, but it’s also applicable elsewhere because it starts with the fundamental question: What is school for?”

• Go deeper on how humans and AI are reshaping the way we work. The Accenture Research Journal lets you query our full body of research and get answers grounded in empirical findings.

Overheard

— Amy Coleman, Chief People Officer, Microsoft, on building an AI-first, human-centered culture

The insights above are made possible by 350 researchers, editors and AI agents across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

In this issue, I’m reading the curve to examine:

• Where the business value is moving in the AI stack

• What’s closing the trust gap

• The human skill that AI still can’t replace

Reading the Curve

As researchers, we live for the moments when a signal bubbles up out of the noise. For the work we do, this could be when a technology turns out to be quietly reshaping how an entire discipline works.

That was at the heart of why we built our emerging tech rankings. We wanted a holistic view of where the AI landscape was shifting, but also the ability to examine what that view might be leaving out: the gaps between what one community is building and what another is studying or buzzing about, at the very same moment.

We introduced the inaugural rankings several months ago, based on research we conducted in December. The concept was to track core technology domains across three places where we know these signals are beginning to surface, but for different reasons: ArXiv, where early research and innovation get discussed and captured; GitHub, where code is built and deployed; and Hugging Face, where models and tools enter practice.

Our first ranking was our baseline. Large language model (LLM) frameworks and model engineering sat at the very top, dominating the stack from the start.

We also saw that classic computing was being rewired for AI, where half the technologies on the list were decades-old disciplines being reinvented for a new era. Databases, compilers and testing frameworks—the machinery underneath every system—were showing up as infrastructure being redesigned for how AI runs.

We saw that the ranking read less like a leaderboard than an architecture, with capabilities at the top, connecting systems in the middle and infrastructure underneath.

And we noticed a trust gap. Technologies that build capability consistently outranked those that make that capability reliable, accurate and explainable. This left us wondering whether we were building these systems faster than we were learning to verify them.

Four months on, some of these threads are starting to take a direction.

The first is what my colleague Jim Wilson, whom I asked to join me on this post, calls the “last mile.” LLM frameworks holding the top spot across every platform signals a kind of standardization. This means that the real differentiation for enterprises has moved downstream, to the work of tailoring a model to a company’s own data and processes.

Subscribe now

The second direction relates to a question we asked back in May: would testing, evaluation and governance start to climb as companies pushed toward production, or would that gap just keep widening? Four months on, we have a clearer read and it’s encouraging.

And the third direction connects to what these developments mean for the people doing the work. As AI tools take on more tasks, the judgment that matters most may be in the kind of tacit knowledge about a business and its culture—risk tolerance, for example—that no machine can supply.

So if the baseline rankings were about mapping where things stood, this next edition is about watching that map begin to move.

To help us make sense of it, I sat down again with Jim, who is Accenture’s global managing director of IT research, along with Shruti Shalini and Rhine Andotra, the researchers whose work built the rankings. I’ll let them take it from here.

AI is moving from pilot to product

Jim here with my Accenture Research colleagues Rhine and Shruti. When we first proposed this ranking, the idea was to build something almost like a real-time monitor from the front lines of tech: a monthly, or even weekly, pulse check on what’s rising and falling across the platforms where technology actually gets built. The thinking was, if a technology jumps to the top of the ranking one month and drops the next, that movement itself is a story and we could talk about why in near-real time.

Two rounds in, it turns out the more interesting story isn’t the week-to-week churn. It’s the slower pattern taking shape underneath all the headlines and hype. We’re starting to see how these technologies shift and recalibrate against each other and we can begin to paint a picture of where organizations are placing their bets and where their real needs lie.

Which brings us to our first topline takeaway. The obvious observation is that LLMs remained at the top of the leaderboard. But the more interesting story, I’d argue, is the movement just below it. That’s where companies stand to gain real advantage.

Increasing focus on the “last mile”

By last mile, I mean something specific: AI is becoming an actual tool you use, not just a system you’re testing. And there are a couple of strong signals pointing to that.

The first is retrieval and vector databases (#10), which jumped from #14, the biggest single move in the entire rankings. Think of retrieval and vector databases as the working memory and search of an AI system. If you’re going to run one of these systems on your own data, you need a way to search that data and you need somewhere for the system to hold onto what it’s learned. That’s what this technology does. It’s a strong signal that agentic systems aren’t just being piloted anymore; they’re also being integrated into how companies run.

The second signal is autonomous testing and verification tools (#9), which also climbed several spots, from #12. If you’re going to have these systems running inside your company, you need quality assurance around them, the same as you would for any product you ship. These testing tools have existed for a long time as programming tools. But the fact that they’re two of the biggest risers in this ranking tells me that companies are moving agents into real practice, not just running them in a sandbox.

We actually saw a version of this play out in our own operations. In one pilot, we found that if you want agentic AI to succeed, it needs a way to learn. Otherwise, it just creates more work than it saves. Getting that right also took investment in retrieval and evaluation together.

There’s a companion story here in what dropped: compiler and runtime innovation (#11), which fell sharply, from #4. My read is that this isn’t a sign that compilers matter less. It’s that we’re moving past the phase where most of the work was building the agentic system itself, largely a compiler’s job, and into the phase where the work is running it well in the real world. That’s exactly what retrieval and autonomous testing are for.

Put together, this is what I mean by the last mile. The great value is in the tailoring: fine-tuning (#2) models to your company’s own data, optimizing inference (#3) for your specific workflows and evaluating (#5) whether the system is performing by the standards your company cares about.

For leaders, this suggests the real question isn’t just about which model you’re using, but how much work you’re putting into making it yours.

The trust gap is starting to close

Back in May, we asked whether the technologies focused on testing, evaluation and governance would begin to climb as companies pushed AI systems toward production or whether that gap would widen. Four months later, we are starting to see an answer.

Evaluation, benchmarking and AI quality (#5) is now the number one topic on ArXiv, displacing compilers as academia’s leading focus. It also climbed from 12 to 5 on GitHub. Indeed, it’s one of the few categories in this dataset, along with Autonomous testing and verification (#9) and Retrieval and vector databases (#10), that moved upward on every platform simultaneously. When an upward movement shows up across all platforms at the same time, it’s harder to dismiss it as a passing trend. In this case, it signals a shift in focus toward the engineering layer that supports and builds agentic systems and LLM frameworks.

That cross-platform alignment is also worth understanding. As Rhine observed, there’s a pattern to how these two communities relate to each other. GitHub, where developers build things, tends to run ahead of ArXiv, where academia writes about those things. As developers build and run into roadblocks, academia picks up on those issues. That’s our read on what could be happening here. As agentic systems moved from prototype into production, reliability became the roadblock that everyone hit and academia converged on it. What’s new this time is that developer tooling is now visibly catching up too. The gap between what we build and what we can verify hasn’t closed completely, but it’s closing from both directions at once.

For leaders, the implication is that evaluation and testing are not disciplines that you bolt on after a system is already live; the organizations that are furthest along in putting agents into production are, not coincidentally, the same ones that are now investing seriously in the means to verify them. If you’re moving fast on capability, this is the moment to be moving just as fast on the means to trust what you’ve built. (We’ve written more on what that looks like in our responsible AI edition.)

The skill gap AI can't close

Last, something that isn’t quite a formal reading of the rankings, but it’s nevertheless a perspective that I find genuinely interesting.

With AI code generation (#6) moving up a couple of spots, someone who just got their computer science degree might be wondering what their workflow looks like, now that Claude Code can do a lot of what they trained for. My honest answer: you’re going to spend more time learning how your organization creates value and then doing the fine-tuning, inference optimization and evaluation work we talked about above.

Companies have traditionally hired early-career programmers for certain skills, like fluency in Python or Java, but AI systems are now automating many of those workflows, thus shifting the bundle of tasks that humans perform. Where AI still has progress to make is in the tacit dimension—understanding what matters to your customers, what resonates with your business leaders, how your company actually competes. That comes from being embedded in an organization.

Take autonomous testing, one of the biggest risers in this ranking. Whether a company leans on it lightly or heavily isn’t really a technical decision; it’s a reflection of that company’s appetite for innovation and tolerance for risk. The person running that testing needs a real feel for where their own company sits on that spectrum. That’s tacit knowledge and no system supplies it on its own.

As we continue to follow how these technologies rise and fall, we’ll be watching this thread too, not just where the tools are headed, but what they’re asking of the people using them.

Worth Your Attention

• From AGI to ASI: A worthwhile read from a team that includes Shane Legg, a Google DeepMind co-founder, on what might come after artificial general intelligence.

• Deep Utopia: Nick Bostrom's earlier book, Superintelligence, asked what happens if AI goes wrong. This one flips the question: what happens if it goes right?

• Crashing Waves vs. Rising Tides: Backed by thousands of worker evaluations, this MIT paper argues that AI progress looks less like sudden breakthroughs and more like a rising tide, gradual, broad and already well underway.

• For a deeper look at what these emerging technologies mean for business, explore the Accenture Research Journal, where you can access the full scope of Accenture Research reports, all grounded in empirical findings.

Overheard

— How AI Will Reshape Jobs Instead of Replacing Them, from Neil Thompson speaking at TEDxBoston

The insights above are made possible by 350 researchers, editors and AI agents across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

In this issue, I’m reading the curve to examine:

• How AI is shifting the competitive advantage in cyber toward human judgment

• The difference between organizations that recover fast and those that don’t

• Why the cyber workforce crisis is a skills issue, not just headcount

Reading the Curve

For nearly a decade, a central thread in our research has been the relationship between humans and machines. Time and time again, we’ve seen the organizations that pull ahead are the ones that figure out how humans and machines make each other better.

In 2018, when our colleagues Paul Daugherty, Accenture’s former technology officer, and Jim Wilson, Accenture Research’s global managing director of IT research, published Human + Machine, many people worried that AI would replace human workers. We presented a different view. The real opportunity lay in reimagining how humans and machines could collaborate. Each would do what the other couldn’t. That argument has been central to our research ever since. (One of Jim’s latest projects is an emerging technologies ranking that we explored in a previous Intelligence Curve issue. It tracks how ideas move from research to real-world use before they surface in the market. Look out for the latest iteration of the rankings in our next edition.)

Nowhere is this more visible right now than in cybersecurity.

The organizations that recover fastest identify who will make critical decisions and how they will respond. Machines can detect and respond at machine speed in the immediate aftermath of a cyber event. Sustained recovery, however, depends on human judgment.

Subscribe now

That is what my colleague Yusof Seedat, Accenture’s global research lead for cybersecurity and co-author of Reinventing the cyber workforce, discovered after examining more than 550,000 job postings and professional profiles over the past year. What he and our colleagues found challenges the way most organizations think about this problem.

Preparation looks different from what most organizations assume.

Better tools detect the breach. Better teams determine how quickly the business recovers.

Almost half of all cybersecurity roles remain unfilled today. That number has become so familiar that most organizations treat it as a given, a hiring problem to be managed rather than a structural problem to be solved.

Yusof’s research explains why. The problem isn’t a shortage of people. It’s that the people available aren’t equipped for what the cybersecurity role has become.

“It’s not about talent being short,” he said. “It’s about what type of talent companies actually require. It’s the right mix of skills.”

Today’s cyber workforce is largely made up of what the research calls “Operators,” professionals who execute technical security tasks, detect threats, manage controls and run tools. What organizations increasingly need are “Conductors,” people who can do that and more: communicate risk to a board, understand what a breach means for the customer and connect a security decision to a business outcome. The demand for that profile is growing, but the supply isn’t keeping pace.

The numbers bear this out. Fifty-nine percent of open cyber roles now require that blend of technical and strategic skills. Only 40% of professionals currently sit in roles that match.

The pipeline feeding the workforce wasn’t built for this. Professionals entering the field haven’t yet developed the hybrid skills employers are asking for. You could hire more people tomorrow and still not close this gap.

The operating model is outdated

The skills gap is hard enough to close on its own. What makes it harder is what happens after a major incident.

The default playbook is familiar: restructure the security team, expand the budget, bring in new tools, appoint new leaders. These moves stabilize operations and satisfy the board in the short term. What they rarely do is address the workforce model that left the organization exposed in the first place.

“After a major cyber incident, what companies tend to do is they’ll go and look at bringing in new security tools to make it stronger,” Yusof told me. “But one thing they don’t do, and we haven’t seen, is that talent realignment.”

That pattern has a compounding effect. The same skills gap that made the organization vulnerable before the incident remains intact after it. The next disruption arrives and the cycle repeats.

Part of what sustains that cycle is how organizations assign accountability for resilience, or fail to. According to our research, fewer than half of organizations treat cyber resilience as a shared enterprise outcome. Responsibility stays inside the security function, even as the risk has moved well beyond it.

Cyber crises are decision crises

That is what our research ultimately reveals. The most critical moments in the face of disruption come down to decisions: Should we keep operating in degraded mode or shut down entirely? Which systems come back first? Who has the authority to delay customer commitments, notify regulators, or accept operational risk while systems are still recovering?

These aren’t questions a security tool can answer. They require judgment, business context and trust across functions and they have to be made quickly, often before the full picture is clear.

This is where the workforce gap becomes most visible. Operators can execute the technical tasks. Conductors can do something more valuable in that moment: help leaders interpret what is happening, understand the trade-offs and coordinate decisions across technology, operations, finance, legal and communications. That capacity to translate a technical crisis into a business decision is what separates organizations that recover quickly from those that don’t.

AI is quietly shifting where that capacity matters most. As it absorbs more of the repetitive analysis and triage work—correlating alerts, detecting anomalies, flagging vulnerabilities—human value moves toward the decisions that machines can’t make.

Yusof is direct about what this means across the organization. When a crisis hits, every function faces decisions it may not be prepared to make. The marketing team needs to know what the company can and cannot say to customers and when. The finance team needs to understand which systems must come back online before the business can transact. The board needs enough context to ask the right questions. That preparation can’t happen in the middle of a crisis; it has to be built long before one arrives.

What a better model looks like

Building resilience requires a different kind of organization.

Yusof describes it as cross-pollination: rotational programs that move people from business functions into cyber roles and back, career pathways that reward strategic thinking alongside technical depth and a recognition that cyber fluency needs to be distributed across the enterprise rather than concentrated in one function.

Some organizations are already moving in this direction, building structured internal pipelines and reskilling programs rather than relying solely on external hiring.

As AI assumes more of the technical execution, the competitive advantage shifts toward people who can align technology, business priorities and leadership decisions under pressure.

“The cyber workforce was designed for protecting technology. The next generation must be designed for protecting business continuity,” said Yusof.

The organizations that fully recover won’t necessarily employ the most cyber specialists. They’ll employ the right mix of Operators and Conductors and they’ll have made sure the rest of the business knows its role too.

Worth Your Attention

• Human + Machine: Reimagining Work in the Age of AI—Paul Daugherty and Jim Wilson reframe the automation debate in this updated version of their 2018 book. Rather than asking which jobs machines would replace, they asked something more useful: How do humans and machines make each other better? Nearly a decade later, the question feels more relevant than ever.

• Degree Requirements Are Poisoning Your Cybersecurity Talent Pool—Forrester analysts Steve Turner and Jeff Pollard make a pointed case that the cyber skills shortage is partly self-inflicted. By requiring college degrees for roles where practical experience matters more, organizations are quietly narrowing their own talent pools.

• World Economic Forum’s Global Cybersecurity Outlook 2026—This year’s report examines how AI is reshaping both sides of the fight—strengthening defense while enabling more sophisticated attacks—and what that means for organizations trying to keep pace.

• Want to learn more about cyber resilience? Head to Accenture Research Journal, where you can query the full body of Accenture Research reports and get answers grounded in empirical findings.

Overheard

— Ann Johnson, Corporate Vice President and Deputy CISO at Microsoft

The insights above are made possible by 350 researchers, editors and AI agents across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

In this issue, I’m reading the curve to examine:

• Why the loyalty that brands built may not survive the agent layer

• How the relationships between consumers and payment providers is changing

• What it will take for brands and banks to remain relevant in an agent-driven market

Reading the Curve

In a recent edition, we introduced the idea of agentic commerce and what it means when AI agents acting on behalf of consumers find, evaluate and choose brands. We explored how the agent layer is reshaping discovery and brand visibility and why connecting with a machine requires a form of trust that machines and humans earn in different ways.

This edition goes deeper into what happens at the moment of purchase and what it means for the loyalty businesses have spent years building.

A simple way to think about what’s changing is to consider how you pay for something today. When you shop online, you probably already know how you’re going to settle up. You add your items to the basket, head to checkout and select your card, Apple Pay or even points, press “place order,” et voila. One to three business days later you’ll receive your package.

That process sounds simple, but it rarely feels that way. Our 2025 holiday shopping research found that 85% of shoppers say they’re likely to abandon their cart altogether out of frustration or indecision. As shoppers, we’ve adapted to this.

In an agentic environment, the checkout page, as we know it, effectively disappears. That final decision to “place order” isn’t there for you to make in the moment.

The agent makes it, optimizing across your payment options, rewards, points, or even financing methods you may not have considered, such as buy now, pay later. That shift has consequences that reach beyond just the payment itself. It touches everyone in the transaction.

Consumers are already letting agents handle the parts of shopping that feel like effort, even if they’re not fully comfortable relinquishing control of the payment quite yet. Our research shows that among active AI users, AI has already overtaken the physical store as the number one channel for product discovery. It’s a signal that the early stages of this shift are already here.

For brands, the risk is invisibility. For banks and payment providers, it’s irrelevance. In both cases, the loss is quiet. Because it’s not necessarily about the decisions a customer makes, but the ones an agent makes on their behalf.

Subscribe now

To understand what it means in practice, I spoke with three colleagues: Jennifer Ducarre, who leads Accenture’s consumer insights globally; Hannes Fourie, our global payments research lead; and Dominika Bosek-Rak, who leads payments research in Europe.

Their insights draw on two recent Accenture Research reports. The first, Agentic commerce rewrites payment choice, examines how the rise of AI agents is restructuring payment selection. The second, Talk to my AI agent: The new rules of brand value, looks at the shift from the consumer side and how agents are changing what brands need to deliver in order to be chosen.

Will the loyalty your brand built survive the agent layer?

Businesses have always competed for a place in the consumer’s wallet: the stored card, the saved credentials, the loyalty app. At the brand level, they compete for a place in the consumer’s mind: the instinctive reach for the same shampoo, the comfort of a familiar hotel chain, the retailer a shopper has come to know and trust.

That competition was won through marketing, habit and the friction of switching.

The agent layer changes the terms of that competition. When an AI agent handles the purchase journey—finding the product, comparing options, selecting the payment method—it doesn’t reach for what’s familiar. It reaches for what’s optimal, including the payment. “You would engage the agent and say, find me this product,” said Hannes. “The agent comes back and says, I found it and by the way, based on all the information I have from you, I can optimize payment using your points from this wallet for that store, and I can make the transaction for you.”

What makes this moment distinctive is that no single player is exempt from the disruption.

Consumers are finding that agents will automate the parts of shopping that feel like effort, but expect to stay involved in the moments that matter to them personally. Brands are discovering that visibility in a search result doesn’t guarantee selection by an agent. Payment providers are learning that being the default card in a wallet is no longer enough. And banks are realizing that the infrastructure they built for human-speed transactions may not be ready for what agents will demand.

We look at this shift from two angles simultaneously: the consumer side and the merchant and payments side. The throughline is that the loyalty businesses built, whether with customers, payment methods or vendors, will only survive if it was built on genuine value. Everything else is being quietly renegotiated as agents make decisions that humans used to make by habit.

Consumers are delegating, not disappearing

The redistribution of roles is already visible, and it looks different depending on where you sit in the transaction.

Accenture’s consumer research maps this progression through what it calls the delegation dial. Consumers aren’t handing over control all at once. They’re deciding, purchase by purchase, how much they want an agent involved and how much they want to stay in the loop.

For example, 74% are already comfortable letting an agent act on their behalf for routine tasks like handling returns or replenishing everyday items. Nearly a third would go further and let an agent decide what to buy for them, from choosing all the items in their grocery basket to deciding which telco provider to switch to.

The hold-out is payments. Only 12% are currently willing to let an agent complete a purchase autonomously and only 9% are open to fully autonomous shopping.

Consumers are willing to extend that trust incrementally, starting small and moving up as confidence builds. One in three consumers told us that success with small, low-stakes purchases will drive adoption of agentic payments for more expensive, riskier items. “They just need to see it to believe it today,” said Jen.

Regardless of this discomfort at the payment level, this is still a huge shift for brands. The world in which the agent is deciding what to buy, even if the consumer makes the payments themselves, reimagines the consumer-brand relationship, explained Jen.

“It changes who is making the purchase decision, it changes what they buy, it changes where they buy it,” she said. “All of that has a huge impact on the ways in which brands generate growth, the ways in which value exists in our industries.”

When the agent picks the payment method

Hannes describes the agent as an entirely new channel, one that banks, payment providers and brands will need to learn to compete within, just as they once had to learn search, social and mobile.

For payment providers, agentic commerce levels the playing field. Traditionally, cards benefited from being the default payment method stored in a consumer’s wallet, making them the path of least resistance at checkout. In an agent-driven environment, however, an AI agent can evaluate multiple payment options, including cards, account-to-account payments, digital wallets, loyalty points and emerging payment rails, and select the one that best aligns with the user’s preferences and the merchant’s objectives. The winning payment method may be the one that offers the lowest cost, fastest settlement, strongest rewards or highest likelihood of approval.

“Payment providers can no longer rely solely on default wallet status,” said Hannes. “They must compete on the value they deliver to both consumers and merchants.”

The fraud question compounds the pressure. Agents transact at any hour, across time zones, at a volume no human can match. Accenture’s research finds that 78% of financial and payments leaders already expect fraud to increase significantly with agentic payments and 87% believe trust will be the key barrier to adoption.

Yet 60% are still relying on standard workflows to investigate agent-driven fraud, a number that reflects how quickly this shift is outpacing the systems built to manage it.

“Banks and payment companies are rethinking everything around authorization,” said Hannes. “The amount of transactions will be higher and they will happen outside of regular business hours.”

In both cases, the mechanism is the same; agents don’t carry loyalty forward unless it’s been earned through actual value. The brands and banks that built market position on friction, habit or marketing visibility are the ones with the greatest challenges at the agent layer.

Three questions worth sitting with

Here’s what stuck with me after these conversations.

Are you building trust or just building agents? Jen notes that the largest global payment companies are already creating their own agentic payment solutions and that consumer comfort will grow as those efforts succeed. But the technology being ready and the consumer being ready are two different timelines. “If trust builds, this could grow,” she said. The risk for payments leaders is investing in capability while underinvesting in the visible proof points that move consumers from 12% comfortable to something larger.

Where do you want to compete? Hannes’s research identifies three positions: own the agent, become the payment of choice by embedding into agent workflows or own the control layer that governs how money moves through rules, permissions and verified identity. The window to choose deliberately is narrow.

Are you using the trust you already have? Dominika points out that banks already hold something the agentic ecosystem needs most: established trust. In several markets, people already use bank credentials to verify identity and access public services. “Banks may be the entities that will preserve this trust and enable agent commerce,” she said. “This is a big opportunity for them.”

Worth Your Attention

• The payment rails that agents will rely on are themselves being rebuilt. In a June report, Citi Institute maps how major financial infrastructure players are moving from experimentation to deployment and what it means for institutions that want to control the rails of the next financial system.

• Accenture’s recent report, AI agents are rewriting the platform playbook, examines why the model most digital platforms were built on is under pressure as AI agents make purchase decisions before a consumer ever reaches a search result and what platform leaders need to do to stay relevant when the decision moment moves upstream.

• Check out Accenture Research Journal, our interactive platform that lets you explore, question and navigate insights from the hundreds of reports we publish annually.

Overheard

— Jack Forestell, Chief Product & Strategy Officer, Visa, speaking at Wolfe Research FinTech Forum in March

The insights above are made possible by 350 researchers, editors and AI agents across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

In this issue, I examine:

• How agentic AI is making the case for adaptive governance

• The four-layer trust architecture that’s becoming the new standard

• What leaders can do to build trust into their agentic systems from the start

Reading the Curve

Something interesting happens when you talk about AI to different audiences. I have been on stage twice this month—once in Europe, once in New York—and I spotted subtle differences. Not in the questions exactly, but in the mood. The anxieties were different. The appetite was different. What struck me, though, is that underneath all of it, people were asking the same thing: can we trust this?

Part of what makes that question so charged is what AI systems actually are. They speak in human language. They take actions that look like human decisions. They learn from how we use them.

My colleague Philippe Roussiere, who is Accenture Research Innovation lead and drives Responsible AI research projects, describes agentic AI as more “human” in nature than other technologies, and I think he’s right. Precisely because of that, these systems require governance that keeps people responsible for defining their purpose, setting their boundaries and deciding when their behavior must change.

My answer, meanwhile, to the question of trust is that transparency is a large part of what makes trust possible. Nobody in the general public asked for today’s AI-driven disruption. It just arrived. It’s also entirely logical that this disruption provokes questions. What helps is not reassurance but clarity. We must invite people to see what’s actually inside these systems and how they’re governed. That is the spirit in which this issue was written.

Subscribe now

I spoke with Philippe and Patrick (“Paddy”) Connolly, global research lead for AI trust at Accenture Research, who has spent years researching what it takes to make AI trustworthy in practice. We discussed how AI trust now requires something more than a policy document or compliance framework; it requires an architecture, one where responsibility is designed in from the beginning.

And we introduce an idea still taking shape: what if trustworthiness could be measured as a standard, persistently, over time? We know from research that trust in AI operates on three levels: cognitive, safety and emotional. The first two relate to capability and security: can the system do what it claims and can it be trusted not to misuse what it knows. But the third dimension, whether the system feels genuinely on your side, often gets overlooked. Therefore measuring trust means accounting for all three dimensions, not just the ones that are easiest to quantify.

We don’t have the answer yet. But the question tells us something important about where this is heading.

Put your AI governance where your agents can use it

Inside most organizations is a document that contains their AI governance policy. A set of principles carefully drafted, reviewed by legal and approved by leadership. It states, in clear language, what the organization’s AI systems are expected to do and how they should behave.

In many cases, the machines themselves have never “seen” it. They weren’t designed to. But agentic AI, systems that act and evolve in real time, has now made that a blind spot that organizations can’t afford to ignore.

As agents begin to act on behalf of organizations and increasingly on behalf of each other, trust can no longer be assumed. “If my agent is buying a sweater off your agent,” Paddy told me, “how can I trust that agent to be doing what it’s saying it’s doing? Now imagine thousands of these running around an organization all day making decisions, taking actions and interacting with customers. How do you retain control at that scale?”

This question points to a change that most organizations haven’t fully reckoned with yet. Governing AI used to mean writing policies, forming committees and running pre-deployment checks. It was often, as Paddy put it, a “side-of-desk” responsibility for C-suite, a secondary priority.

But the old approach is no longer sufficient. “Agentic systems are non-deterministic, dynamic and ever-evolving,” he said. “They will err, they will make mistakes, they will have ‘behaviors’ that shift with new capabilities, environments or interactions . Policies used to get written up. Now they have to be hard-coded as controls that adaptively manage the agent.”

Governance is a growth strategy

For a long time, companies could get away with a traditional risk management framework. AI adoption was low, the models were static and narrow and the potential risks and mitigations were well understood. “With hindsight, traditional risk management seems almost straightforward in many cases, with the majority of work occurring pre-deployment,” Paddy explained. “It was essential, but it struggled to justify its cost in many eyes. This is reflected in our Global Responsible AI Index, where no company was ranked as fully mature.”

Today, however, as organizations move to deploy agents at scale, the risk of getting it wrong compounds quickly. “Anyone can build an agent, but you can’t unlock value without really getting your arms around the problems first. Risk now emerges in production in ways that are hard to anticipate,” Paddy said.

What’s changing, he argues, is how companies are starting to see the problem. AI governance is no longer perceived as a compliance cost. It’s what makes scaling AI possible at all. For organizations that are serious about becoming AI-native, getting governance right is no longer optional. “Organizations are recognizing this, but they can’t do it in a controlled way using traditional approaches,” he said. “Managing risk now becomes a value lever, the question is how do we redesign AI governance for Agents?”

Building a trust architecture

What Paddy and his team developed is a “trust architecture”—four interdependent layers that together define, enforce, monitor and explain how agents behave in the real world.

The governance layer is where the rules are encoded. They’re not written in a document, but are built into the system itself. The governance layer defines an agent’s goals, the boundaries it must operate within and the conditions under which a human must step in. Paddy describes it as the agent’s constitution: the standing orders that shape behavior before action is taken.

The security layer is what enforces those rules at the technical level. It controls how agents access systems, data and tools, and ensures they can only operate within identity-verified, authorized boundaries. This is also where orchestration lives: the logic that allows an agent to interpret, reason, plan and act, within the limits the governance layer sets.

The assurance layer is the monitoring layer. Also known as instrumentation, this tracks an agent’s behavior in real time. “You can track exactly what your agent does, what decisions it took, what application programming interfaces (APIs) it used, what data it accessed,” Paddy said.

The accountability and oversight layer creates the permanent record. It logs not just what an agent did, but why, producing decision traces that can be read, explained and defended long after the fact. This allows an organization to respond to a legal challenge, a regulatory audit or an internal inquiry with confidence. At the base of this layer sits what could be considered something like a “trust score,” a mechanism that aggregates behavioral data over time into a persistent measure of whether an agent is doing what it was designed to do. Paddy and his team are working with Stanford University to develop this concept further.

The architecture is not a one-way flow. Data from the assurance and accountability layers feeds back to the top layer, creating an “adaptive governance.”

“Whereas before we could have set policies that we wanted to live by through the lifecycle of that agent, those policies now have to change and evolve in real time,” Paddy said.

Importantly, this architecture makes human judgment more meaningful. The goal is to keep humans genuinely in the lead. That means giving people the right skills, training and context, not simply assigning them responsibility without the tools to carry out the job. When the latter happens, we talk about the employee becoming a moral crumple zone, and our ongoing work with MIT has focused on the dual role of employees in mitigating risks and optimizing performance.

The better question is how to design the human-AI relationship thoughtfully from the outset, added Paddy. “The real challenge for business leaders is to safely scale decision authority, work distribution, accountability and oversight across humans and AI, while maintaining organizational control. We must remain grounded in our people.”

As part of this, he sees that dedicated “agent ops” teams are already emerging. These people monitor, intervene and update agents in real time, much as cybersecurity teams manage threats today.

Still, new software alone won’t be enough. Organizations will need new roles, new training and a new way of thinking about where humans fit in a system that is, by design, always running.

Where leaders can start

1. Begin with a real problem and a clear reason to solve it. Identify a genuine challenge in your organization where an agent could make a real difference and be honest about why the challenge exists. Don’t deploy agents simply because the technology is available. Start focused, start small and scale autonomy in stages as your people, processes and controls mature alongside the system. The architecture will need to grow with you. As Philippe put it, “It’s never a once-and-done game. It’s constantly evolving.”

2. Build in moments that make people think. One of the most practical ideas to emerge is what Renée Richardson Gosline of MIT calls “cognitive speed bumps,” which are deliberate pauses built into agentic systems that prompt a human to check, question or verify before acting on an output. The instinct in the digital age was to remove all friction. But in the agentic age, some friction is the point. For example, a confidence score that flags a low-reliability answer or a prompt that asks whether you’ve verified a source are small design decisions with significant consequences for trust.

3. Take people with you. Showing your workforce that you care about how AI affects their roles is an act of leadership. And it’s not just the Chief AI Officer’s job to show such leadership. Agentic AI is a profoundly human technology, said Philippe. “It talks to humans and it does actions on behalf of humans that look like they’ve been taken by humans. So that means something fundamentally important is happening.” This reality makes the people inside your company some of your most valuable guides. Survey them. Run a hackathon or a promptathon. Invite them to shape how agents are used in their own workflows.

Worth Your Attention

• Turn AI into a force for responsibility: In a world focused on risks connected to AI, my Accenture colleagues Laetitia Cailleteau, Philippe Roussiere and Josh Elkind make the case that AI can help companies behave more responsibly. Where we focus above on the architecture that governs how agents behave, this article in European Business Review zooms out to the broader organizational question: how do you embed ethics into decisions, develop your people and build a company that can adapt without losing its values?

• Want to learn more about trust and AI? Head to Accenture Research Journal, where you can query the full body of Accenture Research reports and get answers grounded in empirical findings.

The insights above are made possible by 350 researchers, editors and AI agents across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture

In this issue, I examine:

• What the probability stack is and why it matters

• Which business problems it was built to solve

• How leaders can start using it to question old assumptions

Reading the Curve

In business, management guidelines are based around the idea that the right answer exists and the job of a leader is to find it. It’s a worldview the French philosopher René Descartes would have recognized: reason as the path to certainty, logic as the ultimate guide.

Our technology followed the same logic. The computers are viewed as “certainty machines.” They operate on fixed rules and deterministic outputs.

But there is a category of problems that this foundation was never built to handle. Not because the technology wasn’t good enough, but because, as in life, some problems don’t have a single right answer. They live in uncertainty. Demand that shifts without warning. Supply chains disrupted by events no model predicted. Drug molecules with thousands of possible configurations.

For decades, businesses have worked around these trickiest challenges. They built competitive strategies on tolerating them better than the competition.

Today, however, a new capability is emerging with the help of something called the probability stack.

This is the idea that across both software and hardware, we’re gaining the ability to compute in degrees of likelihood, rather than fixed rules, and this opens up an entirely different class of problems for businesses to solve.

As someone who has spent a career looking for what the data says beneath the surface, I’ve been tracking what the convergence of probabilistic software and increasingly powerful hardware means for the decisions that leaders make every day. To help me unpack it, I spoke with my colleagues Ari Bernstein, who leads Accenture’s Technology Vision, the annual report that tracks how emerging technology shapes business; and Prashant Shukla, Principal Director, Thought Leadership Research. Both have been researching what the probability stack means for leaders.

What struck me most in my conversations? The leaders who fall furthest behind the curve won’t be the ones who ignore these new tools. Instead, they’ll be the people who never thought to question which of their industry’s oldest assumptions might no longer hold.

Subscribe now

And as a researcher, the question I keep coming back to is this: Can this new way of seeing the world help us finally anticipate the outliers—the unexpected disruptions, the opportunities hiding at the edges? That may be the most important question business leaders aren’t yet asking.

That’s what I set out to explore.

The probability stack is redefining what’s possible

James G. March in his seminal 1991 paper, Exploration and Exploitation in Organizational Learning reasoned that successful organizations must balance two competing imperatives: exploitation, or the refinement, efficiency and execution of what they already know; and exploration, or the search for new knowledge, experiments and the unknown.

For decades, organizations have relied heavily on exploitation because repetition builds efficiency and scale. However, the environment facing organizations today is different from the one that March described, Prashant told me.

“Specifically, the cycle times of technological disruption and competitive change have compressed dramatically,” he said. “As a result, now companies need to, perhaps for the first time, over-index on building the exploration muscle, the muscle prepared for different probabilities.”

The goal is to complement operational excellence with an equally strong probabilistic muscle. And technology is just a variable in the equation.

Our brains already know how to do this.

Probability is simply a part of how the world works. Our brains are constantly taking in information and calculating probability in real time.

Every time you glance at the sky and decide whether to grab an umbrella or size up a merge on the highway, that’s probability at work.

Those are simple calculations you’re used to making every day. Your brain can run far more complex ones without you even knowing it. Picture a professional baseball player. At the highest levels, a 96 mph fastball moves quicker than the synapses between the brain and the eyes can process. Instead, the batter watches the pitcher wind up and the brain builds a model—calculating when the ball will release, its likely trajectory, when it will arrive—and the batter swings (or not). We call it reflexes or muscle memory, but it’s something much deeper. It’s the brain running a probabilistic calculation and acting on likelihood rather than certainty.

Our brains are built for this but our machines were never built to harness this type of uncertainty. But that’s what’s changing, explained Ari.

“There is an entire side of the world, the messy unpredictable one, that we weren’t able to compute before because our machines were designed to be deterministic,” he said.

To be clear, we’re not conquering uncertainty. Rather, we’re finally building machines that can meet the world as it actually is. The challenge for leaders now is whether they are ready to see what those machines are showing them.

The problem is the lens.

The gap between what machines can do and how we’re judging them is exactly what makes the probability stack so easy to underestimate. Ari shared this famous example:

In 2016, Google’s AlphaGo was playing a grandmaster in the ancient game of Go. On the 37th move, the machine made a play so unconventional that the commentators assumed it had malfunctioned. Not exactly. That move won the game.

“If we’re only looking at it through the lens of how the game is played today, we can see some of these probabilistic outcomes and think, ‘that’s wrong, that’s a hallucination,’ the way these people reacted to that one move,” he said. “But we should be thinking, this can actually teach us how to play the game differently.”

In other words, we have been watching machines make move 37 and calling it a mistake. That’s why the leaders who pull ahead will be those who stop and ask what the machine might know that we don’t.

The assumptions your strategy is built on may not hold.

Every industry has problems so difficult that the entire competitive landscape has been built around tolerating them. In logistics, for example, it’s routing efficiency. In pharmaceuticals, it’s understanding how proteins fold to develop new drugs. In materials science, it’s understanding chemical interactions at a molecular level.

Consider a global products company whose competitive advantage rests entirely on the superior materials in what they make, using knowledge built over decades that was, until recently, nearly impossible to replicate. If a competitor can now use AI to generate and test hundreds of thousands of material compounds in the time it once took to test a handful, how deep is the global products company’s moat, really?

Ari pointed to Insilico Medicine, which used generative AI to identify and nominate promising drug candidates in 12 to 18 months on average, a process that typically takes four-and-a-half years.

“The audacity or ambition to try and tackle some of these problems is something that you should really realize is becoming a capability of the technologies,” he said.

The starting point isn’t a technology budget or a vendor evaluation. It’s a harder internal question: Which of our biggest challenges have we quietly stopped trying to solve?

This won’t happen overnight and that’s exactly the point.

The probability stack is still being built and many of its most powerful applications are still taking shape. The companies that will lead aren’t necessarily the ones moving fastest right now. They’re the ones having this conversation, who are building the understanding, revisiting their assumptions and laying the groundwork before it becomes urgent.

“This isn’t something you should expect to solve or expect to be a competitive threat tomorrow,” Ari told me. “But companies take a very long time to move. It’s not a Titanic moment yet, where you’re about to hit the iceberg. But you are leaving the dock in London without asking whether there are icebergs out there.”

Learning Curve

Not every problem needs the probability stack. Part of developing the right judgment is knowing which problems call for which approach.

The framework below, drawn directly from our research, maps problems along two dimensions: how uncertain the problem is by nature and how costly it would be to get it wrong. Where those two factors intersect tells which method to reach for.

Areas like fraud detection, drug discovery and cybersecurity response are where many businesses are under-equipped because they’ve been forcing deterministic tools onto problems that were never going to yield a single right answer.

Worth Your Attention

• How to get ready for quantum from California Management Review: Quantum computing is closer to practical business use than most leaders realize. This piece, by my colleagues Adam Burden, Carl Dukatz, Shreyas Ramesh and Laura Converso, makes a clear case that the window to get ready is open now. Companies, meanwhile, that wait for broad adoption to force their hand will already be behind.

• Ask us anything! Head over to Accenture Research Journal, where you can query the full body of Accenture Research reports and get answers grounded in empirical findings.

Overheard

“The true winners we have seen in the past are the ones who have moved when things were still ambiguous because it allows them to set the rules of the game … Don’t wait for abundant clarity. The winners will be the ones who move faster than waiting for that ideal moment to arrive.”

— Rajat Agarwal, Commerce Lead, Accenture Song, and co-author of the new report, Agentic commerce: Make your brand unmissable

The insights above are made possible by 350 researchers, editors and AI agents across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture

In this issue, I examine:

• What agentic commerce demands of businesses

• The two paths every company must consider (most are only pursuing one)

• Why personal relevance is becoming a competitive advantage

Reading the Curve

Digital commerce has delivered on many of its original promises. Shopping is faster, more convenient and available at any hour. And yet, anyone who’s tried to book a transatlantic flight with the family cat knows the experience can still unravel quickly. You find the right flight, but then you find the pet policy buried in an FAQ that’s different on every airline, with a cabin limit nobody advertises clearly. You call to confirm. You get transferred. You start over.

Given how far we’ve come, this complexity shouldn’t still exist.

That gap between what digital commerce promised and what it actually delivers is what makes the latest research from our team worth paying attention to. As one of my colleagues put it, digital commerce has largely been tolerated. It’s certainly not loved. Now, the architecture of commerce itself is being reconsidered.

To understand what this change means operationally, I spoke with three of my colleagues at Accenture Research who approach this issue from very different angles: payments, trust and the broader commercial landscape.

Subscribe now

What struck me most is that while each of them is looking at a different part of the problem, they all arrive at the same conclusion: We are at another genuinely transformational moment in commerce, one that rivals the earliest days of e-commerce in its scope. Businesses that move now will have a real advantage.

Your next customer might be a machine

Nearly three decades ago, businesses had to rebuild for the internet age. New storefronts, new logistics, new ways of reaching customers. Companies that moved early defined their categories.

The shift underway now has a similar shape, except that the new customer arriving at the door won’t always be a human browsing a website. It could be an AI agent acting on someone’s behalf, with no patience for friction and no emotional attachment to your brand.

Agentic commerce is the term for what happens when AI agents—software systems that can reason, plan and act autonomously—take over some or all of the steps involved in finding, evaluating and purchasing goods and services on behalf of an individual or organization.

To help me unpack the implications for brand leaders, I spoke with Josh Bellin, who leads research on customers and growth at Accenture and who co-authored the new report, Agentic commerce: Make your brand unmissable; Hannes Fourie, Accenture’s global payments research lead; and Surya Mukherjee, a researcher who has been examining trust and the human dimensions of AI adoption.

Here’s what they told me:

1. The opportunity is bigger than most companies realize

Today, a human browses, compares and clicks. In an agentic environment, that entire process can be delegated. And that delegation changes everything about how businesses need to show up.

There are two distinct ways to respond to this shift and most companies are only thinking about one.

The first is becoming the “choice of agents”—ensuring your products and services are visible, structured and compelling enough that AI agents find and recommend you. “You want to be the solution to a problem. It’s no longer just, ‘how do I show up in search?’” Josh said.

The second is becoming the “agent of choice”—building your own agent that serves your customers directly. Companies with deep category authority, rich first-party data and the ability to orchestrate end-to-end commerce are best positioned to own that second path. But most aren’t there yet.

Josh pointed to an example that illustrates the agent of choice scenario well. Think about a major skincare brand. These companies already possess deep category expertise and deep consumer trust. In an agentic environment, that combination becomes a genuine strategic asset.

As Josh told me, a customer might share a photo of their skin with a brand’s agent, describe what’s bothering them and ask what they need. Or they might return to an agent that already knows their history. It could go something like this: “We’ve talked about my itchy scalp before. Do you have a product that can soothe it?”

The brand is no longer just a product on a shelf. It becomes the trusted guide that helps a customer understand their own needs.

This is exactly the kind of experience that generates net new demand. Our research in partnership with Aaru, which simulated 50,000 synthetic consumer agents across 24 countries, suggests that agentic commerce will generate entirely new buying moments that did not previously exist. Indeed, within 18 to 24 months, a majority of consumers are projected to complete purchases inside AI conversations that didn’t start with an intent to buy.

The agent-assisted purchasing journey will begin with a consumer sharing their need before they even know exactly what they want or whether buying is the answer. As a result, the total number of buying moments for the average consumer is expected to expand.

Not every industry will feel this shift in the same way or at the same pace. The chart below maps industries by two variables: the net-new demand agents are expected to generate and the execution risk organizations face in meeting it.

The demand side gets most of the attention. But business leaders should study the execution risk axis, too. In categories like travel and health and wellness, the ability to deliver on inventory, payments, fulfillment and customer service will determine whether an agent recommends you or moves on.

2. The back end is now a front-line issue

Most conversations about agentic commerce focus on visibility; that is how to get found and how to get recommended. Hannes redirected my attention to what happens at checkout.

In an agentic environment, the checkout page as we know it effectively disappears. An agent that has already found the right product, compared options and verified availability will also select the best payment method. It can optimize rewards, speed, cost and trust on the buyer’s behalf.

“The decision layer is moving up,” explained Hannes. Payment is no longer the last step a human takes. It’s one more variable an agent manages.

Because the agent already knows you and has your payment credentials and even your loyalty points, it can optimize a transaction in ways a human shopper might not think to do. It might apply store points you forgot you had, route the payment through the most cost-effective method or surface a discount tied to your account.

This has significant implications for payment providers, merchants and any business that processes transactions at scale. For example, merchants currently pay around 3% in fees when a credit card is used for a purchase. But in an agentic environment, an agent could route a transaction through a bank transfer or account-to-account payment, saving the merchant money and potentially passing some of those savings on to the customer.

The businesses that act now will have a real advantage. “If your payment is not discoverable, your volume might decline,” said Hannes. “Therefore your revenue from payments declines.”

The risk goes beyond revenue. If your brand isn’t showing up at the agent layer, customers may simply stop thinking about you altogether.

The fraud question is equally urgent. Agents can shop at any hour of the day or night. They can transact across time zones and at a volume and frequency no human can match. Early research from Accenture indicates that a significant majority of financial and payment leaders already expect fraud to increase with agentic payments and that most are still relying on standard workflows to investigate fraud.

“Banks and payment companies need to rethink everything around authorization,” added Hannes. “The sheer amount of transactions will be more and they could happen outside of regular business hours. That’s the power of what agents will bring to this.”

3. The most valuable thing a brand can do now is make its audiences feel seen

Surya has been researching the issue of trust in AI: how it forms, what breaks it and what it requires of organizations trying to build it. He and his colleagues identified something that most businesses entirely miss.

Trust in AI, they argue, isn’t one-dimensional. It has several dimensions.

There is a rational, cognitive dimension: Can the agent do what a competent human would do? Get the answer right, execute the task accurately.

There is a safety dimension: Is my data secure? Is the system behaving within appropriate boundaries?

Most enterprise AI systems are built around these two dimensions. They are necessary, but they aren’t sufficient.

There is a third dimension that companies consistently overlook: the emotional dimension of trust. The sense that the agent genuinely gets you without being creepy. “We used to call it relevance,” Surya said. “But it’s morphing more into the territory of being seen, heard and understood.”

This matters enormously in an agentic environment. As agents handle more routine tasks, such as browsing, comparing and transacting, the moments when a consumer engages directly with a company become fewer. Yet those moments become far more intentional. A customer who shows up in the agentic era isn’t browsing; they’re there for a reason.

“To win in these moments, you must double down on things like intimacy, relevance, experience,” Surya said. Companies that show up with genuine personal relevance will build the kind of trust that both people and agents learn to recognize.

Learning Curve

Digital commerce has always had a patience problem: The majority of consumers are likely to walk away from an online purchase because of frustration or indecision, according to previous research by Accenture. That number has been sitting quietly in the background for years. It’s accepted as the cost of doing business online.

Agentic commerce has the potential to change that. Agents don’t get frustrated. They don’t get overwhelmed by options. They just execute.

That’s why businesses need to determine now if they’re ready for what’s coming before the old model fully breaks down.

Worth Your Attention

• An anecdote about angling with AI technology caught Surya’s attention. In this podcast, Wall Street Journal contributor Ken Wells recounts how technology changes the nature of skill, intuition and discovery in sport fishing. Surya recommends it as an unexpected lens on where AI and personalization are heading.

• Josh recommends What we lose when artificial intelligence does our shopping. This piece, from Mark Bartholomew, Professor of Law at the University at Buffalo and Samuel Becher, Professor of Law at Victoria University of Wellington, is a useful collection of links that deepen the questions we can ask about the future of agentic commerce.

• Ask us anything! Head over to Accenture Research Journal, where you can query the full body of Accenture Research reports and get answers grounded in empirical findings.

— Vidhya Srinivasan, Google’s vice president and general manager of advertising and commerce, in Fast Company

The insights above are made possible by 350 researchers, editors and AI agents across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

In this issue, I examine:

• How foundational systems are being rebuilt for AI

• The signals that leaders should have on their radar

• The growing gap between capability and governance

Reading the Curve

Much of the conversation around emerging technology tends to focus on what’s most visible. The latest models, the newest applications, the breakthroughs that make headlines. These moments matter, but they’re often the result of changes that have been percolating for some time, out of view.

For those trying to understand where the curve is moving, the harder question isn’t just what’s happening now, but what signals are forming beneath the surface and how to interpret them before they become obvious.

That’s what drew my attention to a recent piece of work from my colleagues in Accenture Research, led by , Accenture’s global managing director of IT research. They set out to build a way to track how emerging technologies evolve across different stages of their lifecycle.

What they developed is a ranking that analyzes activity from three environments: ArXiv, where labs and R&D teams post papers ahead of peer review; Hugging Face, where models and datasets are shared and downloaded; and GitHub, where code repositories are built and maintained.

This report is the first edition of an ongoing tracker, not a static snapshot. The findings here are grounded in a carefully validated analysis of activity on those three platforms from the latter half of 2025, which is the foundation on which each subsequent edition will build. The study is designed to evolve: future editions will layer in fresher data, letting readers watch these technology signals develop in near real time.

Read together, these signals offer a more holistic and dynamic view of how technologies move from early research to real-world use.

For example, at first glance, the story might seem familiar: models and frameworks continue to dominate attention. But look more closely and you can see that the foundational machinery of computing—databases, compilers, testing systems—is being reframed and rebuilt for a new AI paradigm.

Subscribe now

What appears to be a simple ranking starts to resemble a layered architecture, where each technology depends on the ones beneath it. And perhaps most notably, there are early signs of a gap forming between the pace at which new capabilities are advancing and the systems required to govern and sustain them.

These aren’t definitive conclusions. They are ways of interpreting the curve as it takes shape.

To explore the ranking more deeply, I asked Jim to walk us through what he sees in the data and what it may mean for decision makers.

Inside the ranking: What’s moving and why

Hello, Jim here. When we started this work, our goal was to find a better way to understand how emerging technologies take shape.

One example I often come back to is the 2017 “Transformer” paper (“Attention Is All You Need”). When it appeared on ArXiv as a research breakthrough, it was one signal among many. A few years later, it became the foundation for systems like ChatGPT. So it got us thinking, how do you spot those kinds of shifts earlier?

So with my Accenture research colleagues Shruti Shalini, Utkarsh Londhe and Rhine Andotra, we built a structured way to track how technologies move across the ecosystem.

To do that, we focused on three environments that capture different stages of a technology’s lifecycle:

• ArXiv captures early research and innovation.

• Hugging Face reflects models and tools entering practice.

• GitHub shows what code is actually being built and deployed.

We then organized 15 technology domains across the AI stack—from models and code generation to infrastructure, testing and deployment—and ranked them based on activity across these three signals.

The goal isn’t just to see what’s popular, but to understand where a technology sits and how it’s moving from early research to real-world use.

Source: Accenture Research analysis based on ArXiv, GitHub and Hugging Face data

But the ranking itself is only the starting point. The more interesting work is to interrogate it. What patterns does it suggest? What might it be missing? And what does it imply about how the field is evolving?

Below are three ideas we found worth examining.

1. Classic computing, rewired for AI

Much of the attention in AI focuses on models and frameworks. They dominate the top of the ranking, as expected.

But nearly half of the technologies in this list sit in the middle and lower tiers. And many of them aren’t new. Databases, compilers, testing frameworks and infrastructure tools have existed for decades. What the ranking reveals is something more surprising: the foundational machinery of computing is being reinvented in real time for the AI era.

Agentic systems require memory, execution, verification, deployment and knowledge. Each of those needs maps to a classical engineering discipline that is now being redesigned to support AI workloads.

At a recent Google Cloud Next conference, compilers were described not as programming tools, but as infrastructure for the agentic era; they’re the systems that translate intent into execution.

In other words, the future of AI may depend as much on the reinvention of existing systems as on the invention of new ones.

Implication for leaders: Don’t just invest in the newest AI tools. Much of the real work is happening in the systems underneath them: data, infrastructure, testing and execution. Make sure your teams are strengthening those foundations, because that’s what will determine whether AI works at scale.

2. A system of layers

If you read the ranking from top to bottom, a different pattern emerges. The technologies begin to resemble layers and the ranking becomes both a leaderboard and an architecture.

At the top are capabilities: models (#1), orchestration (#2), fine-tuning (#3) and inference (#5), with code generation (#8) close behind.

Below them are systems that connect those capabilities to real-world use: evaluation (#6), multimodal frameworks (#7) and edge deployment (#10).

And beneath that sits a deeper layer of infrastructure: operations (#9), cloud environments (#11), testing (#12) and data systems (#13–#15), alongside compiler innovations (#4) that cut across layers.

The ranking suggests that innovation is happening architecturally, not randomly.

So where does an organization’s capability stop? Many teams invest heavily in the top layer because it is visible and easy to demonstrate. The ranking suggests that the harder work—and often the limiting factor—sits below.

What appears to be a measure of momentum may, in fact, be a map of dependency.

Implication for leaders: If your focus is only on the top layer (models and applications), you’re likely building something that looks impressive but won’t hold up in production. Take stock of where your gaps are across the full stack, especially in the middle and lower layers where real reliability is built.

3. The trust gap

A third pattern appears when you group the technologies differently. The gap between what we can build and what we can trust may be widening.

Some are focused on expanding capability: models, code generation, orchestration. Others are focused on making those systems reliable: evaluation, testing, operations and governance.

In this ranking, the capability-building technologies consistently appear higher than the trust-building ones.

That gap raises a question: Are we building systems faster than we’re developing the means to verify and manage them?

This would not be unprecedented. In previous technology cycles, capability often outpaced discipline before catching up. But with AI, which is being designed to take intelligent (and sometimes autonomous) action, it’s much more concerning.

If that is the case here, the ranking may be showing more than momentum; it may be highlighting a structural imbalance that will shape where attention and investment need to move next.

Implication for leaders: If you are moving quickly on AI capabilities, you need to move just as quickly on Responsible AI systems for testing, ensuring safety, explainability, monitoring and controls. Otherwise, you risk deploying systems you can’t fully trust. Therefore, treat governance and reliability as part of the build.

These aren’t conclusions. They’re prompts.

What this ranking offers is a way to observe how technologies evolve across layers and to ask more precise questions about where they are headed and what that means for the organizations trying to build with them.

We plan to revisit this analysis in the coming months, updating the data and looking for changes in the pattern. Will the technologies focused on testing, evaluation and governance begin to move up as companies push toward production? Will new domains enter the ranking as the field expands? Will models continue to dominate or will the underlying systems start to carry more weight?

Those are the questions we’ll be watching. Follow along as we continue to track how the curve moves.

Worth Your Attention

• The method behind this ranking was inspired by Richard Posner, the University of Chicago scholar, who, in Public Intellectuals (Harvard University Press, 2001) ranked influence by triangulating across three distinct signal environments—an approach we first adapted for an earlier ranking and now apply to the emerging technology landscape.

• @importai is an interesting weekly Substack newsletter that explores frontier AI research written by Jack Clark, co-founded of Anthropic.

• @exponentialview is the Substack newsletter written by Azeem Azhar, a UK-based technology analyst, author, and speaker.

• We have made some updates to our Accenture Research Journal. Check them out and let us know what you think!

— Stewart Brand, The Clock of the Long Now (1999)

The insights above are made possible by more than 300 researchers and editors across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

In this issue, we examine:

• Why data alone isn’t enough

• What continuous scenario planning looks like in practice

• How AI agents are changing the way leaders model risk

Reading the Curve

There’s a tendency, in business and in life, to plan around what feels likely. The scenarios we outline and decisions we make are often built on patterns we’ve seen before. But the events that most profoundly influence our lives are often those we don’t fully prepare for, even when their consequences can be severe. In other terms it is critical to anticipate potential events with low probability and high impact.

Author Nassim Nicholas Taleb captured this idea in his book The Black Swan. He used the now famous example of a turkey that grows more confident in its safety with each passing day it’s fed and cared for. That is until it meets its Thanksgiving Day fate. His point was not just about unpredictability (from the bird’s point of view), but about how easily we rely on past experience to guide future decisions.

In today’s environment, profound or life-altering events feel less like exceptions and more like a constant backdrop.

Disruptions—whether geopolitical, economic or technological—can cascade quickly, showing up in ways that are both immediate and tangible. Consider that nearly 20% of the world’s energy supply and roughly 30% of global fertilizer trade move through the Strait of Hormuz. For a manufacturer in Europe or a farmer in the American Midwest, a conflict thousands of miles away can land directly on the bottom line.

My colleague Amir Bagherpour, Accenture’s head of agentic modeling and simulation, has developed Accenture’s proprietary agentic simulation and monitoring capabilities that can be leveraged to strengthen the resilience of organizations. Scenarios can be run at speed to assess not only first order effects but also second order and third order effects of potential events.

Subscribe now

Advances in modeling and AI make it possible to build and test scenarios in real time, helping leaders understand how events might unfold and what they mean for the decisions in front of them.

As Amir put it to me recently, “It’s not about telling people exactly what’s going to happen. It’s about understanding the pathways and the scenarios on how things can happen.”

That shift from prediction to preparedness is where the curve is moving.

Good decisions in uncertain times require a different kind of intelligence

For business leaders navigating constant volatility, the challenge is not a lack of information. Data is plentiful. And geopolitical analysis is seemingly everywhere, 24/7.

The harder problem is translation.

When a major disruption hits, whether a global pandemic or a conflict that suddenly closes one of the world’s most critical energy corridors, the question that leaders pose is not “what’s happening?” It’s “what does this mean for us and what do we do about it?”

That gap between analysis and action is where most organizations have historically struggled.

For decades, scenario planning was the standard response. Leadership teams would gather periodically, map out a handful of plausible futures, assign rough probabilities and build contingency plans around the most likely outcomes. It was a serious and valuable exercise. It was also slow. Building a credible model of how a conflict might unfold or how a supply chain shock might ripple across the business required months of expert work and access to the kind of specialized talent that only the most sophisticated organizations could afford.

What’s changing now is the ability to course-correct continuously, sharpening the analysis as new information comes in.

A different kind of decision intelligence

Two colleagues who spend their days making sense of fast-moving, high-stakes information have shaped how I think about scenario planning. Tomas Castagnino has led Accenture Research’s global economic modeling capability for more than a decade. Svenja Falk heads our Accenture Research Geopolitical Center of Excellence and speaks with senior business leaders almost daily.

Both will tell you that what’s changed in recent years is more than the tools; it’s the nature of the work itself.

Castagnino describes the change through his own career. Early in his work as an economist, he estimates that the majority of his time went to manual data gathering: downloading records one by one, matching datasets and assembling the raw material that analysis requires. The intellectual work—the modeling and interpretation—accounted for 20% of the time in most projects.

“Now that’s flipped,” he said. “We’re spending more quality time trying to understand how to decode the noise in the data.”

This evolution in how economists spend their time reflects something larger. The data available to analysts today is fundamentally different from what existed even a decade ago.

Six years ago, as the pandemic unfolded in real time, modeling teams could track where people were spending their time, whether they were adopting new technologies and how consumer behavior was shifting week to week. Job postings, employee reviews, hiring patterns and technology adoption signals were all data sources that either didn’t exist a decade earlier or simply couldn’t be analyzed at scale. All of these new data streams made it possible to see things that had previously been invisible to analysts. And now agentic solutions are changing not only the speed and the scale of scenarios but the nature of scenarios building to prepare decisions of Executives.

Castagnino uses a black hole metaphor to describe what this data makes possible. Even something as difficult to measure, like pure gravity or how a virus could affect a global economy, can be inferred by observing the signals it creates.

“There are certain things you cannot truly measure (like a black hole), but you can actually get very close to understanding how they are moving just by observing all the data around it,” he said.

For Falk, who works directly with clients navigating such complexity, the challenge is keeping leaders focused on what matters for their business. “For a business leader, the most important thing is actually the decision itself,” she told me. “When I make a decision, what are the implications on my company and how do I take my leadership team along?”

Her team’s work starts with the questions that matter to a specific company. What will this mean for input costs? Pricing? Supply chain? That “translation,” from global signal to business decision, is where the real work happens. All the contextual information and insight is gathered and analyzed in real time through agents such as agents which run PESTLE (Political/Economic/Social/Technology/Legal/Environmental) analysis in seconds vs hours in the pre-agentic world.

A different way to make decisions

A few lessons stand out from my conversations with Tomas and Svenja:

The goal is focus. The volume of information available to leaders right now is staggering and it keeps growing. This is precisely where modeling earns its value, says Tomas. “Everything is data, but not everything is useful data,” he told me. “Use models to reduce the level of uncertainty. It’s like reducing the noise to signal ratio.” Asking the right questions is the first step. Scan available data and cluster them into a small number of drivers that shape outcomes is next. And then develop and pressure test a series of coherent potential futures.

Build agentic solutions combining specialized agents (say financial analysis, white space analysis), agents combining several specialized agents and orchestrator agents (coordinating the drafting of an overall analysis to prepare potential decisions). In the future, scenarios building will systematically include agentic simulations, under the supervision of humans, what we call “Humans in the lead”.

Prepare for more than one future simultaneously. In fast-moving situations, two scenarios can be equally plausible. A leader has to be equally prepared for both. That means understanding what might happen, as well as why and what each pathway would require of the business. We produce such research work regularly for clients and for Industries.

Use it to spot opportunity. A real-time modeling system is not only an early warning tool, it’s an early opportunity tool, says Svenja. “You can analyze intelligently so much faster,” she told me. “And you can match it with activities actually happening in the market.” Investments, competitive moves, shifts in demand: the same signals that flag risk can also reveal where others are placing their bets and where a window might be opening.

Treat scenario planning as a continuous practice. This is perhaps the biggest change. The tools now exist to monitor signals and update assumptions in real time. Organizations that build this into their regular operating rhythm, rather than revisiting it once a year, will be better positioned to act when conditions shift.

Learning Curve

Scenario planning with Agents

When you’re able to do things faster, then you’re also able to iterate more.

“It’s like your iPhone. The version you have now is significantly better than the one you had 10 years ago because there’s always updates and improvements,” says Amir Bagherpour, who leads Accenture’s agentic modeling and simulation.

Except in this metaphor, imagine that those updates occurred over the course of a week, not a decade.

What once required months of specialized effort, the kind of work that only the most elite research teams could pull off, can now be done in days. Sometimes hours. And that window keeps shrinking.

Amir spent his career building models that help decision-makers see around corners, including at the U.S. State Department, where he led this type of advanced decision analysis for the Secretary of State’s office.

Now, sophisticated analysis is becoming common practice in companies.

Amir walked me through the structure behind one of Accenture’s simulation platforms. The simulation itself starts with the kind of modeling that has long existed in academic and government circles, mapping stakeholder behavior, tracking who holds influence, charting the conditions under which a negotiation might hold or collapse.

What agents make possible is the next step: taking that complexity and making it accessible at speed.

Specialized agents handle distinct parts of the work. One continuously reads live data, tracking events and market signals in real time. Another translates that input into forecasts and simulated outcomes. An orchestrator coordinates the two.

Together, they form a system you can ask questions of directly.

Worth Your Attention

• The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence. The essential backstory to everything we explore in this issue. Based on more than 30 hours with the co-founder of Google DeepMind, author Sebastian Mallaby traces how one scientist’s audacious pursuit of artificial general intelligence produced the tools that are now reshaping how leaders model risk and act under uncertainty.

• LLMs as Strategic Actors: Behavioral Alignment, Risk Calibration, and Argumentation Framing in Geopolitical Simulations. This paper, co-authored by Veronika Solopova, evaluates six leading AI models across four real-world crisis simulations, comparing how agents and humans navigate risk and escalation under pressure. Their finding is that models approximate human decision patterns in base simulation rounds but diverge over time, displaying distinct behavioral profiles and strategy updates.

• Superforecasting: The Art and Science of Prediction. Co-author Philip Tetlock’s insight, that forecasting is a disciplined skill built on probabilistic thinking, intellectual humility and continuous updating, is the intellectual foundation of everything that Tomas and his team are building with real-time economic modeling. Taleb’s “black swan” tells us what we cannot know; Tetlock tells us what we can. Read them together.

• Many years ago I learned how to be aware and careful about the influence of implicit assumptions in any model through studying Essence of Decision by Graham Allison and Philip Zelikow.

• Check out Accenture Research Journal, our interactive agent that lets you explore, question and navigate the insights from the hundreds of reports we publish annually. Let me know what you find most interesting. I may cover it in an upcoming edition.

The insights above are made possible by more than 300 researchers and editors across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

In this issue, we examine:

• Why we need to look beyond just jobs to the core components of work

• How to identify the skills that drive disproportionate value

• How the value of skills is shifting based on scarcity and impact

Reading the Curve

There’s a lot of discussion about how AI will change work, often focused at the level of jobs: What will be lost? What will be created? And while that conversation is important, it only captures part of what’s happening. What we’re seeing in the data is something more fundamental: The composition of work is changing.

I see this in the students I teach. They are quick to adopt AI and use it to expand what’s possible. At the same time, what matters most isn’t new. The ability to lead, exercise judgment and use creativity to solve problems remains central. If anything, these skills are becoming more important as AI becomes embedded in how work gets done.

The professional world they will enter looks different from the one that I, or my colleagues, experienced as young strategists. These students can already sense that the work ahead will be less about executing tasks and more about applying judgment, creativity and collaboration in new ways.

In fact, recent analysis conducted by my colleagues across 18 industries shows that more than 50% of working hours could be impacted by AI agents. This new “Age of co-intelligence” is reshaping how work is being performed and where value is created.

This change is happening at the level of tasks and skills.

While some tasks are becoming easier to automate or accelerate, others are becoming more valuable: judgment, domain expertise, coordination and the ability to work effectively with AI systems. As that happens, the value of skills is being redistributed in ways that aren’t immediately obvious. (We have a great illustration of that below.)

Subscribe now

This creates a new challenge for both individuals and organizations.

For individuals, it means thinking less in terms of roles and more in terms of the skills they’re building and combining over time.

For organizations, it raises a more complex question: how do you redefine work itself? How do you align talent, technology and strategy when the underlying unit of value is shifting?

How AI concentrates value at the task level

If the shift to AI were only about efficiency, the path forward would be obvious: automate tasks, reduce costs and move on. The research suggests something more complex is happening.

When organizations analyze work at a granular level, they begin to see that value is not evenly distributed across the enterprise. It concentrates in a small number of high-impact tasks and the skills that drive them.

That becomes clear when you look at the data.

Case study: Where to look for value

In one example shared by my colleagues, they modeled a large enterprise with $60 billion in annual revenue to understand where AI could create value. The results were immediate. Just three task clusters—management, business strategy and writing and recording—represented close to 40% of the total opportunity.

For leadership, the chart above shows that not all work contributes equally to value generated by agentic and other AI. Understanding where that value is actually concentrated—across specific tasks and the underlying skills required to deliver them effectively—can help organizations begin to align talent, time and AI efforts more deliberately around those areas. (See the full case study in our report, “The age of co-intelligence.”)

Acting on this is where many organizations get stuck. As AI creates extra capacity in the form of output or productivity, you can’t just cut headcount or costs and call that value, says my colleague Selen Karaca-Griffin, life sciences and products lead at Accenture Research.

“You need to essentially understand how to redistribute and reallocate these hours to bring growth,” she said.

That also requires a shift in how we think about talent.

Instead of managing roles, organizations need to understand the collection of skills each employee brings and how those skills can be recombined and deployed differently.

“Every employee becomes a skills portfolio manager,” Selen said. “But the second part of that is, individual contributors also become managers of various AI agents.”

This is what we describe as intelligent teaming: humans and AI working in tandem, continuously learning and adapting. Most organizations are still early in this transition.

If you want to make AI a source of growth—not just efficiency—start by asking:

• Where does value actually concentrate in our business, at the level of specific tasks and activities?

• How are we redeploying the capacity AI is creating to drive revenue, not just reduce costs?

• Are we reshaping how work is performed across people and agents and making it both more effective and a better experience?

• Are we evaluating talent through a skills lens or a roles lens?

The opportunity is to redesign work around where value is created and to align people and machines around those moments.

Learning Curve

Not all skills are valued equally

As AI reshapes work at the task level, pay is becoming less about job titles and more about the specific skills that drive outcomes. Some skills are scarce and highly rewarded, while others are widely available and add less value. That balance can change depending on the role or industry.

To make this more concrete, we looked at one industry—life sciences. This chart maps skills along two dimensions: how widely these skills exist in the workforce and how much the market rewards them.

Worth Your Attention

• Anthropic has released a wave of economic reports, including one that suggests AI is not yet driving widespread job losses, but may be starting to slow hiring in certain roles. They’ve also detailed how workers are using AI for specific parts of their jobs.

• A new report from Pearson shows how AI tools can help students move from passive reading to more active learning. What I find interesting here is that when used well, AI can actually deepen how students learn.

• A paper, authored by OpenAI researchers, shows how AI is beginning to perform at an expert level on tasks that directly drive GDP.

• Check out Accenture Research Journal, our new interactive agent that lets you explore, question and navigate the insights from the hundreds of reports we publish annually. Let me know what you find most interesting. I may cover it in an upcoming edition.

• A new report from Brookings highlights how AI may reshape the career pathways that enable workers to transition from low- to higher-wage work.

Overheard

“The lesson of history is that new general purpose technologies will oftentimes automate discrete tasks or aspects of our work. And in some instances, this means that jobs will go away. In other instances, this will mean that the particular bundle of things that you do in your job will shift and new types of jobs will be created.”

— Peter McCrory, Head of Economics at Anthropic, speaking last week at the Axios AI+DC Summit

The insights above are made possible by more than 300 researchers and editors across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

Reading the Curve

Last year, I wrote about what I called a “Big Bang” moment in research. For decades, the process of generating insight followed a familiar pattern: define the question, design the survey, collect responses, analyze the results. It worked well, but it moved at the pace of weeks and months.

What caught my attention then was the emergence of agent-based research—using synthetic agents to simulate how leaders, consumers or employees might respond to a question. Insights could be generated within days, at greater scale and far lower cost.

A year later, I find myself reflecting on how quickly this shift is unfolding. At Accenture Research, we now use AI simulations with agents regularly to assess behaviors and expectations of consumers and executives. Kicking off with last year’s ”Powered for Change” report, where we tested our “human + machine” methodology, adoption of these new methods is increasing even faster than many of us anticipated.

But the most interesting questions that arise from this new way of working aren’t just around speed or scale. They’re about the additional signals we now must interpret.

In early experiments, we saw how synthetic agents answered executive survey questions differently than real leaders. At first glance, that might seem like a drawback of these new tools. In talking with my colleagues, however, we have come to see it another way.

What makes this moment even more exciting is how quickly a new ecosystem is forming around these ideas. Startups are experimenting with synthetic populations to test products, pricing and messaging before they reach the market. One such company, Aaru—founded by Gen Z entrepreneurs—has harnessed this concept to help brands use synthetic simulations to better understand human behavior. (Accenture Ventures is an investor in Aaru.)

For those of us who have spent decades working at the intersection of business, economics and technology, it’s a remarkable moment to observe.

The real questions now are not just about what these tools can do, but how leaders will learn to use them. What happens when we can test assumptions before decisions are made? And what might we learn when simulated leaders think differently than we do?

Subscribe now

In this issue, we’re reading the curve to examine:

• What executives can learn from their digital twins

• Three levels of synthetic personas

• An exciting beta test!

What happens when you compare human and synthetic responses side by side?

This is one of the most interesting ideas emerging from our recent work with synthetic simulations. It’s something my colleagues Gerry Farkova, a thought leadership researcher, and Vincenzo Palermo, our data science lead at Accenture Research, have been exploring closely.

At first, you might assume the goal of synthetic agents is to replicate human answers as closely as possible. But sometimes the most interesting insight appears when the two don’t match.

In some of our early experiments, we began to see how this plays out in practice. For example, when we asked executives about their hiring outlook, many expressed optimism about expanding their workforce. Yet synthetic “executive” agents working from the same prompts forecasted contractions. In another study with an external partner, executives emphasized AI as a growth engine, while the synthetic agents pointed to cost reduction.

These differences reflect the distinct nature of the two perspectives. Synthetic agents are grounded in available data and use consistent logic. Human leaders bring experience, context and long-term optimism into their decisions.

“Machines give you rational responses, while humans are unpredictable. We are the ones who can be contradictory,” Gerry said.

Rather than treating this as a limitation, the contrast between the two can reveal signals that traditional research alone might miss.

This creates a powerful hybrid approach to research. Instead of relying on a single dataset, researchers can now compare multiple perspectives—what people say, what simulations predict and what the gap between them reveals.

A mindset shift

For business leaders right now, it’s all about efficiency: doing it fast and cheap.

“Everyone will be able to do simulations faster with synthetic agents. So, long-term, the real competitive advantage is in changing how we’re asking the questions,” Vincenzo said.

The edge will come from knowing how to question the outputs and recognize when the answers deserve closer scrutiny.

That’s where the value of the hybrid approach begins to show up in practice:

• Testing leadership assumptions: Executives may express strong support for a strategy, while a simulation suggests adoption may be slower in practice. The gap can highlight operational barriers before resources are committed.

• Understanding intent versus behavior: Consumers often say they will adopt a product or service, yet purchasing data tells a different story. Simulations trained on behavioral data can reveal this difference early.

• Revealing hidden risks: Humans bring their biases—a tendency toward optimism or toward the dominant discourse—while simulations are grounded in available data points. Examining the gap can surface risks that leadership teams may be underestimating.

For researchers, this means thinking about how insights are generated differently.

“Sometimes, you’re looking to figure out what you’re missing, for example, because of bias. Then it’s good to look for the gap and figure out what it can tell you,” Gerry said. “That’s what makes it exciting. Now I’m working with a lot more data points. I’m triangulating.”

Staying grounded in the real world

As powerful as these simulations are becoming, they also reinforce the need for real human understanding.

“Regardless of where it goes, you still need to know your own consumer rather than just looking at synthetic results, no matter how precise they are,” Vincenzo said. “Because it’s that actual connection of knowing that allows you to trust the synthetic responses.”

For leaders, synthetic simulations will make it easier to generate insights. But the key advantage will come from knowing how to interpret them: when to trust the signal, when to question it and how to combine it with real-world insight.

In uncertain environments, that kind of discernment may become one of the most valuable capabilities organizations can build.

What questions could synthetic agents help your organization answer? I’d be interested to hear how you’re thinking about this.

Learning Curve

What type of synthetic model is appropriate for different decisions?

Synthetic personas are rapidly evolving from simple narrative constructs into powerful decision systems. But not all persona approaches are created equally.

Accenture Research’s framework shows how leaders can choose the right level of simulation depending on the strategic decision at hand.

Worth Your Attention

• I’m excited to announce that the Accenture Research Journal is now live. We are testing a new interactive agent that lets you explore, question and navigate the insights from the hundreds of reports we publish annually. Be part of our research; check it out and let me know what you find most interesting. We may cover it in an upcoming edition!

• “Acceptance Is Not Enough: Toward a Psychology of Calibrated GenAI Use”: This new research article (PsyArXiv), from Nicolas Bassan, Louise Blart, Charles Ayoubi and my research colleagues Sandra Najem and Philippe Roussiere, reveals that acceptance of generative AI doesn’t necessarily lead to effective use, and that using the technology well doesn’t always lead to higher acceptance; the two (use and acceptance) rely on different skills and mindsets. For organizations, this means unlocking real value from gen AI requires deliberately developing both the right skills and mindsets.

• “2026 DHL Global Connectedness Report”: Steven Altman from NYU just published new data showing that the world is not disconnecting; its connections are evolving instead.

• “The Running Ground” by Nicholas Thompson: Nick, CEO of The Atlantic, recently published a terrific memoir on running (“the simplest of sports”). The book moved me because it connects the sport that I practice earnestly, albeit at a much more modest level than Nick, with other important parts of his life (and my own). After reading it, I even incorporated beet juice into my training regime.

Overheard

“Employers will be looking for social skills. They’ll be looking for capacity to learn new things. The ability to interact with a range of people and be effective in unfamiliar situations. They’ll also be looking for grit and determination.”

— Joseph B. Fuller, Harvard Business School Professor and founder of the Managing the Future of Work project

The insights above are made possible by more than 300 researchers and editors across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture Research

Reading the Curve

Welcome to the first edition of Intelligence Curve. I’m Francis Hintermann. I lead Accenture Research, a global team that studies how markets shift, how new technologies reshape performance and how leaders translate insight into action.

I’ve spent much of my career studying how technologies evolve—how they rise, reshape markets and enable organizations to innovate and reach a higher level of performance. The “curve” has always been central to that work: understanding where we are in an arc of change and what comes next.

Today, that arc is increasingly defined by intelligence itself. Intelligence Curve is built on the premise that intelligence—human and machine—is evolving, and that leaders need a disciplined way to interpret what is signal, what is noise and what requires action.

Technology matters. But so do markets, industry structure, geopolitics and human judgment. The future will be shaped by how organizations adapt along the curve.

Every other week, I’ll use this space to interpret the signals shaping that curve. By subscribing, you’ll also be among the first to access new research from Accenture.

Subscribe now

In this inaugural issue, we’re reading the curve to examine:

• The gap between revenue growth and economic performance

• The accelerating migration of enterprise value in the Age of AI

• The current “AI bubble” mindset

Value is migrating along the curve. Are you ready to move with it?

Revenue growth has returned across many markets. Earnings headlines look healthy. Investment in AI is accelerating. On the surface, this resembles the early phase of a typical expansion.

The underlying economics tells a more complex story.

As my colleagues Jason Angelos, Katie Dunn, Tomas Castagnino and Katarzyna Furdzik found in their research, over the past decade, roughly $27 trillion in enterprise value changed hands. What’s more, that migration has accelerated in the Age of AI—doubling in pace since 2022.

At the same time, more than half of companies continue to earn below their cost of capital despite topline recovery. Growth is visible. But durable value creation is not evenly distributed.

It would be easy to interpret this moment through extremes: that a handful of AI-native firms will capture everything, or that legacy companies are destined to fail. The data suggests something more nuanced. Value is migrating—but unevenly, across industries and regions. Strategic choices still matter.

We are seeing the emergence of “Value Leaders” that are compounding enterprise value at multiples of their peers. They’re not defined solely by being AI-native. Many are established companies that have moved early and decisively.

They tend to share several characteristics:

• They align growth with economic discipline.
  Revenue expansion is paired with returns above the cost of capital and deliberate capital allocation.

• They redesign business models, not just processes.
  AI is embedded into scalable offerings and recurring revenue structures, rather than layered onto legacy cost bases.

• They act before pressure forces them to do so.
  Portfolio reshaping, operating model simplification and capability reinvestment happen proactively, not reactively.

• They convert intelligence into advantage.
  Data, technology and human judgment are integrated into decisions that compound over time.

• What distinguishes them is not access to technology alone.
  It’s the ability to translate capability into durable economic strength.

These patterns are more than observations; they’re also signals. If they distinguish companies that compound from those that merely expand, then the real work is internal. For leadership teams, that leads to a more disciplined set of questions:

Is our growth translating into economic power? Expansion without improved returns or reinvestment capacity is fragile.

Are we redesigning the business model for durability? Future-ready structures signal long-term advantage more than incremental efficiency gains.

Are we moving early enough to reposition? Inertia is often the greatest risk in a period of accelerated value migration.

This is not a story of inevitability or collapse, but of divergence and agency. The companies pulling ahead are structurally stronger.

If you’re thinking through how these shifts are playing out in your own organization, let’s continue the conversation.

Worth Your Attention

• For multinational companies, localization matters more than ever, a Harvard Business Review article written by my colleagues Muqsit Ashraf, Tomas Castagnino and Giju Matthew, and based on research work led by Svenja Falk, shows how leading multinationals are evolving into networks of regionally embedded businesses that still share a unified global strategy. What comes next is the multi-polar company.

• What really made my TEDx talk go viral? Mariana Atencio is an exceptionally gifted Venezuelan-American journalist turned thought leader who continuously inspires my own journey as a French-Swiss citizen working and living in America.

• Where are all the agents? Manlio Carrelli of CB Insights shares data on how AI agents are reshaping the software landscape and how private markets are flooding enterprises with specialized tools. With this, the very economics of SaaS and enterprise value are starting to shift.

• “In Search of Lost Time: A Journey,” by Benedikt Fischer and Fanny Ducommun, offers a lucid examination of how modern tools reshape our experience of time. Drawing on his travels in Myanmar and Silicon Valley, Fischer, a former student of mine, contrasts daily rhythms in low-tech and high-tech societies. Technology, he observes, does more than boost productivity; it makes people impatient and changes the cadence of everyday life.

Learning Curve

Even a market shock won’t slow AI spending

We’re not in a bubble mindset. This is based on our January Pulse of Change report, that shows how recent concerns about an “AI bubble” aren’t dampening executive resolve. Nearly half of leaders say they would continue increasing AI investment even in the event of a market correction, signaling that AI is now viewed as core infrastructure—not a discretionary bet.

At the same time, only 12% of executives cite ROI as the primary driver of investment in AI. This suggests that many organizations are still investing ahead of clear value pathways, thereby raising the stakes for sharper execution in 2026.

Heard This Week

“As AI improves and can no longer be the scapegoat, you get a much clearer view of the real problem—the current state of enterprise data.”

Crawford Del Prete, Enterprise Data and AI – Unlocking The Sleeping Giant

The insights above are made possible by more than 300 researchers and editors across Accenture Research, as well as by the Accenture business leaders who sponsor and shape our agenda and by my colleagues in marketing and communications who help bring these insights to life.

Image credits: Getty Images; Accenture report: The great value migration, 2025; Accenture Pulse of Change, January 2026