When I started designing, it was in Photoshop. Everything was layers, and we used it to build websites, even though it was never really made for that.
Then Sketch came along, and it blew my mind. It was an open canvas where you could just drag frames around and build a design that way.
Then Figma took it further with multiplayer. Other people's cursors became part of the interface, and the canvas turned into a place you work together.
Since then, canvases have been a big part of how I work, in Figma, tldraw and especially Miro. Lately, a lot of that work has been about putting AI on one.
Most AI still lives in a text box. You type, it answers, and the conversation scrolls on. That works well for chatting, and it works well for coding. I'm less sure it works for collaborating. A chat is one person and one AI taking turns, and everyone else hears about it afterwards.
On a canvas the work stays where it was made, and everyone can see it. You can see what the AI is doing while it does it, what it did once it's done, and what your team did around it. The more work AI takes on, the more that matters.
You can see it work
Space shows activity in a way a chat can't. When AI works on a canvas, you can see where it's working. Notes show up in a corner, a frame fills in, a diagram grows next to the brief it came from. It doesn't have to tell you what it did, because you can see it.
For me, that's when it clicked. A chat goes quiet and then dumps the whole result on you once it's done. On a canvas you watch a cursor move around and build things up piece by piece, and it starts to feel like someone is working there with you.
Space also lets work happen in parallel. Several agents can work on the same board at once, each in its own spot, and the whole team can see all of it. Run the same work in separate chats and it ends up split across windows that only one person is looking at.
Multiplayer is why that works. You can tell who's in the file, what they're looking at and when to leave them alone. AI on a canvas can join in the same way, as one more cursor on the board. When it's slow, waiting turns into watching a teammate work.
The board is the prompt
Following the work goes both ways. The AI should be able to follow yours too.
Look at a whiteboard after a workshop. Nobody wrote down what anything means, but everyone in the room can read it, because the meaning is in where things are. Notes placed close together are one idea. The one with a circle around it is the one the group picked. The ones pushed to the edge are for another day.
A text box makes you describe all of that in words. On a canvas you can point at it and say "make this blue", "put three more here" or "turn these into a table". Words like "here" and "this" only work when there's something to point at, and they're how you'd talk to someone standing next to you.
Selection works the same way, because what you've selected tells the AI what you mean. A frame with a few notes, a rough sketch and a logo is a better brief than a paragraph, and a messy board is already good context. You shouldn't have to tidy it up first.
That's when AI starts to feel like part of the team instead of a tool on the side, because you don't have to explain the board to it. It reads it the same way everyone else does.
The work stays
A chat gives you the answer, but a canvas also shows you how you got there.
A Figma file is rarely clean (you should see mine). The final screen sits next to the versions that didn't make it, and that's how the team remembers why. Rows of artboards in Sketch and Figma are how a lot of us learned to decide.
AI makes alternatives in bulk, and a chat shows them one at a time. You end up scrolling up and down to compare option three with option seven. A canvas shows all of them side by side, next to whatever they came from, and the team can point at the one they like.
That history matters even more when AI did part of the work. When something looks off, you want to see what it started from, what it was asked and what it tried first.
On a canvas the brief, the attempts and the result can sit next to each other. In a chat they're spread across one long scroll, and even when you share it, everyone has to read it from the top to find out what happened.
Plans are the same. A plan in a chat scrolls away. A plan can sit on your canvas and take whatever shape the work needs. It could be a kanban board that shows what's moving, a brief the whole team can read, a timeline or a map of who does what. People can move it around, split it up and come back to it, and it stays right next to the work it's planning.
It's also why I like showing work on a board. Slides are still great for presenting, but when I lay out a feature with the demos, the notes and my thinking next to each other, people can see how it all fits together. They can read up on it before the meeting or come back to it afterwards, and it's all still there.
A space to work
A board can be a lot of things. Sometimes it's where you pull the context and documentation together, or present something that's close to done. Other times it's where a team does the work itself, and then it's allowed to stay unfinished.
That changes what AI should make. When a team asks for a retro, the most useful board is a place for people to work in, with a few empty notes, some frames with space to grow and a vote that's ready when the team is.
A retro has a known shape, so AI can set it all up in one go. It doesn't even matter if you've never run one before, because it can walk you through each step while the team fills in the notes.
A lot of work doesn't have a set end like that. When a team is mapping out a new feature, nobody knows yet what the board should look like. Here AI can build it up one piece at a time with them, with loose notes first and frames once the groups appear, so the structure follows what the work turns out to be.
The small details matter here as well. A kanban board made of rectangles looks right, but if you can't drag the cards it's just a picture of a kanban board. And a good board usually ends on something to do next.
Tools made for your team
A board can hold more than notes too. A small tool made for one team or one meeting is now just a sentence away. A scorecard for the candidates you're interviewing this week, a tug-of-war vote, a weather report for the team's mood, or a little game to settle whether a hot dog is a sandwich. Some of them stick around. Most won't outlive the meeting, and that's fine.
Making a tool used to mean building a plugin or an app, so you only did it for things lots of people needed. Now you can make one for your team, for this week, or for one odd thing only you care about.
Because it lives on a shared canvas, everyone can use it the moment it shows up, instead of watching one person use it on a shared screen. I love that it works for both the serious decisions and the small things that give a team its personality.
The tool itself deserves as much design care as anything else on the board. If you ask for a mood check-in, you shouldn't have to write a design brief to get something that looks considered and feels like it belongs there.
Software doesn't have to live somewhere else anymore. It can show up on the board, right when your team needs it.
To infinity…
The infinite part is easy to take for granted. You never run out of page, so you never have to throw something away to make room for the next thing.
That matters more when AI is making things. It can make a lot, and fast, and a canvas can hold all of it without anything getting pushed out. Earlier ideas can stay off to the side while new ones get a spot of their own.
To be fair, that's also how boards get out of hand. Everyone has opened an old board and spent the first five minutes just trying to find their way around.
AI can help here too. Zoom out on a busy board and it can sum up what's there. Zoom back in and the details are right where they were.
A board that never ends is also a lot of context, and an AI that tries to read all of it gets slow and confused.
So instead of showing it everything, you can do what people already do on a canvas and look at what's selected, what's nearby and what just changed, and leave the rest for later. When the AI only reads what matters right now, it gets better, and it also gets a lot faster.
…and beyond
Most AI still makes you wait. You ask, it thinks, and a while later something shows up. We've gotten so used to it that a lot of AI interfaces are designed around the wait, with spinners and little status messages to keep you company.
That's starting to change. Small models can now answer in a second or two, and I think that's what will change how we work with AI the most. When an answer takes a minute, you write a careful prompt and hope it lands. When it takes a second, you can jam with it.
You don't need to know what the end result looks like before you start. You try something, the AI tries something back, you keep the bit you like and push it somewhere else. You get there together, the same way you would sketching with someone at a whiteboard.
On a canvas that raises some fun questions. Do you still need a toolbar if you can point at the board and say "add five stickies here"? Do you need a text box at all? A lot of the answers end up looking like a whiteboard, where all you had was an empty wall, a marker and whatever you said out loud.
The smart, slow models don't go away. What works really well is letting the two split the job.
If I ask for a plan for next week's launch, a fast model works out that it should be a kanban board and puts one on the board right away. Then the smart model fills in the cards. Filling it in still takes as long as it takes, but something shows up the moment you ask, and that changes how the whole thing feels.
When AI is this fast, it can work right alongside you. You can look away, look back and see what changed, and it doesn't need to move your view, touch what you're working on or interrupt anyone. People don't take turns at a whiteboard, and AI shouldn't make you take turns either.
There's still a time for the big generations, where you set something going and come back half an hour later, and that's fine. But the quick part is what I'm most excited about. When AI stops making you wait, it can keep up with a whole team working on the same board, and working with it starts to feel a lot more like working with each other.
Somewhere to come back to
AI is getting very good at making things. It can make ideas, documents, tools and whole boards faster than anyone can read them. The more it makes, the more it matters that the work has somewhere to live, where you can see it, follow it and find it again.
That's what a canvas has always been good at. You can put a thought down before you know where it belongs, move it when you do, and leave room for the next one. It keeps the version you didn't pick next to the one you did, and it keeps your place until you're back.
A good board is never really finished. You just leave it in a state where the next person can pick it up. Everything is where it was put, the reasons sit next to the result, and there's a bit of empty space left.