Group Chats With AI Don't Work — He Built What Comes Next | Steve Ruiz, Founder@tldraw

Why didn't you become a lovable or a boat?

Oh, you sound like my investors though. I bet with TealDraw was like, no, I could probably just make normal website stuff work like a canvas works. And if I could do that, then I would be able to put anything onto that canvas. Video, GIF, interactive stuff, or even instances of TealDraw itself. This did not feel like a product. This felt like a pattern. This idea of taking drawings and diagrams and using them as input to AI models felt like 1,000 products and 1,000 applications rather than just one. Shopify does that. I have a fine art background. I spent a lot of my life working with pens and paint, doing drawing and painting as an artist. Grew out of a kind of an open source project that got really popular. Kind of given myself a couple of months before going to take a job at Adobe, and I was like, hey, let me finish this open source thing that I've been hacking on before that. But then it got popular, and then it kept getting popular, and then now I'm still there. And now that programmers have these tools, what do they do with them? Someone's handed me just like God in a box, and I'm like, all right, how do I make it faster? How do I run two of these things at I want. So like the impulse to innovate, the impulse to engineer. I think what we're going to find is that we've actually been having a really shallow engagement with the amount of knowledge work that could be done historically.

If Canvas really becomes the next UI paradigm after chat interfaces, you are going to be at the center of that thing, right?

I'm building for it one way or another.

You are the founder who is having most fun.

So the first one that we did that was really popular with AI is, was this thing called DrawFast. And in order to appreciate this these days, you have to remember that the year is 2024. GPT-4 with vision has just come out. And that's the first time that developers have had a chance to play around with like vision APIs for language models for AI. And so here's, I want this to be like the, I'm going to just make something. Imagine that this is like a camera feed. And then there's a button down here that says, or is it take photo? And for one of the biggest use cases for TealDraw was like wireframing and things like that. And we actually was a designer at Figma who first came up with this, but they used TealDraw to prototype it because it was, it was easier to do that.

So they choose this stuff instead of Figma?

We would take this and we send it off to GPT-4 with Vision at the time. And now I think this is using Gemini 3, so we'll see how it does. And say, hey, give me like a working prototype based on this image. I think the system prompt had something like, you're a senior developer, you're 4,000 years old, you just, you work with designers and you turn their sketches into prototypes. You just got this design from your designers. Make it real. Respond with a single page HTML and CSS, whatever file. Again, you gotta remember there was no— when we were working on this, there was no Lovable. There was no vibe coding, wasn't a term yet.

This went viral, right? 2 years ago. Are people actually using it?

This got very popular, but it was so new that it really couldn't be a product.

Why didn't you become a Lovable or Bolt?

Yeah. Oh, you sound like my investors. This did not feel like a product. This felt like a pattern, right? This idea of taking drawings and diagrams and using them as input into AI models felt like 1,000 products and 1,000 applications rather than just one.

Do you think Canvas is becoming the new UI? For AI, Chat was the default. Do you think Canvas is the next one?

So I think that Canvas has a place, just I think terminal apps have a place. I guess it's a question about like for whom and in what setting and what are the different constraints of those different platforms. What I think is really cool about the Canvas is that all of the things that make a canvas good for collaborating with other people make it good for collaborating with AI. As an example, when I have a conversation with a model or something, if I'm doing some work with whether that's ChatGPT or whether that's like a coding agent or something like that, there's not a lot of room in that experience for anyone else. That it's like having group chats with an AI actually isn't a very good experience. Group chats in general aren't a very good experience because they're one-dimensional. Like every message it pushes to the end of a big list and the canvas, why the canvas is One of the reasons why Canvas is so good and why it's so popular in situations like Figma or Miro for collaboration is that everyone can be working at the same time in parallel with each other. And there's really no limit on how many people that paradigm can support because you, if it's busy, you just go somewhere else, right? If everyone's working over here, you just start working over here. We did an experiment towards the end of for the year using AI agents that were kind of like spatially located, that lived on the canvas. And they were like virtual collaborators, except they were definitely not as good at anything as people would have been. So instead we decided to make them into fairies.

I bought fairies for my daughter. Oh, this is so cute.

A lot of the design of these things is heavily influenced by whether, whether my daughter can play with it or not. So these are AI agents, just like the, uh, AI agent that I showed over on that other website, but they're located, they're on the canvas. You can move them around and you can select them and you can have like a conversation with one of them, or you can have a conversation with all of them. And when you think about the parallels, and we did a lot of thinking about the parallels, thinking of these as agents, like that type of conversation with multiple agents at the same time, or the way that they self-organize and things like that, I think is, these are really hard problems to solve. In any situation, but they, some of them are problems that like have been solved on the canvas already. So some, one of the questions is like, where is my agent? What are they working on? Or something like that. And on the canvas, they're working on wherever they are.

The fairies are moving them one by one, one stroke at a time. That's so silly.

There we go. All right, cool. Shoes on the cat. And now if I wanted to do something with all three of them. I could do that as well. Make a few more animals. So when working together, when addressing all three of these things, they work a little bit differently. This one, one of them gets elected as the leader and it creates a plan. Their wing color changes to indicate that they're part of a group. And then now they can, this one will begin giving tasks to the other ones, which they will complete in parallel. Agents on the canvas working together.

Yeah. Visualize. This is visualize the agents. There's a fair amount of fear, especially in the US market. There was a lot of layoffs in the FAANG companies. How are you preparing your kids for this? For her?

For the future of work?

Yeah.

I'm like positive on, it's probably an unpopular view, but I think that what I've seen with even in the last 2 months with people coming on to Claude Code 4.5, which if you're viewing this in the future, seemed like a tipping point moment for coding AIs. And I think what I'm recognizing is that, wow, there's actually a lot of work to do. There's, there's a lot of stuff that like we, we were never trying to solve, like in my startup and in my team, in my monorepo, in my like life, a lot of like tedious knowledge work, a lot of like things where, wow, if I had 15 really smart consultants, like I could do a lot with that. Other programmers have these tools. What do they do with them? They start engineering on top of them and with them. It would be like someone's handed me just like God in a box and I'm like, all right, how do I make it faster? How do I run two of these things at once? How do I optimize my God box in order to make this thing more powerful? That the impulse to innovate, the impulse to engineer, the impulse to like just like do more. Yeah. Do more and to do better and to do, to go deeper and stuff like that. I think what we're going to find is that we've actually been having a really shallow engagement with the amount of knowledge work that could be done historically.

The world will become a better place. Hopefully.

I hope so. Or, or else paperclips, but I hope, yeah, hopefully not.

If you would recommend like a book to, let's say, the lost new grads starting their career, what would they be and why?

I really liked a series of books written by Kim Stanley Robinson called Red Mars, I think is the first one. I read them in university. They're big chunky books. He's a geologist, so there's a lot of text about rocks in there. But the thing that I liked about this series of books was that the, as a narrative conceit, people lived a long time. The first page of the book is, yeah, don't ask us how, but we've extended life to 500 years. And what I really liked about the book is that because folks just had more time. There was a complete fearlessness about learning something new. The idea that like, I've been a physicist my whole life, but now I'm— I need to solve a problem that requires me to have a PhD-level understanding of biology. All right, I got time. Let me spend the next 12 years studying biology in order to be able to solve this problem. And being able to think of those kind of like long-term thoughts and the, again, the kind of the fearlessness about spreading yourself too thin or not being able to engage deep enough. Who cares? Life is long, and I think that is— that's always been a very inspiring thing for me.

Tools like Teodraw or Clockhold are extending our lives in a sense.

I never thought that I would be doing firmware for little development boards, but I'm doing that now, thanks to AI.

What strikes me most about this conversation is how much you care about the craft, how a line feels, how it curves, and the vast amount of imagination, of creativity that's potential there. A lot of people are chasing the next AI hype.

Cycle.

I feel you're adamant about building infrastructure that lasts. I don't know if Canvas really become the next UI paradigm, you are going to be at the center of that thing, right?

I'm building for it one way or another.