Leonardo Da Vinci, Ada Lovelace, Benjamin Franklin. I’ve always wanted to have a conversation with these polymaths, but it’s pretty hard to get them on a podcast nowadays 😉But this week’s guest comes close. At 14, he built an AI-like tool so he could automate property reports for his father. He then became a computer scientist, product designer and strategy consultant.

This week on Prompted, I sat down with Tey Bannerman, a former McKinsey partner who’s spent the last decade advising some of the world’s biggest companies on how to actually use AI, not just bolt it on.

Tey’s path is a strange and brilliant one. He started as a software engineer, moved into design, then spent years at McKinsey helping organisations implement machine learning and early generative AI. He now teaches AI to design and business leaders and co-authored The AI Revolution.

He kept coming back to this idea that coding taught him the discipline of properly defining problems and being clear with what you’re trying to achieve. That kind of constraint, he argues, is exactly what’s missing now that AI can produce anything, instantly, with no friction at all. We got into what fills that gap, why he thinks trust (not capability) is the real reason AI rollouts fail inside companies, and a four-mode framework he uses for pretty much every AI task.

Before you dive in, a request: if you don’t already, please remember to hit “subscribe” and leave a rating or review on your podcast player of choice. Or better yet: share it with a friend or colleague. This helps a *ton* when it comes to helping more people discover the show and in finding great guests for you to listen to! And a reminder: my inbox is always open for guest ideas, questions and feedback.

👉 Listen to Episode 20 with Tey Bannerman

YouTube

Spotify

Apple Podcasts

Key Ideas: Constraints, The ‘Four Modes’, and the Gap Between Search and Intent

1. Constraints Still Matter, Even When AI Has None

Tey learned to code as a kid, building a database tool to help his dad generate property valuation reports. That early experience shaped how he thinks about creativity today.

He believes the old constraints haven’t disappeared, they’ve just changed shape. Deeply understanding your audience, being specific rather than vague in your inputs, and giving AI real context are the new constraints that separate distinctive work from generic output.

“The definition of a constraint may have changed with AI tools being completely boundless. But I do believe that understanding who you’re really designing for is an important constraint that means you’ll get a better outcome.”

2. Context Is the New Creative Currency

Tey doesn’t just prompt AI with a task, he feeds it context: email threads, relationship dynamics, personalities. Before starting a project, he’ll ask AI to analyse past correspondence to understand how the people involved communicate and relate to each other.

That context becomes part of every prompt going forward, producing far more considered output than a generic request ever could.

He’s also deliberate about what he leaves out. Rather than dumping everything into an AI tool, he curates inputs carefully, believing that the more control he has over what goes in, the more he can trust what comes out.

3. The Four Modes: Expansion, Compression, Reflection, Execution

After a year of informally cataloguing how people were using AI, Tey realised almost every use case falls into one of four modes: expanding your thinking, compressing information into something usable, reflecting on your own judgment, or simply executing a task.

His advice: stop starting with “which tool” or “how do I prompt this,” and start by asking which mode you’re actually in.

Tellingly, when asked which mode he personally skips most, Tey didn’t hesitate: reflection.

“It’s tough to have yourself challenged, so it’s what I’m most actively trying to get myself to do more.”

4. Trust Is the Real Adoption Problem

Tey has seen AI tools fail inside organisations again and again, and it’s rarely because the model wasn’t good enough. It’s because employees never trusted it in the first place.

Drawing on decades-old research about trust between employees and managers, he argues the same three conditions apply to AI: people need to believe the output is valuable, believe it’s working in their interest, and feel safe if it goes wrong.

The fix isn’t better marketing. It’s involving the people affected by AI in the process of designing and rolling it out, before it ever reaches their desk.

5. Systems Are Built for Precision but Humans Search in Metaphors

While working with a grocery retailer, Tey discovered that a huge number of shoppers searched for “purple vegetable” instead of aubergine or eggplant, and got zero results.

It’s a small example of a much bigger problem: most systems are built assuming people know the exact thing they want.

Real human intent is fuzzy, associative, and often searches by colour, feeling, or vague memory rather than precise language.

Closing that gap, he argues, is where a lot of the real design work in AI still needs to happen.

The Prompt Segment

Long before Tey Bannerman was advising Fortune 500 companies on AI strategy, he was a 14-year-old kid in the Pacific Islands, staying up far too late building a database tool so he wouldn’t have to keep typing the same property reports for his dad.

Using Microsoft Access and a Mail Merge trick he taught himself off the early internet, Tey automated the report his father needed, freeing up his afternoons to go play football with his mates instead. It was, in many ways, his first product.

Tey’s sketch:

Our Designer, Brody Cox, teamed up with AI to transform Tey’s request for a “neoclassical” reimagining into a full gallery show: portraits, posters, and a masterpiece titled Efficiency, depicting a 14-year-old Tey coding through the night.

Tips from Brody:

1. Be specific with prompts.

Reference style, medium, mood, era, or a specific artist/brand’s visual language rather than generic terms. “1970s Kodachrome travel poster, warm oranges, hand-lettered type” beats “vintage poster.”

2. Use AI to explain, not just generate.

Feeding a brief or reference image into a model and asking “what’s working/not working about this composition” can be a fast way to get a second opinion before you commit hours to a direction.

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: You started out as a software engineer, then moved into design, and eventually became a partner at McKinsey. How does your coding brain shape the way you think about creativity and strategy today?

Tey: “I feel so fortunate to have learned to code at a young age because I really believe that coding teaches you about constraints, and I think constraints breed creativity. When you code, you might have the most brilliant idea, but the code doesn’t care about your idea. There’s a back and forth, multiple times a day, where your ideas are being tested against reality. If I think back, it taught me a lot of humility, taught me discipline, and I think those are important things to learn, especially when you’re young.”

Cam: What does “constraints breed creativity” mean in today’s AI world, where there’s pretty much no constraints?

Tey: “I think there’s a lot of bad work out there that has been where people have used AI, but I think the best work is where people either have the constraint of, ‘here are my users,’ or ‘here’s my audience.’ I think that in itself is a constraint — really deeply understanding who you’re designing something for and ensuring that what you design is actually going to be fit for purpose. So I do think that maybe the definition of a constraint may have changed with AI tools being completely boundless, but I really do believe that understanding who you’re designing for is an important constraint that means you’ll get a better outcome.”

Cam: What does a typical Tuesday morning actually look like for an AI strategist?

Tey: “I will have a project set up, and that project has all the context around this is what I’m doing. The most valuable thing I’ve found is it also has context around the people and the relationship dynamics between them. I tend to do that by going, okay, at the beginning, here are all the emails that we’ve exchanged, and I’ll ask Claude or ChatGPT to go, ‘based on these emails, what can you draw about the personalities of each person and their relationship dynamics?’ And I have that as part of the project so that if I’m putting together a strategy, help me think through things — because I have insight about those relationship dynamics, I tend to get an output where it goes, ‘here is how you should consider this because this person is in this particular situation.’ It just helps me think far broader than I would.”

Cam: You’ve come up with a framework called the Four Modes. Can you explain it?

Tey: “It essentially came from a place where over the course of a year or so, I’d had so many conversations with people that asked, ‘how do you use AI?’ And I got to a point where I realised it’s not about the tool, it’s not about the specifics of what you’re trying to prompt, but essentially everything boils down to four modes. Are you trying to expand your thinking? Are you trying to compress, which is you have a lot of information and you’re trying to boil it down to make sense of it? Are you trying to reflect, which means are you essentially trying to question your own thinking or judgment? Or are you trying to just get something done, to execute? Every single thing on that list of tasks people were using AI tools for fell into one of those four modes.”

Cam: The purple vegetable test — what did grocery search queries teach you about the gap between messy human intent and good corporate systems?

Tey: “We were working with a grocery chain to build a new platform to help people buy groceries online, and we quickly started to realise that the way people actually really search for groceries is very different to how the original systems were architected. Most people would say aubergine or eggplant, but a lot of people go, ‘what’s the name of that vegetable again?’ So they’d just type ‘purple vegetable’ instead. But because the systems are built in a manner where it has to be an exact name match, you will not get an aubergine, you’ll not get a purple carrot. Fundamentally, it happens because the way these systems were built were not built with human intent as part of how they were designed.”

Cam: How have you seen trust become a problem in AI projects?

Tey: “Trust is probably the biggest barrier to AI adoption, and I think a lot of companies are going in thinking, ‘we just put the technology in there and people will just start to use it.’ What you need to solve for is not does the tech work, or do you have the best model, but do people actually trust what you’re saying and do people actually trust that this thing will work in their best interests. If the first thing you do in ChatGPT, you immediately get a hallucination back, it’s unlikely that you’re going to go back and keep using it.”

But the menus! It’s the detail you overlooked as you hustled to get the counter top installed for the opening. Exhausted, you don’t have any more mental energy to expend, so ... you just ask AI for it. And it gives you what it’s good at: the best generated menu that 100 different restaurants could all call their own. Something that pleases everyone and amazes no one.

That same pattern repeats when your quirky, personality-filled Halloween party setup is summed up by a flyer that looks remarkably similar to your neighbour’s. Or your business card looks just the same as that ex-colleague competing with you for customers.

AI didn’t fail any of these tasks. It succeeded at guessing what’s expected, and handed you the same thing it hands everyone else. That sameness has a name now: AI slop.

First writing — 😉 — now design

Much to my dismay, AI first killed the em-dash. I was —ing before it was cool. But now? I can’t use it. The neat, and previously underused, punctuation mark became synonymous with AI writing and immediately discredited any writing that contained it. But then we noticed more AI writing tropes. “It’s not this, it’s that.” The overuse of hyperbole. The long, breathless sentences. The adverb overuse.

The same thing’s happening in design now.

There’s a certain look and feel to AI design that’s done without imagination. There’s a lot of text-heavy work with brushstroke flourishes and on-the-nose colour scheme. You may have seen flyers and posters with crowded columns and heaps of extra information crammed along the bottom. On websites it’s the same pill-shaped buttons, sans-serif fonts, and beige and grey palette. A look that grabs your eye but you immediately forget because you’ve seen it a hundred times today already.

The Independent has been tracking how poster slop is taking over pub noticeboards across the UK, and the New Yorkers are noticing AI-generated signs and menus throughout the city. A group of academics recently looked at why so much AI content gets made in the first place, in a paper called “Why Slop Matters.” The TL;DR is that people get genuine enjoyment out of creating something low-effort, even when almost nobody else finds it interesting, but it’s a different activity entirely from producing a flyer, campaign, or pitch deck that’s actually meant to move someone else.

That distinction is crucial. Most of what’s flooding Facebook groups and LinkedIn feeds isn’t someone quietly enjoying themselves. It’s work that’s intended to find an audience, and that’s exactly where the low-effort version starts to cost you something.

Average to the MAX

Look closely at what slop actually is, and it’s oddly competent: it predicts what a thing is supposed to look like, then dials that average up to 11. A flyer becomes a wall of text because the model has learned flyers contain information, so it crams in all of it at once. A brochure for a real estate business includes images of blazer-clad professionals, thanks to decades of corporate stock photography. An AI-generated Spring Fayre poster gets an avalanche of florals (“Florals? For spring? Groundbreaking.”) Those AI slop stereotypes you see are simply the model taking the data, doing its job, pattern matching at scale, then handing you the result dressed up as a finished idea.

The moment you treat that first result as done, you’ve outsourced the one part of the work a model was never built to do: making a choice with all of your experience and context behind it. The model doesn’t respond to your prompt with a long-term plan in mind for your project’s ultimate success. No amount of prompting can replace the final choosing of what you deem worthy of sharing with your fellow humans. The first thing any model hands you, however good it is, is a guess; based on data, not a decision. Turning a guess into something worth keeping takes someone with a vision to shape it into something unique and appropriate.

The swing back to obviously-human AKA “authenticity”

I certainly see the benefits of using AI in my own work. It helps me get to 80% sooner, helps me organise disparate, loosely connected thoughts, and generally distill the noise to figure out what matters. But to counter the tide of AI slop I’m also placing a much higher quality bar on myself than I have of late, tightly editing before anything goes out the door.

For the last couple of years, AI experimentation was fuelled by speed: more drafts, more options, more content, faster, faster, faster. Quantity felt like progress, and in a narrow sense it still is ... if you already have taste and intent going in.

But now the novelty of speed is wearing off. And many people are going to extremes to prove their work is human-made.

Artists are deliberately leaning into rough, handmade, analog textures because they are unmistakably human. People are trading ChatGPT flyer callouts the way we used to trade memes (404 Media called it a ChatGPT flyer pandemic) and posts naming the pattern, like the one Emily Sundberg shared, are getting real engagement because people are hungry for a point of view, not another average.

This *re*action to AI’s *action* is shaping society and the way we communicate forever. Go back through this article and count the number of commas. There’s at least twice as many as I would have had in 2019 because I’m trying to avoid those — god — damn — dashes. Contorting your grammar so it doesn’t look like it was written by a machine. That’s what authentic looks like nowadays.

But that’s also what I love about language and writing. It’s forever changing and there is never a “perfect”. To maintain a unique voice (or, if you’re a business: a unique brand) and introduce new ideas we will always need to change, adapt, and move on.

Stop shipping slop

Look, I (like you) now use AI a lot in my daily life. And, since it’s the topic du jour here on Substack: I do use it to help formulate my ideas and add polish in my writing. But whether it’s writing, creating a deck, or doing a design, I have a few checkpoints I hold myself to. None of them require being “more artistic,” they just help me keep my brain in the loop a little longer.

• Think about the message before you open the tool. One sentence: who it’s for, what you want them to feel or do. If you can’t say it clearly, the design’s about to compensate with noise.
  

• Ask for three directions, not three versions of the same idea. If the options look like triplets, you haven’t actually explored anything.
  

• Feed it your specifics. A real photo from last year’s fete, your actual brand colours, the way you want the event to feel and what you want people to do after seeing the design. Specific inputs are the fastest route out of average outputs.
  

• Ship the edit, not the draft. A first pass is fine. Treating it as the finish line is where it all goes wrong (more on that here). First drafts are the original slop and I don’t think any self-respecting writer, designer, music producer *or* restauranteur would ship a first draft; AI or not. It’s a standard we hold ourselves to, and we need to hold AI accountable to the same editorship. Just because AI can make a flyer faster, it’s still a first draft that needs feedback and fresh thinking.

The tools that keep imagination front and centre will also make editing your AI work easy, with precise regeneration, structured layers, and movable elements instead of locked pixels you either accept whole or throw away. Yoko Li’s piece on the next frontier of visual AI being code gets at exactly this. The interesting frontier is output you can actually keep working on, not a better one-shot generation.

Slop is intelligence without imagination. Artificial intelligence without human creativity is just artificial. Maybe this needs its own “Slip Slop Slap“ moment. “Ship slop? Stop!” OK, that might be a first draft...

Thanks for reading Prompted. If this resonates, or you have your own tips to keep exercising your imagination, I’d love to hear them below.

Cameron

There is a moment in every growing company when a simple question starts taking an absurdly long time to answer.

You know the answer exists. It’s probably in a Slack thread, a doc, an old deck, or with the person who wrote it and they’re now asleep in another time zone. After 20 minutes, you start wondering whether it would be quicker to make the decision without it. But you fire off a message to someone asking to track the answer down. And wait. It’s *another* decision that got delayed 24 hours, and all of these papercuts add up to months of inertia that’s slowing your organisation down.

I wanted to talk to Arvind Jain, founder and CEO of Glean, because he’s been working to solve this pain point far longer than most, since waaaaaaay before we had MCP and apps inside of LLMs. At Rubrik (the previous company he founded) the team grew to 2,000 people in 5 years but somehow they were getting less done. People simply could not find what they needed. And this frustration is what drove him to build Glean.

All our AI tools are supposed to help with this, but an AI tool is only as useful as the context it has. If it has to search blindly across dozens of disconnected systems every time you ask a question, the answer gets slower, more expensive and often less reliable.

We talked about why context is becoming the real advantage in enterprise AI, why token costs have become an unexpectedly urgent issue, and what happens when every employee starts building their own agents, including ones that can think a little like their boss.

👉 Listen to Episode 19 with Arvind Jain

YouTube

Spotify

Apple Podcasts

Key Ideas: Context Quality, Token Costs, and Your AI Work Double

1. Context Is the New Competitive Edge

Arvind doesn’t think the underlying AI models are the differentiator anymore, since they’re increasingly interchangeable and all improving at different things.

What matters now, he says, is the quality of context you feed them: how comprehensive it is, how well it distinguishes fresh, relevant information from outdated noise, and how well it understands who’s asking and why.

“Context is another way of basically saying all the information, the data, and the know-how of how work happens inside your company,” he explains.

Businesses that invest in that context layer, rather than relying purely on whichever model is newest, are the ones building a durable advantage.

2. Token Costs Have Become the Center of Every Conversation

Just two months ago, barely anyone was talking to Arvind about token costs. Now it dominates nearly every customer conversation he has.

He’s seeing companies blow through entire annual AI budgets in under a month, largely because many AI tools take a “brute force“ approach: repeatedly querying live systems until they land on the right context.

Glean’s approach is to do that aggregation work ahead of time. Arvind says pairing Glean’s pre-organised context with a model like Claude has cut token usage by more than 50%, while making responses twice as fast.

3. Let a Thousand Agents Bloom, But Build the Guardrails

Employees inside companies are now building tens of thousands of their own AI agents, often duplicating each other’s work with no way to know what’s actually useful.

Arvind’s answer isn’t to lock innovation down. He favours a hybrid model: letting anyone experiment and build freely, while also funding dedicated, top-down teams to build vetted, department-level agents.

The unlock, he believes, is applying the same search problem Glean already solves for company knowledge to agents and skills themselves, surfacing the right one for the right person at the right time.

“Just like knowledge, you will have this proliferation of skills, you’ll have this proliferation of agents, and you will have to solve that problem again through the search problem.”

4. Everyone Might Eventually Have an AI “Work Double”

Inside Glean, employees have quietly built agents that mimic Arvind’s own thinking: one drafts messages in his voice for the sales team, another simulates how he’d react in a product review before the real one happens.

He’s cautious about building a single, fully-formed digital clone of a person, admitting it “felt odd” internally and that customers haven’t gravitated toward it either.

But he sees real potential in one specific use case: capturing the expertise of long-tenured employees before they retire, preserving institutional knowledge that would otherwise walk out the door.

5. Leaders Need to Get Their Own Hands Dirty

Arvind’s advice to leaders trying to drive AI adoption isn’t just “encourage your teams to use it.” It’s to build something yourself, first.

He’s built his own agent to help review every job offer at Glean, and he’s actively working with his chief of staff to automate parts of that role too.

“If you are not hands-on, you will just not have that instinct. You will not be able to drive the company fast enough with AI.”

For Arvind, personal, hands-on experimentation is what gives leaders the credibility and instinct to push their teams further.

The Prompt Segment

Before Arvind Jain was building AI systems used by some of the world’s biggest companies, he was a ten-year-old kid in India, watching his older brother boot up a 286 PC with a 5¼-inch floppy drive to play Pac-Man.

Video games weren’t really a thing where he grew up yet, so the moment his brother explained that these games could actually be built, not just bought, something clicked. That curiosity about what was happening inside the machine became the thread that ran through the rest of his career, all the way to a computer science degree and, decades later, to founding Glean.

Arvind’s sketch:

Our Designer-in-Residence, Matt Momtaz, teamed up with AI to turn Arvind’s childhood memory into a playable 8-bit game, styled like the Commodore 64 he grew up on, starring a miniature Arvind “gleaning signal from noise” just as he has throughout his career:

Tips from Matt:

1. Mine the guest’s story for the medium.

Arvind’s story was that his brother got him a computer in the 80s and he played Pac-Man on it. That one detail gave us everything: the game format, the Commodore 64 boot screen, the 8-bit aesthetic. When you’re creating for someone, listen for the moments that suggest a form, not just a theme.

2. Make the interaction mean something.

Instead of a straight Pac-Man clone, we tied the gameplay to Glean’s mission of finding signal in noise. Green dots are signal (worth more), gray dots are noise. The game mechanic literally is the company’s story. If you’re building something interactive, ask what the action should say, not just the visuals.

3. Layer the world-building.

The experience starts with an AI-generated video zooming into a Commodore 64 screen, then a period-accurate boot sequence typing out, then the game itself. Each layer makes the next one more believable. Small touches like CRT scanlines and 8-bit sounds made from raw audio waveforms do a lot of heavy lifting.

4. Iterate with AI like a playtester.

The first version of the game had bugs where characters wouldn’t move. Describing exactly what I saw (”it only moves a little when I press arrow keys”) got them fixed faster than guessing at technical causes. Report symptoms, let the AI find the disease.

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: You spent years building search at Google, but at some point your focus shifted from helping people navigate the internet to helping companies understand themselves. Was there a moment that inspired that transition?

Arvind: “We grew about 3X that year, and it felt like everything was falling apart. We had a lot more people in the company, but we were actually getting less work done. The biggest complaint from them was that we cannot find anything in this company. This enablement problem was a huge issue for us. Even at Google, it was incredibly hard to find a design document that you were looking for.”

Cam: What’s the difference between using something connected into ChatGPT or Claude versus using Glean?

Arvind: “These AI tools are very brute force in nature. They have to do this retrieval pass once, connect with all these systems live, and then they come back and see what came back. Usually they’re not satisfied, so they make another round, and you go into this loop of doing lots of retrieval passes. That ends up taking too much time and burns a lot of tokens. Versus with Glean, we’ve done the hard work behind the scenes already, so we can deliver the right context in one shot.”

Cam: How many conversations are you having now with enterprises about token usage and cost?

Arvind: “Nobody was talking about token costs in a big way two months back. But in the last month, that’s become the centre of every conversation that I’m having. Companies are telling us that they’d allocated a budget for AI for the year, which they’d already consumed, not even one month in.”

Cam: You’ve talked about the idea of having everyone with an AI work double. Is there an Arvind Jain work double?

Arvind: “There’s no double for me, like a video version of me. But internally there are a lot of specific agents which I keep learning about, like people are building these without telling me. There’s a ‘sound-like-Arvind’ agent that our sales team uses to craft messages they want me to send to customers and prospects.”

Cam: What’s one thing every leader should make sure their teams are doing right now?

Arvind: “You as a very senior leader, how hands-on are you yourself with AI? If you are not hands-on, you will just not have that instinct. You will not be able to drive the company fast enough with AI. So you have to actually build some automations yourself.”

Cam: And what’s one skill every young person should still learn, even with AI everywhere?

Arvind: “They should learn most of the skills that they were learning before, like how to be analytical, how to actually write well. Even though AI helps you write, that is actually a skill that’s going to be one of the most valuable ones, especially in this new world where everybody can write. You can set yourself apart by being a really good writer who’s concise and to the point.”

Welcome to Prompted, a newsletter and podcast about practical, human side of working with AI, and what it means for the future of creativity. Subscribe to receive future editions.

A couple of weeks ago, Substack CEO Chris Best published a piece called Against Claudefishing. In it, he announced a partnership with AI detection tool Pangram, giving readers the ability to scan posts and see how much of the text was likely written by hand versus generated by AI.

I applaud Chris’s belief in transparency. It’s one I happen to hold as well, particularly in the domain of visual content (such as photographs) where I think “truth” is held to a higher standard than anywhere else.

But in an art form like writing I feel like just having a tool like Pangram won’t get us there.

Here’s what Pangram promises to measure: whether AI was used to produce the text. Here’s what it cannot measure (by Best’s own admission): whether great human care went into creating it. Conflating those two things is where the promise of AI writing detection goes wrong.

A commenter in Best’s thread named Monica Hebert put it better than I could:

We (well, Monica) could just drop the mic there and move on. After all, trends by nature are transient, and online outrage oscillates to the next object with the frequency of a floor fan. But I do value transparency. And as I wrote a few months ago, sharing your workflow including the prompts you discarded, the rough layouts, and the “vibe-coding” prototypes, is the ultimate signal of authenticity. It proves there’s a human at the helm, making nuanced value judgments that a model can’t replicate. I also think that if more people in my position are honest about their process, the more useful this conversation gets.

How I actually work

When I sit down to work, I’ll sort through ideas I’ve gathered: I might read something somewhere, spot an inspiring piece of art, or overhear something when I’m travelling. All this informs the interviews I do for my podcast or the posts I do here and on social media. Sometimes, my schedule and my passions for a particular topic allow words to flow through my fingers right then and there. But more often than not I take my perspective and position, along with some sourcing I’ve already done to direct a draft: prompting with the structure, the argument, and the tone. Then I click “publish” and I’m done. (OMG, could you imagine if AI wrote well enough to do that? 🤣)

No, what it gives me is a starting point, not a finished product. That’s when I take it through the last mile, and decide what to keep and what to cut, but mostly, whether the piece is actually saying what I want it to say.

Sometimes that last mile actually becomes the first mile. The draft just isn’t working; the argument isn’t persuasive; or the narrative is all in the wrong order. But whatever path the process takes for any given piece of writing, I think it’s always in service of helping me make a better end product. Getting me started. Aiding exploration. Giving me a thought partner.

An AI detection tool might flag a draft if I’ve kept a phrase or sentence that worked well for me. But it won’t flag the thinking that shaped the brief, the revisions that changed the argument, or the accountability I carry for every word that publishes under my name. Those things don’t leave a detectable signal. They’re just all part of the work.

Why we’ve reached an inflection point

Almost everyone I talk to — founders, executives, creators — (forgive the em dashes, they’re a favourite punctuation mark of mine and I’m trying to claim them back from AI’s clutches) has insights and experience, challenges and opportunities that can be genuinely helpful to a reader facing something similar.

But the barrier between having that experience and *communicating* that experience has always held us back from getting some of the best stories that we’ve never heard. This is a principle that has been the foundation of Canva since we started: democratising access to easier content creation benefits us all. Because of that, we’ve seen design open doors to hundreds of millions of people who wouldn’t have been able to access it without Canva.

I think it’s the same with writing. Although writing skills are much more widespread than visual design skills, AI gives us the opportunity to be *better* writers, and for more people to create compelling content that others want to read.

That accessibility certainly doesn’t delegate the responsibility for providing credible arguments backed by experience and data. You don’t have to be a journalist, but taking a page from their playbook by checking facts and offering primary sourced citations for claims or data points should continue to be best practices.

And none of this means readers should switch off their critical instincts. Just as they couldn’t *before* the AI age, either.

I think it’s important to be skeptical about sourcing. You want to know whether the claims in a piece are real and whether the data is verified. AI makes it easier to generate plausible-sounding material that fails that test, which means a reader’s healthy skepticism matters more now.

But detection scores won’t get us there. They measure the origin of the writing but they’re not capable of calculating the integrity of the thinking. Publishing AI output straight from the machine is a real problem. It wastes readers’ attention and erodes the trust that makes platforms like this valuable. The answer to that problem, in my mind, is accountability for *ideas*. The individual who brings unique perspective, checks their sources, and stands behind the work will be the most compelling. I’ve always believed that quality wins in the end.

Modern writing is naturally blending the tools we have (computer, spell checker, AI) and the human who knows what they’re doing with them. The time is still spent. The craft is still required. The process just looks a little different now. And we’re all still figuring out what to call it.

Addendum: to prove a point, this post is entirely written by human hands. No AI involved; 100% free range organic. For two reasons; one: to show that I can still string a few compelling sentences together; and two: if you run this through Pangram it still says “50% AI”. So when it comes to those tools, I give up 🤷‍♂️

Cameron here 👋

I recently had one of my favourite Prompted conversations live in Los Angeles with Debbie Millman: designer, author, educator and host of Design Matters, one of the longest-running podcasts about design.

Debbie has felt like an extension of our team at Canva for a couple of years now; she’s a part of our Designer Advisory Board, a group of professional designers who advise on Canva and Affinity products. I wanted to have her on the show, and (designer to designer) go deep into whether all of the fear around what AI will do to design is warranted.

Her optimism and yearning to try out new forms of creativity comes through infectiously, not least because she’s a veteran speaker and thinker who’s spent decades interviewing some of the world’s greatest artists, designers and creatives, while building an extraordinary creative career of her own. She’s witnessed wave after wave of technological change that others heralded as the end of creativity, but which she embraced. All of that gives her a balanced view on what we’re going through right now. Rather than replacing creativity, she believes AI is forcing us to think more deeply about where creativity actually comes from, and why our judgment, perspective and lived experience have never mattered more.

Just this week, Debbie penned an opinion piece for the New York Times with the cheeky headline “If You’re Over 40, You’re Ready to Use A.I.” because it’s at that age that you’ve developed enough of a voice and unique viewpoint to critique AI’s output. In the episode, we discussed how she tries to set guardrails with her students so they learn to truly grapple with language before they outsource it to AI.

👉 Listen to Episode 18 with Debbie Millman

YouTube

Spotify

Apple Podcasts

Key Ideas — Originality, Judgment, and the Future of Creativity

1. Fear of New Technology Isn’t New

Debbie has lived through several huge technological shifts, including the rise of desktop PCs and the Internet, and sees today’s AI debate as another chapter in a familiar story.

She points out that history is full of people fearing innovation. Socrates worried writing would destroy memory. Early critics of the printing press feared it would eliminate jobs for scribes. Designers once believed computers would “take the soul away from design.”

As Debbie says, “Humans have always been afraid of change and innovation because of something that they couldn’t predict would happen.”

Instead of fearing every new tool, she believes we should stay curious about what it will make possible.

2. AI Is Best When It Expands Your Thinking

Debbie doesn’t see AI as a replacement for creativity, but as a research partner.

She now relies on AI to prepare for interviews and uncover new ideas faster, but she’s careful about where she draws the line. AI can gather information and accelerate exploration, but the insight, curiosity and human connection still have to come from the creator.

3. Original Thought Still Belongs to Humans

One of Debbie’s biggest concerns isn’t that AI will replace designers. It’s that people will stop exercising their own creativity.

She encourages her students to use AI as an equaliser or advantage, never as a substitute for thinking.

As she puts it, “Not as a replacement for creativity, not as a replacement for original thought, because that’s not where you’re going to get original thinking.”

The value isn’t simply producing an output. It’s developing the ideas behind it.

4. The Next Generation Still Needs to Learn to Make Things

Debbie worries less about jobs than about how AI is shaping young people’s creative development.

She believes students should first learn to solve problems themselves before turning to AI for help.

“I’m more worried about students’ brains not being able to craft certain things on their own before going to AI.”

She argues that making, drawing, writing and creating are how judgment is formed. And those experiences shouldn’t be skipped.

5. Consciousness Is Still the Human Advantage

Debbie isn’t worried that AI will become truly human anytime soon.

For her, creativity begins with consciousness, lived experience and perspective, things we still don’t fully understand ourselves.

AI can generate extraordinary work, but until it possesses consciousness, it remains a powerful tool rather than a creative equal.

The Prompt Segment

Prior to Debbie Millman becoming one of the world’s most influential voices in design, she was a little girl struggling to write the letter “H.”

After struggling to sign her name “Deborah,” her grandmother decided the mistake was worth keeping, and from that day on, everyone called her “Debbie.”

It’s a story that’s both endearing and revealing because throughout her career, Debbie has embraced the idea that imperfection, curiosity and unexpected detours often lead to our most meaningful creative work.

Sometimes the mistake becomes the masterpiece!

Debbie’s sketch:

Our Designer-in-Residence, Matt Momtaz, teamed up with AI to transform Debbie’s childhood story into an interactive design that reveals how one imperfect letter gave her the name she’s known by today:

Tips from Matt:

1. Start with the story, not the tool.

For the Debbie piece, the whole concept came from something she said during the recording — she couldn’t draw the letter “H” as a kid, so her grandma gave her the nickname “Debbie.” The interactive design came from that moment, not from a pre-planned template. The best AI-assisted designs start with a human story and work backwards into the medium.

2. Use AI as a creative collaborator, not just a generator.

I used Claude to help brainstorm the concept, write the code, and iterate on the interactive design — all in real-time during the podcast recording. The trick is to treat it like a back-and-forth conversation: describe what you’re feeling, not just what you want it to make.

3. Connect the interaction to the meaning.

The brush-reveal mechanic isn’t just a cool effect, it mirrors Debbie’s story of letting go of what doesn’t work to find who you really are. When you’re designing something interactive, ask yourself: “Does the way someone experiences this reinforce what it’s about?”

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: You started your career in a moment when people thought computers might really ruin design and creativity. I’m interested in how you felt at that time, and whether you see any parallels between then and what’s going on now.

Debbie: “Humans have always been afraid of change and innovation because of something that they couldn’t predict would happen. Plato and Socrates were opposed to writing because they thought it would ruin our memory. When Gutenberg first came out with the press, there were people that were so opposed to the press because it would take away jobs from scribes and monks. If we can’t predict what will happen, it triggers our feelings of unpredictability and uncertainty, which then creates a need to protect ourselves.”

Cam: Do you see what we’re going through now, in terms of artificial intelligence and other technologies that are coming to the fore, is it just another tool shift? Or is there something more fundamental here about how we think and how we do the creative act?

Debbie: “Having lived through three massive technological changes, there’s never been a time where designers didn’t think our profession was doomed. In the ’80s, the great designers were talking about how the computer was taking the soul away from design, we were told the internet was going to destroy the design business. Again, the opposite happened. I think when we don’t understand something, we associate that with something threatening. There’s an opportunity here to do things that we’ve never seen before.”

Cam: Is there a project in your career where you used one of these moments of technological shift and really embraced the technology to do something very different?

Debbie: “AI has been particularly good for me for research, and that’s where I rely on it. My ChatGPT knows Debbie Millman’s style of research, so it gives me a lot of support when I’m researching an interview for Design Matters. I don’t craft my questions that way because they don’t have the soul I want. Could this help me cut down that amount of time in a way that still keeps the integrity of my work in tact? The research did, but the questions didn’t.”

Cam: You’re an incredible writer yourself, and you’ve talked about how writing helps you explore ideas and figure things out. What do you see happening now that most of the world is going to AI for writing?

Debbie: “AI helping people write certain things can be phenomenal, especially for students who aren’t speaking English as their first language but are competing against students who do. I talk to my students about how to best utilize AI to create advantage or as an equaliser, but not as a replacement for creativity, and not as a replacement for original thought. That’s not where you’re going to get original thinking.”

Cam: You’ve said that AI doesn’t have a soul. It doesn’t have its own point of view.

Debbie: “We can’t even pinpoint the origins of our own consciousness. We can’t replicate our consciousness, that is one of the big mysteries of being alive. Where does our consciousness come from? Where does it get ignited? Where does it go? Energy isn’t destroyed in the universe, it’s just recycled. I am not worried about computers becoming sentient in the next century.”

Cam: What do you think creatives should be doing in 10 years? What should we hold onto as all this change happens?

Debbie: “To lead it. To be at the front lines, to experiment, to not be as afraid of the new tools. There’s no technology that has come out where there wasn’t some backlash to that technology. If we look at that as a pattern in our humanity and in our way of thinking, it could make us less afraid of what this tool can do and more empowered to leverage it for our creativity and to stop the bad acting.”

In this episode, I sat down with Michele Catasta, President and Head of AI at Replit, to talk about one of the biggest shifts happening in software today.

For decades, building software meant learning programming languages, frameworks, and years of engineering practice. Today, we’re entering a world where you can simply describe what you want, and AI helps build it alongside you.

If that sounds familiar, it’s because it’s the same shift Canva helped bring to design, and we’ve been doing the same with code on our own platform with Canva Code.

Michele has spent the last decade working towards this moment. From researching AI at Stanford and Google to leading the development of Replit Agent, he’s been at the forefront of turning natural language into working software.

And where this all goes? It’s somewhere that Michele believes isn’t making human creativity less important. It’s making judgment, taste and agency even *more* valuable.

👉 Listen to Episode 17 with Michele Catasta

YouTube

Spotify

Apple Podcasts

Key Ideas — Vibe Coding, Taste, and the Next Generation of Builders

1. Software is becoming a creative medium

The phrase “vibe coding” gets thrown around a lot, but Michele thinks it actually undersells what’s happening.

We’re moving beyond AI that simply writes code. Increasingly, these systems can help create products, presentations, research, workflows, marketing assets and entire businesses from a single prompt.

In that world, the limiting factor isn’t technical skill.

It’s having a clear idea of what you want to create.

2. AI raises the value of taste

One idea came up repeatedly throughout our conversation: taste.

As AI takes over more of the execution, choosing what to build becomes more important than knowing how to build every component yourself.

Michele believes the most successful people in an AI-native world won’t necessarily be the strongest programmers.

They’ll be the people with good judgment, strong product instincts and the ability to recognise quality.

3. Creativity isn’t disappearing. It’s evolving.

One of my favourite moments came when we talked about whether AI can actually be creative.

Michele argued that prompting itself has become a creative skill.

Just as artists have always drawn inspiration from previous work, AI gives people another medium for remixing ideas, combining influences and developing entirely new creative languages.

The creativity doesn’t disappear.

It evolves.

4. The future belongs to people with agency

Technical knowledge still matters, but Michele believes something else matters even more: agency.

The people thriving inside AI-native companies aren’t waiting for permission.

They’re experimenting, building prototypes, asking for feedback and improving quickly.

AI dramatically lowers the cost of trying ideas.

Which means curiosity and initiative become even bigger competitive advantages.

5. The next generation will learn faster than any before it

One of Michele’s most optimistic observations was about people entering the workforce today.

With AI accelerating prototyping and shortening feedback loops, young builders can iterate far more quickly than previous generations ever could.

Instead of spending months building something before getting feedback, they can build, test and improve every week.

That enables them to compound remarkably fast.

6. The hardest problems are still human

For all the discussion about AI agents writing code, Michele kept coming back to one point.

Companies aren’t built by software alone.

Ideas, execution, resilience, leadership and community remain deeply human challenges.

AI can remove enormous amounts of friction.

But it can’t replace vision.

The Prompt Segment

Long before Michele was building AI systems, he was a kid in Italy dreaming about California while watching Top Gun.

For him, the film wasn’t just about fighter jets. It represented ambition, high performance and the belief that you could build a bigger future than the one in front of you.

That early fascination ultimately led him to Silicon Valley and shaped the way he approaches leadership today. 😎

It also inspired this sketch:

Our Designer-in-Residence, Nellah Ocampo, teamed up with AI to transform Michele’s drawing into an animated tribute to that childhood inspiration.

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: Vibe coding has exploded over the past year. What changed to make this moment possible?

Michele: “There was an inflection point in LLM capabilities somewhere around mid-2024. I’d been waiting for that day for many years. Going from something barely working to something ready to ship took three years.”

Cam: You’ve said that AI makes taste even more valuable. What do you mean by that?

Michele: “The human taste, the understanding of what to build, how to build it, and when it should be done, should still come from the person who’s running the process.”

Cam: In an AI-native company, what makes someone successful?

Michele: “It’s a combination of agency and taste. Have the willingness to get things done. Ask for forgiveness, not permission. But don’t just create AI slop.”

Cam: Some people worry AI can only produce the average of everything it’s trained on. Can AI really be creative?

Michele: “We learned how to become creative through prompting, which is a completely new skill. We all take inspiration from what came before. The same is happening with AI.”

Cam: You often talk about the possibility of a one-person unicorn. Where do you think the real value comes from?

Michele: “The value creation is the idea, the execution, the grit. It’s not just about creating software.”

Cam: If AI can build almost anything, what should humans focus on?

Michele: “I don’t think of developers as people who manage every technical detail anymore. The real value is thinking about the whole product, the experience, and what should exist in front of the user.”

You know that moment when the AI’s output looks good enough to make your finger hover over the send button? That’s a trap. AI will easily get you to something that’s good enough. But your job isn’t to make it good enough.

AI gets you 50%, 80%, sometimes closer to 90% of the way there. The very ease of reaching that threshold is what makes it dangerous. Sure, when the palette feels considered and the assets look like they belong together, it feels done. That’s when it takes active discipline to keep editing rather than ship.

The last mile is where the magic happens

Professional expertise has never been about the ability to get a first draft in place. It starts with the spark of an idea, or the recognition that a strategy needs to shift, and comes into its own in the last mile: deciding what to cut, what to sharpen, and what to rethink entirely.

In practice, that means recognising when your eye lands in the wrong place, or the visual weight is over-optimised for symmetry rather than meaning. It means questioning whether the brand is present or just implied, whether the emotion the work is supposed to carry actually comes through, or whether it’s just ... competent. The creative direction that separates a piece from what it’s trying to do to and whether it does it, remains entirely human.

Think of AI like a high-speed train that drops you near the city. Fast. Efficient. Impressive. But the last mile? That’s you on foot, dodging scooters, reading street signs, choosing that hidden gem of an alley where you can really connect with your friends.

What happens within that last magic mile

AI excels at processing information at scale. It can analyse patterns, synthesise data, and generate outputs faster than any human. But there’s a category of judgement AI cannot touch: reading the room. The last mile requires understanding the current zeitgeist, who your audience is, what they’re talking about, and what they’ll respond to.

Research confirms the gap. A report from Cornell showed that while AI increases productivity, quality and creative value rise when people play a substantive role in shaping and refining the output.

Skip that role and the errors are rarely subtle. In design and marketing, a campaign can be both technically polished *and* emotionally tone-deaf. The kind of thing that makes people squint; then screenshot; then dunk on you in public. In finance, an AI draft can follow all the rules and still miss timing, context, and second-order effects. Speed combined with shallow understanding is how reasonable-looking decisions become very expensive ones.

The examples are already piling up: a 2025 AI-generated Coca-Cola holiday ad that misread cultural sentiment and sparked backlash; or the Harvard Business Review study that found executives who solely used AI to forecast stock prices got more confident and made worse predictions than people who just talked to peers. And then there’s the nightmare fuel: a government-commissioned report that cited fabricated sources . *That’s* a credibility problem that cannot be corrected with a revised PDF.

What makes human judgement so difficult to replicate is the capacity to understand the full context and the possible subtexts: when a line is technically correct but socially wrong, when boldness tips into recklessness, when caution becomes forgettable. It can create that delicate balance between “familiar enough” to feel grounded and “sharp enough” to feel alive. It’s taste, context, and lived experience combined. Brands are built on trust, and every piece of work either reinforces or erodes it. AI can generate content but protecting that trust remains a human responsibility.

Building a last-mile practice

To do this well, treat AI output as raw material requiring expert curation, not a finished product needing light edits. A strong last-mile discipline has four parts:

• Rapid pattern recognition: spot the signal in the noise.

• Amplification expertise: take the buried insight in an AI draft and make it the centrepiece of the work.

• Context integration: deliberately inject what AI cannot know about audience, timing, culture, and stakes.

• Quality obsession: resist the pull to ship at 80% when 100% is achievable. That last stretch is where expertise justifies itself.

AI can synthesize the past. It can’t lead the future. The campaign that starts a genuine conversation, the visual language that gets copied, the narrative that lands across cultures, all originate with people making deliberate, yet risky, creative decisions. They are what will train the next model. They can’t come from the current one.

The teams who build a real last-mile practice have the discipline to use AI well and stay in the work past the point where it looks done. When AI raises the floor for everyone, the last mile is what separates work that’s passable from work that’s sharp, timely, and trusted. Human work that matters.

Thanks for reading - we’ll be back with a new episode of Prompted next week.

Cameron

Our People team built an entire interactive session calendar for 60+ sessions across every global time zone, without writing a line of code.

A salesperson built “Cliff Notes.” No, not an avatar of our COO Cliff telling you to move faster, but an AI-powered prep and notes app that pulls Salesforce, Slack, and voice memos into one place.

These are opportunities that only came to be because of time. The time to explore. The time for human creativity to transfer into action. Time unlocked by AI.

This is what AI Discovery Week 2026 looked like at Canva.

We paused the entire company, all 5,300+ Canvanauts, for the second year running to give everyone time to build and join sessions we hosted with speakers from OpenAI, Anthropic, Google and more. On top of that, we held a two-day hackathon to close the week. By Friday, our teams had registered 467 ideas – 137 more than last year.

What people actually built

A marketer built an AR camera that lets you stand on a street corner and see your billboard at true scale before you book the media buy.

A pair of engineers built ‘Claude Code Pets’, a macOS menu bar companion that gives engineers real-time visibility into what Claude Code is doing across multiple sessions, without ever leaving their window.

A team built Readables, an adaptive reading experience inside Canva’s Learn Grid that personalises literacy support to each child’s reading proficiency, flags comprehension gaps, and gives parents a dashboard to track progress.

Now that we’ve seen this in action multiple times the pattern has become clearly visible: it’s easy to identify pain points, but it takes a combination of time to think and collaboration to build a meaningful solution. The most compelling projects came from teams that spanned disciplines (like marketers paired with engineers) and this dedicated week allowed them to connect and create together without the pressure of deadlines.

The longer tail

Last year, a hackathon team built a personalised annual memory of your time in Canva. That became the first iteration of the About Me feature in Canva AI 2.0, which we launched at Canva Create earlier this year.

A Principal Engineer used last year’s Discovery Week to build a tool for managing parallel AI software development agents. It now has hundreds of engineers using it inside Canva and has been formally adopted by our DevX team.

Our Legal team built a Slack bot at last year’s hackathon that lets any Canvanaut check whether a fictional brand name is safe to use in marketing materials. They’re lawyers, not engineers, but they still shipped it in a few days. It saved over 1,500 hours of trademark review work in its first year.

The week sends ideas downstream that often don’t reach their complete form until months later.

Show and Tells, every week

AI Discovery Week is the loudest version of something we’ve now turned into a weekly routine in Product.

Every week, we run an AI Show & Tell. People bring a new feature or process they’ve built and walk the room through it. When people see their teammates building things in minutes that would have taken weeks, the mindset shifts faster than any manifestos could manage, and gives oxygen to the best ideas that can go to our customers.

Why we do this

Pausing work across every function and time zone is not a small call, but we’ve now done it twice.

The reason is simple: our team told us they needed it, and meeting that need has translated into results we can see all across the company. Canvanauts already had access to plenty of resource libraries and Slack channels of links. What they wanted was protected time to learn together, without a deadline waiting on the other side.

We’ve been deliberate about the framing, as it’s not about asking people to use AI to do more with less. We’re asking them to use it to get closer to the work that actually matters: the creative thinking, the judgement, and the bigger and better goals they can tick off as a result.

What other companies can learn

The big AI shift has already happened, and whether you’re on the latest model release or the one before it doesn’t matter much when it comes to company-wide transformation. The question now is how you get these tools to hundreds of millions of people in ways that actually work for them, not just the ones who know how to structure a .MD file to hook up a long-polling calendar bot to an MCP server.

For other companies looking at AI Discovery Week from the outside, the lesson is probably less dramatic than it looks. We didn’t unlock anything by spending more on tools. We unlocked it by giving 5,300 people permission to stop, learn, and try something that might not work.

A lot of companies are talking about AI adoption – but this is the real, sometimes chaotic, but really fulfilling *human* work that’s required to actually make it happen.

I recorded this episode live in New York with Alex Kantrowitz, veteran tech journalist, founder of Big Technology, and one of the sharpest observers of the AI industry, during our AI Vision event in New York City.

Alex has spent years interviewing the people shaping modern technology: Sam Altman, Mark Zuckerberg, Demis Hassabis, Larry Ellison, and many more. Through his reporting, podcast, and newsletter, he’s become one of the clearest interpreters of where AI is actually heading; beyond the hype cycle.

What came through wonderfully during our chat wasn’t just his perspective on AI, but his perspective on adaptation.

Because the people and companies succeeding in moments of technological change aren’t necessarily the ones with the best predictions. They’re the ones willing to keep climbing even when the path ahead is uncertain.

👉 Listen to Episode 16 with Alex Kantrowitz

YouTube

Spotify

Apple Podcasts

Key Ideas — AI, Reinvention, and Staying in Motion

1. The biggest risk is standing still

Early in the conversation, Alex compares technological change to climbing a mountain. Fear is natural, but if fear dictates your decisions you stop moving.

That mindset shows up repeatedly in how he thinks about AI adoption. The companies getting the most value from AI aren’t treating it as something to avoid or contain. They’re creating environments where people are encouraged to experiment, explore, and build with it.

The real danger isn’t using AI imperfectly. It’s refusing to engage with it at all.

2. Reinvention is a discipline

One of the central themes in Alex’s 2020 book Always Day One is that the best companies constantly reinvent themselves. Even when what they already have is working.

That’s difficult for organisations because success creates inertia. Processes calcify. Legacy systems accumulate and teams become optimised for the current version of the business.

But the companies that endure treat every moment like “day one.” They stay adaptable. They question assumptions. And they build systems that allow people to evolve alongside the technology.

That lesson feels especially relevant right now.

3. AI changes jobs — but it also creates new ones

Alex shared an example from Amazon’s early AI automation efforts that stuck with me.

Optimisation and efficiency often lead to automation, and the traditional expectation was simple: automation replaces people.

But inside Amazon, many employees whose workflows became automated transitioned into entirely new roles: product management, systems thinking, operational design, and invention.

That doesn’t mean technological shifts are frictionless. They aren’t. But history repeatedly shows that new tools often expand human capability rather than eliminate it.

The important question becomes: how quickly can people adapt?

4. Diverse experience creates better insight

One reason Alex has such a nuanced perspective on technology is that he didn’t come up through a traditional journalism path.

Before becoming a reporter, he worked in sales, marketing, advertising, and media operations; experiences that gave him firsthand exposure to how businesses actually function.

That operational understanding shaped the way he reports on technology today. Instead of viewing companies abstractly, he understands the complexity behind building products, making decisions, and navigating constraints.

It’s a good reminder that some of the most valuable perspectives come from non-linear paths.

5. Persistence compounds

There’s a great story in this episode about Alex trying – multiple times – to get a job at BuzzFeed.

Rejected once.

Rejected twice.

Still kept going.

Eventually he sent an email with the subject line: “I want in.”

That persistence changed the trajectory of his career.

There’s a tendency to think successful careers unfold cleanly and logically in hindsight. In reality, they’re often built on repeated attempts, uncomfortable leaps, and moments where people decide not to stop after the first “no.”

6. AI adoption is ultimately cultural

The last thing Alex said that really stayed with me is that successful AI adoption isn’t just about tooling, it’s about *permission*.

The organisations moving fastest are the ones giving employees room to go deep with these systems. To experiment. To fail. To rethink workflows.

Technology shifts don’t happen because software appears.

They happen because people change how they work.

And that requires culture as much as capability.

The Prompt Segment

Long before he was interviewing the biggest names in tech, Alex was obsessed with mountaineering. Especially the mindset it takes to keep climbing when conditions get difficult.

That mentality carried into his career too: pushing through rejection, taking risks, and eventually leaving a stable job in 2020 to build Big Technology from scratch.

An early fascination that became a philosophy for navigating change.

This found its way into Alex’s sketch:

Brody Cox, Designer-in-Residence, teamed up with AI to turn Alex’s sketch into a complete brand built upon that insight into adversity.

Tips from Brody:

1. Knowing which AI platforms do what. If I have a project I want to first understand the tools I will use and what they can achieve, and how easy it is to move content from one platform to another. Having a good workflow with reliable tools is important.

2. Drawing by hand and then having AI take my sketch and transform it is something I like to do. That way I start with control by sketching, and then getting an output that looks and feels like my idea I put on paper. It is a good starting point in the process to see your ideas reflected early and see generations that actually are based on your pen to paper creation.

3. While I’m working, I still have the ability to research anything and everything with AI, so I like to look up more info on topics surrounding a project I’m working on. This helps me learn about the topic, style, content, etc. Why this is so great is that by learning more, I can get new ideas and more insight to push my project further. It allows me to think in ways I couldn’t before, so I can keep elevating the project with new ideas.

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: You didn’t come up through a traditional journalism background. How do you think that shaped your perspective?

Alex: “Spending time in industry gave me a real rooting into the way businesses work… and I think that continues to serve me even today.”

Cam: One of the key ideas in Always Day One is that great companies constantly reinvent themselves. Was there a lesson there that stuck with you?

Alex: “The best thing about being a startup on your first day is you don’t have a legacy business… all you do is build what the market needs at that moment.”

Cam: Was leaving your job in 2020 to start Big Technology a difficult leap?

Alex: “People said it was risky to leave a job in that moment. But honestly, it would’ve been more risky to stay.”

Cam: Your path into BuzzFeed is a great reminder not to take no for an answer.

Alex: “I just feel fortunate that my young twenty-something self didn’t take no… and really went for it.”

Cam: What separates the companies succeeding with AI from the ones struggling with it?

Alex: “Those that are succeeding are embracing it… and giving their employees permission to go deep into it.”

Cam: You’ve talked about mountains as a metaphor for life and work. Why does that resonate with you?

Alex: “The mountain is a symbol of what people can do when they’re determined… humanity’s ability to push through difficult things.”

I recorded this special live episode with Refik Anadol (AI artist, pioneer of “data painting,” and co-founder of Dataland) at Canva Create in Los Angeles.

What’s most compelling about Refik isn’t just the scale of his work (we’re talking billions of data points, supercomputers, and installations seen by millions). It’s how he thinks about AI itself.

He sees it not as a tool. Not as a replacement.

But as a mirror.

Because if AI is trained on the data we leave behind… what it creates is, in some sense, a reflection of us.

👉 Listen to Episode 15 with Refik Anadol

YouTube

Spotify

Apple Podcasts

Key Ideas — Memory, Machines, and Collective Imagination

1. AI as a mirror of humanity

Refik describes AI as “a mirror that reflects who we are,” and the more you think about it, the more that makes sense. These systems don’t emerge from nothing. They’re trained on human output at scale: our images, our writing, our histories, our biases.

What that means is that AI isn’t generating something from thin air, it’s synthesizing what already exists across humanity. The outputs might feel novel, but they’re grounded in collective input. That puts a lot of responsibility on us because shaping AI ultimately means shaping the data—and the values—behind it.

2. Data isn’t numbers—it’s memory

One of the biggest mindset shifts in Refik’s work is how he interprets data. Instead of seeing datasets as abstract or technical, he treats them as living archives—records of moments, behaviors, environments, and experiences.

That perspective unlocks a different kind of creativity. When data becomes memory, working with it becomes less about analysis and more about storytelling. It also reframes the role of AI—from a generator of outputs to a collaborator that helps surface, reinterpret, and visualize those memories in ways we couldn’t access before.

3. Creativity lives in the imperfections

There’s a tendency to think of AI as something that should be optimized—more accurate, more efficient, more correct. But in creative work, perfection is often the least interesting outcome.

Refik leans into the opposite: the glitches, the breakdowns, the moments where the system behaves unexpectedly. These imperfections create space for interpretation, surprise, and originality. It’s a reminder that creativity doesn’t come from eliminating error—it often comes from exploring it.

4. Human + machine is the real model

Rather than framing AI as a replacement for human creativity, Refik sees it as an amplifier. The most interesting work happens when human intuition and machine capability are combined—each doing what the other can’t.

Machines can process scale, complexity, and patterns beyond human capacity. Humans bring context, emotion, and intent. When those two sides work together, you get something neither could produce alone. That collaboration—more than automation—is where the real shift is happening.

5. From individual memory to collective dreams

Refik often talks about “collective memory,” but he takes it a step further: to the idea of “collective dreams.” The datasets we create as a society don’t just document who we’ve been—they shape what we imagine next.

In a world that often feels fragmented, that’s a powerful concept. AI can surface shared patterns, shared histories, and shared possibilities. Used thoughtfully, it becomes a tool not just for expression, but for connection, helping us find common ground in what we create.

6. Rethinking what a museum can be

All of this thinking culminates in Dataland, Refik’s attempt to build a new kind of cultural institution. Not a static space, but something dynamic that’s powered by real-time data, AI models, and sensory inputs that allow it to evolve continuously.

What’s interesting about Datalandisn’t just the technology, it’s the shift in experience. Instead of viewing art as something fixed and finished, it becomes something alive and responsive. A museum becomes less about preserving the past, and more about engaging with the present as it unfolds.

Final thought

There’s a moment early in the conversation where Refik talks about watching Blade Runner as a child and being struck by the idea of borrowed memories.

That idea stayed with him.

Now, decades later, he’s building systems that work with memory at a planetary scale.

If machines can learn from everything we’ve created they can also help us imagine what comes next.

The question is: what do we want that to be?

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: You’ve described AI as a mirror. What do you mean by that?

Refik: “If you think about the fundamentals of machine intelligence, it truly comes from the data that we left behind as humanity.”

Cam: How do you think about data in your work?

Refik: “To me, data is an invisible, poetic force… I see it as a form of memory.”

Cam: What happens when a machine learns from all that data?

Refik: “If a machine can learn, can it dream? Can it hallucinate?”

Cam: Where does creativity actually happen when you’re working with AI?

Refik: “We artists don’t like perfect systems… art happens in glitch, error, imperfections.”

Cam: Do you see AI replacing human creativity?

Refik: “I always see AI not as a replacement… but as a great friend… I hope creativity stays in the hands of humans.”

Cam: What does the ideal relationship between humans and machines look like?

Refik: “I see it as machine-human collaboration—50% machine, 50% human.”

AI adoption at work is now a given. But meaningful team-level transformation that truly reinvents how work is done and delivers organisation-level success is far less common.

In this episode of Prompted, we’re seeing no shortage of enthusiasm from leaders at some of the most forward thinking companies in tech, from Google to Anthropic.

I sat down with Sherif Mansour, Head of AI at Atlassian, to unpack why.

Sherif has spent more than 16 years at Atlassian shaping products like Jira and Confluence. He’s lived through the web era, the cloud shift, mobile, and now with AI, he’s in the position of rebuilding the apps he’s worked on for over a decade.

We dug into AI’s shift from single player (prompt > output) to multiplayer mode, where teams are effectively using agents and LLMs together in a shared workflow - and Sherif has a few ideas on what it really takes to do this well.

This conversation was recorded late last year, and in that time the use of agents for knowledge work has gone from future-state to something that’s here, now. The examples from inside Atlassian are a great starting point for anyone looking to get their teams collaborating with agents more, but aren’t sure where to start.

👉 Listen to Episode 14 with Sherif Mansour

YouTube

Spotify

Apple Podcasts

Key Ideas — Operationalizing AI

1. The shift most teams haven’t made

Sherif described a progression he sees across companies. Most teams begin by using AI individually (asking questions, drafting emails, experimenting). That can boost productivity, but it rarely changes how the organization runs.

The real shift happens when AI is embedded inside a team workflow.

Not replacing everything, but improving a specific step.

*Examples*: Before a customer meeting, Sherif can generate a briefing that pulls from Slack, proposal history, CSAT data, and internal systems. That’s not just a well-written prompt; it’s AI integrated into a defined process.

Atlassian also uses AI to analyze thousands of customer feedback inputs each week, grouping themes and suggesting next steps. Humans remain in the loop. AI handles aggregation. People handle judgment.

That’s the difference between experimentation and operational change.

2. Fix the workflow, not the use case

Many organizations search for a “killer AI use case.” Sherif sees the teams making progress by taking a simpler path.

They examine what already exists and consider:

• Where is there friction?

• Where is work repetitive?

• Where are decisions delayed?

Instead of inventing something new, they improve an existing process.

It’s less about novelty and more about clarity.

3. As AI increases output, strategy becomes higher leverage

AI allows teams to produce more, faster. More ideas, more prototypes, more feedback cycles. But increased output doesn’t create direction.

As Sherif noted, humans increasingly become the constraint, not because they can’t execute, but because they must decide.

*If strategy isn’t clear, teams simply move faster in the wrong direction.*

In that environment, leadership needs to shift from doing the work to architecting it: setting goals, defining constraints, and maintaining coherence as execution accelerates.

4. Adoption requires psychological safety & synchronised experimentation

Behavior change doesn’t happen through policy. It happens through shared experience.

Sherif shared how Atlassian ran an AI Builder Week, pausing normal work and encouraging teams to experiment together. The goal wasn’t perfection. It was learning.

Psychological safety and synchronised time made experimentation legitimate, not extracurricular. It reduced guilt.

And that’s often what turns curiosity into capability.

A Note on Talent

The next generation of builders will be AI-native. While many are worried about career prospects for junior talent, Sherif actively wants to hire them.

The Prompt Segment

At the start of the episode, Sherif shared a childhood story about collaborative comic drawing, each person adding to the narrative panel by panel.

It’s a fitting metaphor for how he sees AI: not as a replacement for human creativity, but as a collaborator in an evolving story.

Our Designer-in-Residence, Yvonne You, teamed up with AI to turn Sherif’s sketch into a short visual narrative.

🎨 Tips from Yvonne:

When prompting a video animation with Canva AI, think of it as giving clear, simple instructions so the tool understands exactly what you want.

• Start by explaining the goal and where the video will be used. For example: a school project, a birthday invite, a social post, or a product promo.

• Describe the movement, not just what appears on screen. Use everyday action words like slide in, fade out, zoom closer, pop up, or move slowly across the screen.

• Add a few style and mood words to shape the look, such as minimal, bold, playful, calm, professional, or cinematic.

• Be clear about the pace and energy. Should it feel relaxed and smooth, or fast and exciting? If your video has several moments, break it into simple steps or scenes so it flows in the right order.

• And if you’re not sure how to describe the style or camera angle, explore the options under the “Style” and “Framing” tabs in Canva AI - Video Clip. They’re a great starting point and can help you build confidence as you refine your prompt.

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: Most AI tools feel like they’re in single-player mode. Where does AI need to go to become truly collaborative?

Sherif:
“I have this framework internally around the maturity… You always hear three rough stages of the story. Stage number one is like, I type things into the AI, and I get answers to my questions. Stage number two is the aha moment that, wait. It can help do things for me, generate that media for me, give me a draft idea, help design the website for me. That’s sort of the second stage. The third stage is like when we start to deploy these, I call them virtual teammates, these new teammates with us in team workflows every day. You’re getting an AI agent and you’re deploying it in some sort of business workflow.”

Cam: Your research shows 96% of companies haven’t seen major AI transformation yet. Why is that?

Sherif:
“I think it’s pretty spot on. Companies are taking the first step where they’re saying there is a lot of individual productivity, but it’s quite hard to measure. It’s like fluffy. The ones that are getting the best impact, they’re looking at, what are my teams already doing? They’re coming at the problem from quite a different angle… What’s my business already doing and where can I improve things?”

Cam: You mentioned agents working best inside workflows. Where are you seeing that actually work?

Sherif:
“The companies that are doing amazingly well with AI and they’re deploying agents and AI already have a workflow modeled in their business… at each step of the workflow… they’re looking going, oh, I could add AI here to either reduce time or maybe increase the creativity… And that’s where we’re seeing the most leverage happen.”

Cam: As AI increases output, do humans become the bottleneck?

Sherif:
“You’re spot on. There’s more work to do. AI does an amazing job and helps us prioritize, but what’s then more important is really around directing the ship and the strategy becomes the most important high-leverage exercise. You could be building a lot of stuff, but the boat’s headed in the wrong way. The worst thing we could do is feel like we’re a bottleneck and keep scatter gunning everything without any sense of direction, and then we’ll just be going around in circles.”

Cam: How do you actually make adoption happen inside a company?

Sherif:
“What I’ve seen have the biggest impact, is we ran an AI builders week where we had a thousand people stop what they’re doing. We created some psychological safety.  It was, one, everyone sit at a synchronized time. We’re stopping so you don’t feel guilty about taking time. Two, we’re giving people space and we’re giving them space to say that the success of this week is learnings. It’s not an outcome. You’re probably going to get lots of failure. We’re going to celebrate what you learned.  People want to rethink how they could do their job better. They just haven’t had the space. And as leaders, our job is to work out which constraints we want to relax, and be okay with that to encourage that behavior and reward learnings. At the end of the week, we’d celebrate the biggest fail as a joke, but also as a celebration of the most ambitious project that didn’t work.”

For this episode, I sat down with Noam Lovinsky, Chief Product Officer at Superhuman (formerly Grammarly).

Noam has spent his career building products millions of people use every day –  from Meta to YouTube – and he has helped transform Grammarly into the AI-first company Superhuman.

Our conversation focused on a simple shift that is happening quickly ⟶ Building software is getting dramatically faster.

When the barriers to building drop, everything else changes.

Product teams get smaller.
Experimentation speeds up.
And the hardest problem becomes deciding what is actually worth building.

We talked about how product teams are evolving, how leadership needs to lead, and the future of AI and user interactions.

👉 Listen to Episode 13 with Noam Lovinsky

YouTube

Spotify

Apple

Key Takeaways

1. Product teams are condensing, lines are blurring between roles

Product teams are starting to look different.

Traditionally, a zero-to-one team might include a product manager, a designer, and several engineers. Now, you can get a lot done with just two people.

AI tools are blurring the boundaries between roles. Instead of large teams divided across specialties, we may see smaller groups of builders moving quickly together.

Design, coding, research, and prototyping can now happen much closer together. The number of ideas teams can push to production is expanding.

2. Leaders: Set aspirational north stars. Do not prescribe the path.

As building becomes faster, leadership becomes more important.

Noam believes leaders should set ambitious north stars that reflect what the new tools make possible.

One example he shared is the idea that product managers should eventually be able to push code to production themselves.

Another is a goal that any PM should be able to build an idea and get customer feedback within 24 hours.

The point is not that teams must immediately operate this way. The value comes from the progress made toward those goals.

Noam emphasized that leaders should not be overly prescriptive about how teams get there. Most builders are already motivated to learn these tools and stay relevant. Instead, leadership should set the ambition, clear away obstacles, and create the space for teams to experiment.

For his team, that approach tends to be more motivating than mandates.

3. AI Product Strategy: Moats and Castles

All companies want moats around their products. Grammarly’s biggest moat is that they are proactive and ubiquitous. They appear everywhere people already work and do not wait for users to prompt them.

This may be the future of many products. Noam argues that chat is not the interface of the future. Instead of standalone chat apps, tools may become more embedded and ubiquitous, appearing directly inside the places where work happens without needing the user to ask for them.

That said, companies may still need castles. These are places where users intentionally go to find and engage with your product.

As AI becomes more ambient and embedded across workflows, product leaders need to think about both sides of the strategy.

Build the moat so your product shows up everywhere.
And make sure you still have a castle where users know where to find you.

4. Design meetings so AI can produce the artifact

We can change how we run meetings now that AI can capture and process them.

Many meetings exist to generate something. A summary, a memo, a decision, or shared context for the team. Instead of writing that artifact later, Noam structures the meeting knowing the transcript will be used to generate it.

The meeting is recorded. Afterward, the transcript becomes the raw material. AI extracts decisions, action items, and context, and helps generate the document that comes out of the discussion.

Meetings are naturally high signal but also high noise. The goal is to pull the signal from the noise and get that context out to the wider team.

Noam is also experimenting with summarizing his meetings from a given day and posting them internally so anyone in the company can follow along if they want.

A big part of his job is creating shared context across teams. AI dramatically reduces the effort required to capture and distribute that context.

Instead of conversations living only with the people in the room, they become centralised resources the rest of the organization can quickly scan.

The Prompt Segment

At the start of the episode, Noam shared one of his earliest builder projects.

As a teenager, he created his own bulletin board system, an early online community people connected to through dial-up modems.

He called it Crash Landing, and ran it under the screen name Marvin the Martian.

A member of the community designed an ANSI art loading screen showing Marvin crashing onto the surface of Mars as the system started up.

Noam’s sketch recreates that moment.

Noam’s sketch:

Our Designer-in-Residence, Alvin Zhang, teamed up with AI to turn Noam’s sketch into something new:

🎨 Tips from Alvin:

1. Lead with vision, not prompts

   Sketch or write out what you actually want before touching any tool. AI is great at execution, but taste and intent are yours to bring.

2. Use an LLM as a translator layer

   Describe your idea casually, then ask it to rewrite as clear instructions for Canva Code (or whatever tool you’re using). It catches gaps in your brief that you’d otherwise hit mid-build.

3. Always ask for variations

   First output is rarely the best. Ask for 3–4 directions, pick what’s closest and iterate. Treat it like a back-and-forth with a collaborator, not a vending machine.

4. Keep the human in the loop

   AI can execute fast, but you’re the one deciding what’s good. Don’t outsource judgment.

Thanks again for tuning in.

Keep creating,
Cameron

Co-founder & Chief Product Officer
Canva

Interview Excerpts

Cam:
You’ve argued that chat shouldn’t be the primary interface for AI, and I’m interested in what you mean by that.

Noam:

“I think chat is now a UX primitive in the same way that a button or a dropdown is. Fundamentally, it’s not where we spend most of our time when we work. Most of the time is spent in the artifacts we create and how we collaborate around those artifacts — code, documents, presentations, design files So chat will exist in those applications, but the idea that everything should be designed chat-first is something I believe less.”

Cam:
 Are you starting to see AI actually change the structure of your teams as well? Do you need less teams? Do you need less people at your organization to build product now?

Noam:

“ I don’t know about you, but I have never been part of an organization that has a finite list of things they want to build. I always have an infinite list of things that we want to build, and I do think that we’ll just end up doing more. We’ll just do more for customers and customers... But I do think that the makeup of a team can change quite a bit. Maybe before zero-to-one team, you’d have like a PM, a designer and five engineers. Now you could probably get away with like a Design EPM, you know, an engineer or a product designer and an engineer. They can get pretty far, just like two or three of them. That’s very exciting. And I would just have more of those sorts of teams, not, not do fewer things.”

Cam:
 What are you most excited about in the world of AI and product building for the next three years?

Noam:

“The thing I’m most excited about is life extension. It’s less about discovering ways to stay alive longer, which I’m sure we’ll do, as well. It’s more about the time that I am alive, getting to use that time in ways that I want to use it more.  I’m a power user of Waymo. I use it two or three times a day. I’ve clocked, you know, it tells you, like 10,000 minutes on Waymo. I see those minutes as life extension because those are 10,000 minutes that I spent in a way that I actually wanted to spend them, rather than spending it driving.”

I recorded this episode with Rong Yan, CTO of HeyGen, just after our AI Vision event in New York.

What struck me most wasn’t just the technology he’s building around AI-generated avatars and video. It was his perspective on what actually matters now that these tools exist.

Because if anyone can generate content, code, or even a human presence on screen… what differentiates people?

Rong’s answer is simple, but not easy: it’s no longer about what you know. It’s about how quickly you can learn, adapt, and apply.

That shift has pretty deep implications—for how we build teams, how we communicate, and how we show up in a world where AI can increasingly act on our behalf.

👉 Listen to Episode 12 with Rong Yan

YouTube

Spotify

Apple Podcasts

Key Ideas — Learning Velocity, AI Companions, and Communication

1. The end of knowledge as a moat

One of the most direct things Rong said stayed with me: “I don’t believe in accumulations… A 10x person is not winning by accumulating knowledge… [they’re] the one that can learn the quickest, that can adapt the quickest, and they can leverage the tools they have at hand to make themselves better.”

For most of our careers, expertise has been built on accumulation. Years of experience, deep domain knowledge, hard-earned intuition. But in a world where tools can generate, translate, and reason alongside you, that advantage compresses quickly. What doesn’t compress is your ability to keep updating yourself.

Rong’s framing is that velocity has become the new moat. Not just learning fast once—but continuously letting go of what you knew and rebuilding on top of what’s changed.

2. AI isn’t replacing people—it’s pairing with them

There’s a lot of conversation about automation replacing human work. That’s not how Rong thinks about it.

At HeyGen, they’re exploring the idea of every employee having an AI “companion”—something that sits alongside you, learns from your work, and helps you execute.

“We always believe AI is not a replacement for humans. It’s a tool… if you use it well, you can actually improve the quality of the humans.”

What I like about this framing is that it keeps the human in the loop, but expands what they’re capable of. Not less work. Different work. Less repetition. More direction. And importantly, more leverage.

3. Communication is shifting from text to video

One of the clearest use cases Rong shared was internal communication. Instead of sending a roadmap as a spreadsheet, he generates a short video—his avatar explaining it, with visuals layered in. It takes minutes.

“I just changed it to a two minute video… my avatar speaking for the roadmap… it takes me about five minutes to generate the video.”

The interesting part isn’t the tool itself. It’s the shift in medium. Most work today is still text-based. Docs, slides, emails. But if video becomes just as easy to produce, that changes how ideas are shared and consumed inside companies.

More context. More clarity. Less friction to engage.

4. Trust becomes the real constraint

As AI-generated video gets more realistic, the technical challenge becomes less about quality—and more about trust. Rong has a simple test for whether something should be AI-generated:

“If you openly, explicitly tell them this is an AI generated video, and they don’t feel offended… pass the test.”

That’s a useful heuristic. It suggests that transparency isn’t a limitation—it’s a design principle. And that not every use case benefits equally from AI. In some cases (like influencer content), authenticity still depends on a real human presence. In others (like knowledge sharing), value matters more than origin.

5. More content isn’t the problem—better content is

There’s a common concern that AI will flood the world with content. Rong sees the opposite problem.

“It’s not about the volume of the contents… we don’t have good content… the producing cycle of traditional videos is so long… months is too long.”

If creation becomes faster, the bottleneck shifts. Not distribution. Not production. Taste. Judgment.

What’s actually worth making—and worth watching.

6. Ride the wave, don’t resist it

One of the underlying themes in this conversation is how quickly norms shift—and how little control we have over that. Rong draws a parallel to previous platform shifts, where initial resistance eventually gives way to adoption.

“When a paradigm is shifting… the only way you can deal with it is to accept it and make the best use out of it… you just need to know how to ride a wave instead of blocking the wave.”

This feels particularly relevant right now. There’s a natural instinct to question, slow down, or even reject these changes—especially when they affect how we work or how we communicate. But the people who seem to benefit most aren’t the ones debating whether the shift should happen.

They’re the ones learning how to operate inside it early.

That doesn’t mean blind adoption. It means active participation. Understanding where the technology creates leverage, where it introduces risk, and how to shape it through how you use it.

Because whether we like it or not, this wave is already here.

The Prompt Segment

Rong’s story starts with a moment that feels familiar to a lot of builders.

At nine years old, his dad brought home their first computer. It wasn’t common at the time—he estimates maybe 1% of families had one.

That moment sparked everything.

He started coding, hacked together his first game in BASIC, and stayed up until 2am building it.

A small moment. A long trajectory.

Rong’s sketch:

Frasat Farooq, Designer-in-Residence, teamed up with AI to turn Rong’s sketch into an interactive storytelling experience.

Tips from Frasat:

1. AI is an enhancer, not a mind-reader
Come in with a thorough idea. The more detail and direction you give, the better the output. Always ask the chat model (Grok, ChatGPT, etc.) clarifying questions, it forces both of you to get on the same page.

2. Iteration is everything
Whether it’s lyrics, prompts for images, or music: generate → refine → generate again. Don’t settle on the first result.

3. You are the studio
Every AI tool is just one crew member (Grok for ideas & lyrics, Suno for music, Leonardo for imagery or video, ElevenLabs for voice, Canva for final assembly). The more familiar you are with the tools, the closer you get to 100% of your vision.

4. Reference images are magic
Using Rong’s original drawing as a reference let me put that exact game on the TV screen in front of his younger self, that level of consistency and storytelling is hard to get without strong references.

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: How do you think about what differentiates people in an AI world where tools can do so much?

Rong: “I don’t believe in accumulations… A 10x person is not winning by accumulating knowledge… [they’re] the one that can learn the quickest, that can adapt the quickest, and they can leverage the tools they have at hand to make themselves better.”

Cam: Do people still need years of experience to build meaningful things?

Rong: “I feel like almost my knowledge accumulating over the last 20 years are all gone… the important thing is to really embrace the fact that velocity is really fast right now… you just cannot stop.”

Cam: There’s a lot of fear around AI replacing people. What are you actually seeing?

Rong: “We always believe AI is not a replacement for humans. It’s a tool… if you use it well, you can actually improve the quality of the humans.”

Cam: What does this shift look like inside a company day to day?

Rong: “I just changed it to a two minute video… my avatar speaking for the roadmap… it takes me about five minutes to generate the video.”

Cam: How should we think about trust in a world of AI-generated content?

Rong: “If you openly, explicitly tell them this is a AI generated video, and they don’t feel offended… pass the test.”

Cam: Are we heading toward a world of too much content?

Rong: “It’s not about the volume of the contents… it’s about the volume of good contents… we don’t have good content… the producing cycle of traditional videos is so long.”

Not long ago, creating a film meant months of planning, production, and post. You needed an extensive team, expensive equipment, and a fairly high tolerance for slow feedback loops.

Today’s guest told me it once took him three *years* to get feedback on a film.

With today’s image and video models, that same loop has collapsed. You can have an idea, prompt it, see it, and iterate on it almost instantly with strong prompt adherence and object permanence across each iteration.

That shift changes more than just speed. It changes who gets to create, what gets made, and how we think about how we define art that is the result of collaboration between man and machine.

So I wanted to talk to Alejandro Matamala Ortiz, Chief Innovation Officer at Runway, because he has been building at the centre of this shift. Runway is one of the most compelling AI companies in the creative space right now, pioneering generative models like Gen-3 Alpha that are pushing the boundaries of quality with text-to-video; producing cinematic results that were unthinkable just two years ago. In fact, last time Alejandro and I chatted actually was two years ago, and we were able to reflect on just how much has changed since then.

👉 Listen to Episode 11 with Alejandro Matamala Ortiz

YouTube

Spotify

Apple Podcasts

Key Ideas — Storytelling, Tools, and What Comes Next

1. What makes a great storyteller now

If anyone can create a film, what actually differentiates great storytellers?

Alejandro’s answer was that the quality of a story always wins, and that comes down to the questions you ask. When new tools emerge, most people use them to do the same things they were already doing, just faster. But the real breakthroughs tend to come from people who ask a different question.

Not “how do I do this more efficiently?”, but “what can I do now that I could not do before?”

He referenced early cinema. Everyone had access to the same camera, but most people used it to record reality. Then filmmakers like Georges Méliès started asking whether film could show something that did not exist.

That shift created an entirely new form of storytelling. We are in a similar position now.

The gap between average and great is not going to be how well someone can prompt a tool. It will be how they choose to use it.

2. From carving to building

One of the more interesting ways Alejandro described working with AI was this idea of carving.

In the early days, building AI products did not feel like traditional software development. You could not fully design the product upfront, because you did not know what the model was capable of yet.

So instead of building to a plan, you would explore the model, experiment with it, and gradually shape something useful out of what you discovered and anticipated might be possible.

He is not the only one thinking about it this way. Joel Lewenstein at Anthropic used a similar analogy, describing AI as more like a volcano than a skyscraper. You are not constructing something layer by layer to a predefined design. You are seeing what emerges, then figuring out how to work with it.

What is changing now is that models are becoming more capable and more predictable. Which means we are starting to move back towards building. You can begin to design products with more confidence about what the underlying models will do.

3. World models: from content to environments

We also spent time on what Alejandro sees as the next phase in video generation, which is world models. Today’s video generation has some constraints, and world models are pushing beyond them.

Instead of just generating something that looks right, they are trained to simulate how the world actually works. That includes things like motion, physics, continuity, and cause and effect. They need extensive data to learn - video, sensor data, robotics data, and other forms of real-world interaction.

The goal is not just visual accuracy, it is behavioural accuracy. You are no longer just producing something to be watched. You are creating something that can be explored. This shift will have an impact on industries beyond entertainment.

4. Place many bets and create space for them

Another theme that came up was how teams should operate in this kind of environment.

Historically, building a company meant focusing on a small number of bets. It was expensive to build, slow to test, and risky to spread yourself too thin. That constraint is loosening. When it becomes faster and cheaper to prototype, it no longer works to focus on one strong idea.

Alejandro recently stepped into the role of Chief Innovation Officer, and one of the things he is setting up is an internal incubator at Runway. A smaller team focused on exploring new applications of generative video and world models across different industries.

It also connects closely to something Allie K. Miller talked about on the show, which is the idea of small strike teams; groups that are given the freedom to move quickly, experiment, and learn without being slowed down by the rest of the organisation.

Alejandro’s framing was similar.

In a fast-moving environment, you cannot afford to wait. You need to be proactive, place multiple bets, and start exploring what is possible now.

5. Keep humans in the loop and expand what you can do

There is a lot of discussion right now about automation and replacement.

That is not how Alejandro sees it. From the beginning, Runway was not designed to be a one-click system that generates a finished output. It was designed as a tool for creators.

Something that helps you move faster, test ideas, and bring what is in your head into the world.

What we are seeing as a result is not a narrowing of roles, but an expansion.

He shared an example of a surgeon who always wanted to be a filmmaker. That was not a realistic path before. Now it is.

More broadly, people are no longer defining themselves as a single type of creative. Designers are working with video. Writers are thinking visually. Photographers are experimenting with narrative.The boundaries are starting to blur.

The people who benefit most from this shift are not the ones trying to remove themselves from the process. They are the ones staying close to it, and using these tools to extend what they are capable of doing.

6. Test your product by creating with each update

One habit Alejandro mentioned stood out because of how simple it is.

Every time Runway improves, he recreates the same short film. It gives him a consistent way to understand what has changed and how the model is working. It is a very practical way to stay grounded in the product.

For founders and builders, it is a useful reminder. If you want to understand what you are building, you need to use it in the same way your users would. And ideally, you do that repeatedly over time, so you can see the progression clearly.

The Prompt Segment

At the start of the episode, Alejandro shared a childhood memory of growing up in a home that doubled as a ceramics factory.

Shelves filled with objects at every stage of creation. Clay, glaze, finished pieces.

It gave him an early understanding that things do not just appear. They are made.

Alejandro’s sketch:

Catie Takimoto, Designer-in-Residence, teamed up with AI to turn Alejandro’s sketch into an interactive storytelling experience.

Tips from Catie:

1. See it before you make it
Before I opened a single AI tool, I envisioned what I wanted to build: an experience where you’re looking at someone’s interconnected memories and clicking into them to see what each one is holding. That overarching vision carried me through 200+ iterations to get it right. AI tools are great at execution but the vision has to come from you. The clearer your mental image before you start, the more useful it becomes when things aren’t working. And they won’t work for a while!

2. Start with pen and paper
My first move wasn’t opening an AI tool, it was picking up an old school pen and paper. A rough sketch forces you to visualize something concrete before the tools take over. It doesn’t need to be good. Going back to pen and paper is honestly one of the most underrated moves in an AI-heavy creative process! (See my poorly drawn original sketch!)

3. Use Claude to write prompts for other AI tools
Describing what you want emotionally is very different from writing a prompt that Runway or Suno will actually respond to. I’d describe the feeling I was after and Claude would translate that into the specific technical language each tool needed. You bring the vision, Claude does the prompt engineering.

Thanks for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: In a world where anyone can create anything, what actually differentiates great storytellers?

Alejandro: “I think the most important thing is that stories will always win. If you have a good story, something that people want to watch and makes sense, that will always be the key piece. But I think the most important question is how differently can I use these tools… how can I use them to create new ways of stories?”

Cam: You dropped out of film school because the feedback loop was so slow. What changed when you saw these models?

Alejandro: “I remember seeing the first models and I thought, I have an idea and I can visualise it right away. Even though it looked very bad, it was a shocking moment. It was everything that I was missing when I was trying to become a filmmaker.”

Cam: How do you think about staying ahead when things are moving so quickly?

Alejandro: “We are not playing the game of seeing what everyone else releases. We are always trying to push for what’s next… what is the next best tool that makes sense for creators.”

Cam: There is a lot of fear around job replacement. What are you actually seeing?

Alejandro: “I think what has happened in every technological transition is that it creates new markets and new economies. I am seeing people not defining themselves as a single type of creative anymore. You are not just a designer or just an illustrator. People are blending roles.”

Cam: How do you think about agents in the creative process?

Alejandro: “We have never wanted to create solutions that are one click and you get the full output back. We wanted to create tools for people who want to tell the stories they have in their heads. I think human involvement is going to remain really crucial.”

We ignored them all.

The Speed Trap

Leading with an MVP mindset is undoubtedly alluring. It enables you to put one foot in front of the other with the belief that you can find your way to success just by operationalising the collection of data.

AI has made this pressure exponentially worse because execution feels deceptively easy. What used to take a team two months to prototype can now take as little as three days (or three minutes!). But there’s a difference between throwing a million things at a wall and building a thoughtful path to the world that you envision.

In Startupland, product-market fit is critical to getting your vision off the ground, and I’d argue it’s the root cause of every type of failure in that list because you won’t be able to raise funds, generate profit, or compete with rivals if you don’t have the right product. Yet the journey to product-market fit still takes anywhere from two to six years, regardless of so many accelerating technologies at our disposal.

Today though, the bottleneck has shifted. Building your vision is not the challenge. Knowing what to build is the challenge. So it’s never been more critical for teams to understand their problem space and what the right solution should look like. No matter how fast you build, if you’re sprinting in the wrong direction you’ll never succeed.

Speed Without Direction Is Just Chaos

At Canva, we have 2,500 people on our product team who can now prototype in days what used to take months. We can test more ideas, iterate faster, and lead with more confidence than ever before. This means that our confidence in what we’re shipping has never been higher

But our confidence in what we ship doesn’t come from velocity alone. It comes from product intuition.

This distinction is where companies with strong product sense can win against competitors with more resources. Large organizations have brand recognition and the budgets to match, but they’re often slow to kill bad ideas because of organizational inertia.

A team with deep user understanding and AI-powered experimentation can test ten hypotheses in the time it takes an enterprise to get approval for one, not because smaller organizations are building faster, but because they’re deciding faster.

Speed amplifies whatever direction you choose. If that direction is wrong, you just get lost faster.

The Discipline of Rapid Testing

None of this is an argument against rapid iteration. If anything, it makes it more important.

But it has to be applied with discipline.

Every prototype should answer a specific question about user behaviour or market need. If you can’t clearly articulate what you’re trying to learn, you’re not experimenting—you’re accumulating decision debt at high speed.

At Canva, we often prototype end-to-end experiences across different parts of the product: content management, AI writing, photo editing, presentation generation; because the real experience lives in how those pieces come together. It’s how we catch misalignments early, when they’re still cheap to fix, rather than discovering them once they’ve hardened into the product.

The real benefit isn’t shipping 20x faster.

It’s making decisions that are 20x better before you commit to building.

Strategic Patience, Tactical Speed

Before celebrating how many weeks you’ve shaved off your roadmap, it’s worth pausing for a second.

Are you building from conviction about what your users truly need? Or are you outsourcing your judgment to feedback loops and hoping clarity emerges on the other side?

Speed without clarity is expensive. It doesn’t just waste cycles—it chips away at trust. And that’s a cost you don’t easily win back.

The temptation to be first to market isn’t wrong. But in a world where everyone can build quickly, it’s not much of an advantage either.

The winners will be the teams that sharpen their intuition so that speed becomes a force multiplier—not just a goal.

Taking our time at Canva wasn’t a gamble. It was a deliberate bet that understanding the problem deeply would lead to a better product.

That was true then.

It’s even more true now.

Because today, once you know what to build, you can move faster than ever.

But if you don’t—

you’re just moving faster in the wrong direction.

Cameron here 👋

I sat down with Peter Deng, a visionary product leader turned investor who has been at the forefront of shaping the major technology waves of the last two decades.

Peter founded the Messenger team at Meta, led Instagram during its breakout years, and drove major launches like ChatGPT Enterprise, voice, and memory at OpenAI. Today, he’s a General Partner at Felicis, where he invests in founders building AI-native companies.

Peter has a unique perspective on building AI companies and evaluating the market today. He shares what makes a startup worth investing in, whether the LLM race has already been decided, and where new companies still have room to win.

👉 Listen to the full episode with Peter Deng

YouTube

Apple

Spotify

Key Ideas - AI Startups

1. Old Problems, New Capabilities

Peter believes the startup paradigm has changed with AI. It’s no longer about identifying an unsolved problem. Now, the most interesting companies are revisiting issues that have been solved with old technology, and solving them better with AI.

Instead of chasing novelty, founders can look for places where people have quietly accepted inefficiency. AI has changed the constraints.

2. AI as a New Building Material

Peter compares this moment to discovering a new element in the periodic table.

The founders who succeed will not treat AI as a shortcut. They will experiment with it, understand its strengths and limits, and design around its properties.

3. Success in the Next AI Wave is About Tapping Into Human Knowledge

The race to win in the next wave of startups may have less to do with models and more to do with data.

He says, “That extraction of the chain of thought is going to be massively differentiated for startups who are building today.  The startups that can extract that knowledge out of people’s heads will really have an advantage in the next round.”

Peter’s questions for founders:

•  Do you have access to a new set of data through your product?

• Are you creating something new?

• Will you get any insight into how people think about problems and their chain of thought, as they use your product?

4. Advice for Founders Building in AI

Peter’s advice is steady and direct: Solve a real problem. Expect it to be hard. Have grit.

“Nothing worth building is easy to build.”

He also believes taste and product design will matter more as AI becomes more capable. When many products can generate similar outputs, people become more selective about what they use and trust.

Peter’s Co-creation with our Canva Designer-in-Residence and AI

During The Prompt segment at the top of the show, Peter sketched a childhood memory that perfectly captures his lifelong builder mindset.

In second grade, he and his dad took apart a remote-control car and turned it into a remote-control boat. They cut a two-litre bottle in half, repurposed the motor, built a homemade propeller, and glued everything together to make it float.

It was scrappy. It was improvised. It worked.

Peter’s sketch:

Our Designer-in-Residence, Jade Nelson, teamed up with AI to turn Peter’s drawing and story into a looping cinematic trailer titled Peter’s Infinite Machine. The short film traces his journey from tinkering as a child to building products used by billions, blending the physical and digital worlds in a way that reflects his view of AI as a new building material.

Plus, a sneak peek behind the scenes - Jade even had time to mock up a couple of movie posters based on her trailer.

🎨 Tips from Jade:

• “AI can help in crafting the best prompt for the scene you are after, but I also found that I got better results when I didn’t overcomplicate my prompt. Being clear and direct was my best chance at getting the results I was after.”

• “Create a storyboard of concept art using AI for each scene. You can then use these as reference images for your animated scenes, which allows consistency and structure throughout your animation.”

• “Try using different models to see how each interprets your prompts. Some models worked really well with my prompts and reference images, others were very clunky.”

• “Just play! Test, try, fail, try again! I learnt so much from just doing, asking the questions like “I wonder if it could do this” and using the results to inform my next step. “

Thanks again for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: Having been part of really pivotal teams inside Meta and OpenAI, what was the cultural difference in how they approach building for billions of people?

Peter:
“I’m really honored to have had this front row seat. The biggest difference can be summed up very simply in that these were different technological waves. Instagram was the mobile wave. Facebook was more of the Web 2.0 wave. And ChatGPT, you just had a different technology available to you. You had intelligence. You had AI that could talk, that could reason, that could remember, that could think. I think that’s the biggest cultural difference. You just had a different set of raw materials to build with.”

Cam: From an investor’s perspective now, what feels most interesting in this current wave of AI startups?

Peter:
“What I love about this vintage of companies is that there are so many people solving problems that have been solved before, but with old technology, and now they’re solving them with AI in a way that’s like ten, a hundred times better. The most interesting companies are coming from people who really deeply understand what AI is and can do, and are re-looking at these problems from first principles and this new lens.”

Cam: Do you think AI needs to be core to a startup now?

Peter:
“It’s got to be AI native because otherwise it’s just not interesting. I think with foundational models, the winners are probably the top four and it’s going to be really hard for others to catch up there. But I don’t think that’s where the game is being played. Having a great foundational model does not equal having a valuable product or a valuable business.”

Cam: If the advantage isn’t at the model layer, where does it come from?

Peter:
“People assume that the data in the world is in the models. Well, it’s not. A lot of it’s behind VPNs and within a company. But even then, if you had the data of every single meeting note from a company, that’s still not good enough because even then there’s still data stuck inside people’s heads. And that’s the most valuable data. The startups that are able to extract that knowledge out of people’s heads, I think will really have an advantage in the next round.”

Cam: What advice would you give founders building in this environment?

Peter:
“There’s a lot of hype out there. It’s easy to get caught up in it. But the biggest piece of advice is go out and solve a real problem. Meet a fundamental need. Nothing that is worth building is easy to build. And I think the next eighteen months, product taste and product design is going to be even more important. Every other product that’s being vibe coded looks the same, and people are going to be discerning and say, I actually want something that’s a little bit more crafted.”

Subscribe now

Cameron here 👋

For this week’s conversation, I was thrilled to sit down with Allie K. Miller, one of the most trusted AI advisors in the world and an influential AI tastemaker for the nearly nearly two million people who follow her work.

Allie has spent her career at the intersection of cutting-edge AI research and real-world business application, from building IBM Watson’s first multimodal AI system to leading Amazon’s multi-billion dollar AI startup business. She knows what works.

But the real reason she’s gained such an audience, I think, is because her excitement is palpable when she’s talking about the unexpected, deeply human impact AI can have. We got stuck in to stories like a woman with no tech experience who coded her own Mahjong app, or the importance of social cues and empathy when using AI in a public setting or meeting.

She has an intensely practical approach to upskilling and offers some amazing tactical advice in this episode, but she also had the most interesting prediction on the future of AI I’ve heard on the podcast to date!

We talk about:

• Practical frameworks and tips to try as you up-level your AI expertise, like:

  • Giving multiple models the same task and seeing which does best

  • Going “context hunting” and feeding everything back into your assistant

• Why multi-threaded brains have an advantage in the AI era

• Allie’s wildly ambitious 10-year prediction about perceiving the invisible

👉 Listen to Episode 9 with Allie K. Miller, Time 100 AI Influencer

YouTube

Spotify

Apple Podcasts

AI Action Items 📝

Try This 🚀 #1: “I’m a Blank Who’s Trying to Blank” Prompt Structure

Allie’s go-to framework for getting better outputs from any AI system (especially if you are newer to longer prompt outputs):

1. Start with context:

   • “I’m a [who you are] who’s trying to [what you want to accomplish].”

     • Example: “I’m a 50-year-old woman learning how to draw.”

2. Break into subtasks:

   • “I want to not only learn how to draw, but I want to learn specifically how to use pastels and how to draw envelopes with Sharpies.”

3. Add context about goals, fears, and background:

   • “I’m worried that I’ll never be able to draw hands. I took one class in drawing when I was eight years old.”

4. Provide resources and constraints:

   • “I can spend up to $1,000 to learn how to draw.”

   • “I want to learn in less than three years.”

   • “I don’t want to have to travel for any drawing class. I want to do it all from home.”

Try This 🚀 #2: The 45-Minute Rant Walk

1. Open Otter.ai or another transcription service and start recording (*Allie said she liked Otter.ai because you won’t lose your rant if you get a call or notification, thus avoiding lost-recording-heartbreak)

2. Go for a walk and talk through everything: tasks, problems, strategy, difficult conversations, email drafts

3. Copy the full transcript and paste it into ChatGPT, Claude, AND Gemini separately as context

4. Ask each to LLM to support the next task: draft emails, break down strategy, or help with challenges

5. Compare the outputs and take the best parts from each

You’re converting thinking time into amplified output, getting away from your computer and maximising your effort by getting multiple opinions from different models.

Try This 🚀 #3: Use Context from the Physical World

Take photos of your messy Post-its, conference slides, and whiteboard scribbles at the end of the day. Upload them to your AI tool and ask:

• “Can you make sense of what happened today?”

• “Pull out three big themes and turn it into a LinkedIn post.”

• “What’s missing from these notes?”

Everything around you is potential context. The more specific you get, the better your outputs become.

Try This 🚀 #4: ChatGPT Karaoke

Setup: Try out Allie and her friends’ ChatGPT Karaoke Game (then send us the video, obviously).

How to play:

1. Have two bowls with slips of paper. Bowl 1 has song titles. Bowl 2 has random themes (tennis balls, space travel, breakfast cereal).

2. Draw one slip from each bowl.

3. Give both to your ChatGPT Prompt:

   • “We already have a pre-done prompt and a chat thread going. So you go to it and you say, okay, I need a song about tennis balls, but to the tune of “Don’t rain on my parade”. And then the AI rewrites it and you have to perform it in front of the group.”

4. AI will rewrite the lyrics.

5. You perform it for the group.

6. Equal parts embarassment and hilarity ensue.

This makes AI social and fun, opening people’s minds to possibilities through low-stakes creativity.

Allie’s Co-creation with our Canva Designer-in-Residence and AI

During The Prompt segment at the top of the show, Allie shared her heartwarming childhood memory of launching her first business with her sister: a postal service that only ran inside their house. They called it “Lick, Stick and Send.” They hand-made envelopes, hand-drew “Sealed with a Kiss” lips on each one, and delivered mail between their parents.

Allie’s sketch:

Tisha Esquejo, Designer-in-Residence, teamed up with AI to turn Allie’s sketch and story into a vintage ad:

From Tisha: If creating videos, it’s best to generate different iterations of still images first that can be used as references when generating the video clips.

Thanks again for tuning in.

Keep creating,

Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: You’ve talked about this idea of the “productivity trap.” What do you mean by that?

Allie: “I think productivity is a trap. I do want to caveat - it is also helpful. A lot of really strong AI use cases drive productivity, but there are so many other things that AI can enable. It can enable you to do better work. It can enable you to do new work that you weren’t able to do before. What you have now is a lot of executives, maybe it’s CMOs, yelling at creative directors. And they’re overly driven by squeezing, saying, ‘Hey, we want to still produce these 10, but we want to do it with fewer people.’ And I think that sort of mindset among an executive team can be really detrimental and toxic for people below them.”

Cam: So how do you actually get people over that initial resistance to AI?

Allie: “You know, I have found that even if I sat in Madison Square Park in New York for five hours and just helped people with their AI questions, set up a booth, set up my laptop, and just sat there talking to strangers... It was a fun test to see in the real world when people are very, very hesitant toward these tools. What is that aha moment? And how do they get there? And as much as I would love to believe as a business person that it’s solving all of their business woes and showing them how Claude or Chat GPT or whatever can create the most beautiful financial analysis of their life, what ends up driving the bigger mindset shift as a first step is something fun, is something creative.

And what I usually default to is I ask them, or maybe it’s their kid, you know, what is a hobby? And so maybe it’s someone being able to say, Hey, I’m a, you know, single father. I have a five-year-old son. He’s obsessed with dinosaurs. Can you give me five ideas of things that I can cook that are dinosaur themed? That’ll be easy. And also we wanna watch Lord of the Rings while it’s happening.”

Cam: You mentioned that people with “multi-threaded brains” have an advantage. What did you mean by that?

Allie: “I meet with, you know, executives and leaders at a lot of the AI labs, and one of them told me that the really successful people with agents and skills and everything we’re talking about, they tend to have multi-threaded brains. And I was talking to friends and asking followers, what type of people have really strong multithreaded brains? And they were talking about project managers that they have to keep all of these things in their heads at one, they were talking about parents with a lot of children that you have to manage a very complex household. They were talking about people who grew up with a lot of siblings and have always had to manage five different conversations at the dinner table. Like I love thinking about what types of people have been naturally, uh, groomed for this multithreaded [work] without realizing it. I think people with like ADHD are actually at a pretty strong advantage in the AI age because we can manage all these tasks.”

Cam: What’s your wildest prediction for AI 10 years out?

Allie: “I think 10 years ago or something, there was a gentleman who was colorblind who created him, you know, he turned himself into the world’s first cyborg. Have you ever seen this? He like did some sort of brain computer interface so that a sound or a frequency he was playing depending on what color he was seeing, and he was able to retrain his brain to see things that he literally could not see from birth. I think there is a world in which with some sort of brain computer interface or new sort of AR overlay to the world that we could see things that are completely imperceptible to humans today. Like maybe you have a heat tracker and a facial recognition system, and the ability to see how solid the ground is that you’re walking on, and the ability to see the air pressure around you and the color that you’re looking at and the emotion of the people around you. Like there is a world where we go from the colors that we can see in the peripheral vision that we can see in a 3D only on earth plane to such big advancements in AI native interfaces and multimodal work and world models that are multimodal that we could perceive more than we’ve ever perceived as humans.”

Cam: Why do you think entrepreneurship feels so different now?

Allie: “Entrepreneurship is democratized. It used to be that only people that were willing or able to slave away for days to crack how to list something on Etsy or whatever it is, could figure it out. No, it is democratized. And so there’s gonna be other points of friction, but certainly getting your website up is not one anymore.”

After years of watching creative tools evolve, I’ve seen the landscape shift dramatically. The technical barriers are falling, but something more interesting is happening: the skills that matter are changing.

I recently shared an overarching vision for the year ahead in Fast Company: 2026 belongs to those who bridge a deep understanding of AI’s technical capabilities with their own discernment around what’s worth making.

With this in mind, I wanted to list out the biggest shifts I think we’ll see in AI and AI-powered creative work this year:

1. Voice mode will enter the mainstream

Voice is probably the most critical unlock that AI brings to the majority of the world who have found typing (and reading) a cumbersome burden. For those whose language isn’t accurately represented by a QWERTY keyboard, or those who haven’t had the opportunity to build literacy, speaking and listening to AI can be the unlock that gives them the keys to technology-driven opportunities.

This is something we’ve always believed at Canva: you should be able to bring an idea to life even if you’re not able to type or click your way there. Recent chats with Oliver Jay (OpenAI) and Allie K. Miller (OpenMachine) have affirmed this for me.

Voice interfaces remove the friction that’s kept billions of people from participating in creative and professional opportunities. When creation becomes as simple as speaking, we’ll see an entirely new wave of voices (literally) enter the conversation.

2. AI partners that push back will become the norm

Tools that give the right amount of friction to push your creative ideas will become increasingly sought after. Not just an obsequious yes-man (or a punishing jerk), these tools will help you as a true creative partner – one that challenges and adds to your ideas rather than just being a research assistant or a “me too.”

The best AI tools will be “yes, and...” and “what if?” The value shifts from compliance to collaboration; from tools that say yes to everything to partners that help you find better questions.

3. “Not made with AI” will coexist with “AI-generated” labels as standard practice

I think there will be plenty of public “not made with AI” statements (and even social networks) that hone in on human-made craft, but behind the scenes using AI will become the norm. Whether that’s just creative inspiration, scale, or something deeper where it’s a true collaboration between a human artist and an AI artist. We’ve already started to see this in domains like music where AI “artists” are getting signed to record labels. This will continue to push into other types of artistic domains as well, such as movie direction, visual art...perhaps even sculpture! (I’m sure some of you can point me to an example of just that.)

The core tension around what is deemed acceptable will shift from pure authorship, to the person’s intention:

• Did the creator choose speed or depth?

• Exploration or persuasion?

• Expression or optimization?

I believe that in time, AI will become morally neutral in creative culture and motivation will become the differentiator.

4. AI will make apps quieter, not louder

Despite fears of constant AI chatter – and the fact that most technical advances of the last 15 years have only contributed to notification overload – 2026 will see consumer experiences become quieter, more restrained, and less attention-seeking.

AI’s biggest UX contribution will be:

• Fewer notifications

• Fewer decisions

• Fewer interruptions

Great AI won’t talk more, it will know when to stay silent.

Also, as we more intrinsically link the data in our lives (calendars, email, messages, docs) into our AI assistants, AI EAs (executive assistants) and AI PAs (personal assistants) will start to become the norm. Helping us to streamline our lives and remove the humdrum necessities of things like meeting scheduling, travel planning, bookings, responding to communications (particularly messaging, since email has already seen some movement over the last couple of years).

5. The full spectrum of AI-human collaboration will emerge

We’ll start to see the full spectrum of artistic expression: AI-only, human-only, AI + human. AI + human attributed. AI + human unattributed (just part of doing business). All of these modalities have valid expression, will create different types of art and will add different types of value.

Just as TV displaced radio, but created space for radio to find its true place of value, AI content will displace current (particularly social) content, but define a valuable space for human-made content.

Getting to the right solution is still the key skill for human creators. AI offers a world of possibility and the ability to express yourself in entirely new ways, but pressing the button to send something to someone is still the key skill: what do you send, when do you send it, who do you send it to. Those decisions shape your brand, your identity, and your message.

6. New roles will design human-AI workflows

Building on the prediction I made last year that a new role for those who specialize in managing automated agents to accomplish tasks and projects will emerge, I believe companies will begin hiring people whose job is to design how work flows between humans and AI.

They’ll answer questions like:

• Which decisions should stay human?

• Where does automation create fragility?

• How do we prevent cognitive overload?

The single most important tensions leaders must manage: What is the core value that people bring to a process and an organization? How do you maximize that human-ness while benefiting from AI?

In software development, I think rather than the traditional division of labor that has previously surrounded designing/PMing/engineering, those areas are already starting to overlap and what your work looks like will be more about what problem domain you are attracted to. Designers might focus on more creative/experiential problems and opportunities, engineers on more functional and logical domains, PMs might focus on workflow domains. But each type of person will be capable of delivering complete solutions within those domains and iterating even more rapidly.

7. Judgment, taste, and intention become the most valuable skills

In 2026, AI doesn’t make work or creativity easier; it makes judgment, taste, and intention the most valuable skills we have. Bringing those to your team in an AI-enabled way will be how the most successful people approach their jobs.

The most in-demand professionals will be those who can:

• Reframe messy questions

• Challenge false assumptions

• Decide what not to optimize

• Find the biggest way to have an impact

Anyone who bills hourly is ripe for disruption. New ways of framing value (and cost) to clients will be required as AI supercharges people’s ability to execute.

These are just some of the ways I see AI reshaping creativity this year. The next 12 months will be defined not by what AI can do, but by how thoughtfully we choose to use it, and what we decide is worth making in the first place.

Cameron here 👋

At Canva we think deeply about our ways of working, determining if anything needs to change so that we can have a bigger impact at the product, team and individual level. That’s why this week’s conversation with Natalie Piucco, Field CTO at Google Cloud, really got me energised.

One of the biggest things I took from this conversation is the mental model shift every person on earth needs to navigate right now. Things we’ve taken for granted as ‘impossible’ now may actually be achievable with the help of AI.

This feels especially relevant in these first few weeks of 2026, where Claude Cowork has shown us that more than ever, non-technical professionals can build their own software solutions. Real human-agent collaboration is much closer than it seemed a few months ago, and we should approach every task with the question: “Is there a better way to do this?”

Natalie has had a front row seat to the entire AI wave while at Google. For example, the world is still figuring out AI wearables; Natalie was testing them back in 2017. She has insightful perspectives on AI applications in areas we haven’t covered much yet on Prompted, like higher education, medicine and virtual reality. This conversation was not only about the uses of AI, though - Natalie walked us through the exact frameworks she and her team have used to build AI products over the past few years.

There is so much in this week’s episode that any team leader or product builder can immediately apply to your own ways of working. I’m excited to hear what you think!

Episode 8: The Art of the Possible in an AI World
🎧 Prompted: AI, People, and the Creative Spark

We talk about:

• How AI is changing higher education.

• Embracing AI as a ‘permission slip’ to do things differently.

• Why the future belongs to high-agency creative generalists.

• Natalie’s learnings from seven years of building and shipping AI products at Google.

👉 Listen to Episode 8 with Natalie Piucco, Field CTO, Google Cloud

YouTube

Apple

Spotify

Hit reply: What’s one place in your work where waiting to feel “ready” might be slowing progress?

Frameworks Natalie Shared

1. The Google Moonshot Creativity Framework

Natalie outlined a three-part “recipe” used inside Google’s Moonshot Factory for thinking creatively. Look for the intersection of:

• A huge problem worth solving (one that affects millions of people).

• A radical, sci-fi-level solution (aiming for a 10x improvement, not 10%).

• A breakthrough technology (something newly possible that makes the solution viable).

In the overlap of those three categories is the creative solution.

2. “Price the Problem” Before You Build

Instead of starting with tools, start with repeated problems and assign each a cost.

Every problem carries a price:

• Time

• Talent

• Money

• Customers

The higher the cost, the higher the priority, with one important twist. Sometimes solving a smaller problem creates the most momentum and unlocks belief faster than tackling the biggest one.

3. Redefining Progress (The Progress Principle)

Natalie referenced research from Harvard Business School showing that the strongest driver of motivation at work is the feeling of making progress on something meaningful. Not necessarily big breakthroughs, but forward motion.

She applied this to creative and AI-driven work, where progress often feels slow if it’s measured only by outcomes. Instead, she suggested redefining progress to include:

• Testing an idea.

• Killing an idea.

• Learning from feedback.

Momentum comes from banking small wins, not waiting for breakthroughs.

AI Action Item 📝

Try this 🚀

To do this, I created a GPT that takes screenshots of my calendar for the day, as well as screenshots of my browser history (documents I’ve opened, etc.). It then spits out a chronological list of all the things I worked on.

Here is my prompt for the GPT (which was generated by ChatGPT after a back-and-forth conversation the first time I tried this workflow):

You generate concise daily work logs from screenshots of a user’s calendar and browser history. You analyze one or more screenshots of a calendar day view and Chrome browser history for the same day, synthesize them, and produce a single bulleted list of activities in chronological order from earliest to latest. Treat the calendar as planned work and browser history as evidence of actual execution; prioritize what was actually done. Ignore calendar blocks titled exactly or effectively as Cycling or Lunch (any variant). De-duplicate aggressively: merge the same work appearing across sources, repeated opens of the same document, deck, video, or tool into one item; treat logins, Zoom joins, Canva home/profile, and access requests as context rather than activities. Prefer visible timestamps; otherwise infer order using browser history order, typical workflow, and calendar structure. Use short, concrete, note-to-self phrasing with minimal filler verbs. Output only a bulleted list with no preamble or commentary unless explicitly requested. Make reasonable inferences for ambiguous items rather than asking follow-up questions. When a browser item maps clearly to a known project, name the project rather than raw tab titles. Maintain accuracy, brevity, and a human tone.

Natalie’s Co-creation with our Canva Designer-In-Residence and AI
During The Prompt segment at the top of the show, Natalie shared a childhood memory of working on her family’s banana farm. She sketched the banana shed, her family eating pasta together, and the banana boxes her great-grandmother designed using a reusable template.

Harriet Van Tongeren, our Designer-in-Residence, teamed up with AI to turn Natalie’s sketch and story into something new:

A few ideas from Harriet to try:

• Start with a whiteboard brain dump, including written notes on the story or direction.

• Decide what vibe you’re going for (dark moody, futuristic, etc.) and head to your inspirational sites to gain some style references (try things that aren’t Pinterest so you’re not in a popular trend trap).

• Feed inspo to an LLM to help write a prompt. This can make it more detailed and add metrics you might not have thought of.

• Run a few static graphic tests, generating images that support the story and fit the style you’re going for.

• Continue iterating, being super specific, and creating a few versions of the static scene, then move into animations

Thanks again for tuning in.

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Interview Excerpts

Cam: So what’s your approach for helping leaders and organizations foster rapid learning with new technology?

Natalie:
“Everything we know about business and technology is being rewritten. A lot of businesses are faced with this question of how they possibly keep up when the stakes keep getting higher. There is no book you can give leaders to navigate this pace of change.

If you are leading a business through change, at minimum you really have to upgrade your organization’s learning operating system. Some businesses give everyone baseline AI literacy and a platform to experiment, which is important. But one of the less obvious ways to foster rapid learning is for leaders to create space for teams to start without having all the answers.

Progress comes from beginning, not from certainty. You cannot wait to feel ready. Readiness is not a feeling. It is a decision.”

Cam: Google studied its highest-performing teams. What stood out?

Natalie:
“Google did a lot of research on what made our highest-performing teams successful. Of everything we looked at, the number one trait was psychological safety.

If you have a group of people who feel safe to admit they do not know, they are far more likely to learn together. That applies not just to the team you have today, but also to the people you hire and bring in over time.

Without psychological safety, people do not ask questions. And if people are not asking questions, learning slows down.”

Cam: You talk about the ‘art of the possible.’ What do you mean by that?

Natalie:
“What I am seeing right now is that everyone has almost been given a permission slip to do something different. Ideas that were once too complex, too time-consuming, or just impossible to execute now feel achievable.

Before, you might have had an idea but not the computing power, the tools, or the time. Now people can go to their leaders and say, imagine if we did this, imagine where we could go. And they are rewiring how they think about every part of their business.

That is what I mean by the art of the possible.”

Cam: What’s one thing leaders misunderstand about creativity?

Natalie:
“I think the biggest mistake businesses make is thinking creativity is always forward motion. It is not. It is messy. You cannot put it neatly in a spreadsheet.

To be creative, you have to stay in that uncomfortable window of not knowing. A lot of businesses are not prepared to sit there. They want certainty too quickly.

The only way to get measurable impact from creativity is to redefine what progress means. Testing an idea is progress. Killing an idea is progress. Learning what does not work is progress.”

Cameron here 👋

When we launched Canva from Australia in 2013, global expansion and product localisation weren’t just a way to tap into more markets, but fundamental to fulfilling our mission of empowering the *world* to design. So after we’d found product-market fit we prioritised localisation by launching in Spanish in 2016, followed by rapid expansion into over 100 languages.

Today, that playbook looks totally different at a company like OpenAI. Large language models don’t “launch” into new markets so much as they are discovered there—often before a company has teams, policies, or infrastructure on the ground. ChatGPT was used globally from day one, with adoption accelerating faster than traditional SaaS companies ever experienced.

As a result, scaling AI internationally is less about unlocking new markets and more about catching up to reality. Localisation at this scale now includes understanding cultural norms, education systems, media environments, and political contexts, as well as recognising how AI can create very different risks and opportunities in different parts of the world.

Off the back of OpenAI’s new office opening in Australia, we invited Oliver Jay, Managing Director of OpenAI International to join us for our first-ever AI Vision event in Sydney. While in town, he came on Prompted to go even deeper into what this all means not just for OpenAI, but the people, governments and companies who use their products.

Episode 7: AI at Global Scale and What It Takes to Make AI Work for Everyone

🎧 Prompted: AI, People, and the Creative Spark

In this episode, we talk about:

• Why AI adoption follows familiar product-led growth principles, but places increased priority on rapidly localizing to different markets

• How OpenAI approaches localisation, from language to cultural context

• Why voice could be a major unlock for narrowing the digital divide

Oliver has spent his career scaling products and is now thinking about scale at a country and global level. This conversation is a reminder that making AI useful is not a technical challenge...it’s a human one.

What stood out to me was how practical Oliver and his team are about embracing this opportunity to scale rapidly in a pragmatic and responsible way.

He explains that the familiar chat interface and immediate utility offer speed of growth, but this isn’t enough on its own. Working directly with governments, rethinking education models, and tapping local experts to localise ChatGPT on a country level are a few of the more intentional steps OpenAI is taking to ensure the whole world can benefit from AI long-term.

It was also reassuring to hear that (as a part of OpenAI, but also for the sake of the entire tech ecosystem) Oliver enthusiastically reiterated the importance of partnerships – from developers to governments – in building an AI-first future. Not only is the opportunity shared, but the responsibility is, too; even for companies who aren’t building models themselves.

👉 Listen to Episode 7 with Oliver Jay, Managing Director, OpenAI International

YouTube

Spotify

Apple

Hit reply - What is a cultural nuance you wish your AI tools understood better?

AI Action Items 📝

Here are the three big ideas I’m taking from my conversation with Oliver Jay.

1. AI is universal, but local

ChatGPT has scaled to 800M+ users globally, and unlike other products, its growth happened everywhere from day one rather than market by market. At the same time, Oliver highlighted how usage still varies significantly by country. One of my favorite examples was how people in India were using AI to learn, which directly led to the creation of Study Mode and then scaled back out to the rest of the world. That is a powerful model for product development.

Try this 🚀

• Pay attention to how people in different regions, teams, or roles are already using AI, especially in ways you did not anticipate.

• Curate a panel of local experts to help you navigate cultural nuances.

• Treat local behavior as an input to design, not something to smooth over.

2. Upskilling is the real multiplier

OpenAI knows that to succeed they have to work on upskilling talent. Between OpenAI’s academy certifications, and the upcoming jobs platform, Oliver made it clear that they see AI readiness as being about having the right tools and resources as much as it is about raw experimentation. This is a useful signal for any organization adopting AI. The companies that invest in frameworks for learning may be able to scale their AI use more sustainably.

Try this 🚀

• Create a clear path for people to become more AI fluent, regardless of function.

• Make learning visible and shared across teams, not siloed with technical experts.

• Treat AI capability as a core organizational investment, not an optional skill.

3. Use AI to make work more human, not less

One idea that really stuck with me is that AI helps you get to the hard parts faster, but it cannot replace empathy, creativity, or leadership. AI gives people time back, and what matters is how we choose to spend that time. As leaders, our responsibility is to reinvest those gains into the work only humans can do.

Try this 🚀

• Audit where AI is saving time today and explicitly reinvest that time into human work.

• For leaders, block time for coaching, development, and strategic thinking that used to get deprioritized.

• Ask regularly: what is the work only humans should be doing here?

Keep creating,
Cameron
Co-founder & Chief Product Officer, Canva

Subscribe now

Interview Excerpt
Cam: You have worked on product-led growth at companies like Dropbox and Asana. How does the adoption curve of ChatGPT compare?

Oliver:
“I feel very blessed to have had a chance to be on a few rocket ships now. When I think about my experience at OpenAI so far, I would say that a lot of the principles of how you scale a product-led growth company stay consistent, but the tactics have definitely changed.

When I think about the adoption curve, it is similar, but the slope of that curve is a lot steeper because AI is this phenomenon that has been adopted so quickly in so many places, in all facets of society. The principles do not change, but the tactics definitely need to change.”

Cam: Why has OpenAI taken such a different approach to international expansion compared to other companies you have worked at?

Oliver:
“This time last year, we had a small office in London and one person in Tokyo, but at the same time AI was being adopted everywhere. So we had to think about doing things very differently. Right now, we are opening six offices at the same time in parallel.

AI is happening everywhere. That means we cannot do things sequentially. We have to do many things in parallel, whether that is improving the model, improving the product, or being closer to our users and customers. The principles stay the same, but the way you execute has to change.”

Cam: There is a real risk that AI could widen the digital divide. How do you think about access, especially in regions with lower connectivity?

Oliver:
“I think about this in a top-down and bottom-up way. Top-down, this is the purpose of OpenAI for Countries. We need to work side by side with governments to make sure countries are ready for what is coming and what is already here. Upskilling is a global need, and that responsibility is shared between governments, companies, and OpenAI.

Bottom-up, I am personally most excited about voice. Most of our digital experiences so far have been through a keyboard, and without that keyboard you do not get access to a lot of digital benefits. Voice changes everything. It opens up access in a way that can actually help narrow the digital divide, not widen it.”

Cam: AI gives people time back. What should leaders and teams do with that time?

Oliver:
“The most concrete benefit AI gives you is time and space. It takes away the less joy-giving work. For teachers, that means less administration and more time with students. Teaching is not just injecting content. It is helping students gain confidence and reach their potential.

For me as a leader, AI helps orchestrate work more efficiently so I can spend more time coaching, developing people, and helping leaders see around corners. Those are the intangible things that often get deprioritized, but they matter the most.”