Canva is one percent of the way there

I spent three years inside the most ambitious company I’ve ever seen, and the way it works is a preview of how every company will soon have to work.

The phrase you hear most inside Canva is we’re one percent of the way there. Said casually and meant completely, it is also the largest thing I’ve ever heard a company say about itself. Mel Perkins was saying it when the company was a few hundred people. She was still saying it at five thousand people and a quarter of a billion users.

My three years there began in March 2023. I initially led Discovery, incubating much of what became Magic Studio, then ran the generative AI group that shipped Canva AI, and finally launched Canva AI 2.0 as product lead in April 2026. I left the following month. Three roles and three org shapes in three years, all pointed at the same problem: how fast good judgment can reach the work. When the person with the clearest picture of the product decides something, how long until the thing being built reflects it? I’ll come back to that, because it’s about to be everyone’s problem.

It should not have worked

None of it makes sense without the origin story, which you’ve probably read: Perth, the most isolated city in the world; more than a hundred investor rejections; Mel learning to kitesurf to get in front of an investor. Two things came out of it. Profitability, because when capital is hard to get you build something that doesn’t need it: Canva has been profitable every year since 2017, and it once spent two years rewriting its entire codebase while shipping almost no new features, which you can only do if nobody is running out of runway. And a habit of deriving things from scratch, half a world away from whatever San Francisco considers normal.

People at Canva are cheerful in a way that might take an American a while to trust, and underneath the cheerfulness is the hardest-working group I’ve been around. I started calling it laid-back intensity, and I still don’t have a better phrase. Nobody is stressed for show. Everybody is moving extremely fast.

“Canvas Chef” becomes Canva AI

Canva is the only place I’ve worked where a document written before the company existed still functions as a spec. When thousands of people can predict what the founder would say, they stop needing to ask, and she gets to choose where to be involved rather than being the bottleneck everywhere.

In 2011, before there was a company, Mel made a deck for something called Canvas Chef. The pitch was that you could describe an idea and have it appear, designed, without knowing anything about design. Fifteen years later she stood on stage at Canva Create and quoted her own deck back to the room while launching an AI that does exactly that.

From the original 2011 Canvas Chef deck, via NBT’s archive of the pitch.

A spec like that never lets you feel finished. We shipped Canva AI, and within months the conversation was already about an agent that does the whole job, which is roughly what we announced a year later as Canva AI 2.0. The target never stood still, because the ambition is genuine. If the objective is to empower the world to design, then the world is large and design is most things, and there is no version of the job where you’re done. Mel holds ambition at that size without ever letting it curdle into fantasy. I haven’t seen anyone else manage it.

The blank design

Mel is in all the details, and so is her cofounder Cliff Obrecht. Paul Graham named the pattern founder mode in 2024, eleven years after Canva started running it, and the debate that followed gets stuck on whether a founder should be that involved. That’s the boring half. What’s worth studying is what the involvement produces.

On Canva AI 2.0 we were designing memory for the agent. What we brought to the review was a strong version of the pattern every assistant had converged on: remembered facts, stated back as a list. Mel’s objection wasn’t that it was wrong. Her word was techie: memory the way engineers understand memory.

So she opened a blank Canva design and started drawing.

What she prototyped in the room was a library. Not a list of remembered facts, but a shelf of books organized by category: the big events in your life, the campaigns you run, the things that are simply true about you. Then she pushed it into the onboarding, where a single button would read everything you had ever made in Canva and write an About Me document to seed the library on your first day. It shipped as the Memory Library, and the first book in it is called About Me. She didn’t write a brief or ask for three options. She made the artifact herself, in the product, in front of the people who were going to build it.

The design we brought into that room was weeks of work, and it was good work. Ten minutes later it wasn’t the plan anymore.

At most companies that pattern would be corrosive. The thing that keeps a team fast is believing that the work they’re doing now is...