Hi everyone, it's Tuesday, February 17th. Welcome to Tech Check live. I'm Georgia. Most software CEOs are struggling with how to survive when AI can do what their product does, but Dylan Field decided to partner with the company most likely to disrupt him. Figma today announced Code to Canvas, a feature built with Anthropic that lets users bring code-generated designs into Figma. Dylan Field, CEO Figma, joins me now. Dylan, thank you for joining us.
Thanks for having me. Good to see you.
So maybe let's start. How did this partnership come together? Did you call Anthropic or did they call you?
We've been exploring various things with Anthropic historically. This is an evolution of our existing workflow. We already have a way to go from design to code through dev mode to MCP. Now we're enabling the reverse: from code to design canvas. This lets you bring AI-generated code into Figma and benefit from the infinite canvas, collaboration, and the ability to iterate visually. The round trip is extremely important.
So you're combining the tools. There were posts showing Anthropic using Claude Code and naming Figma before this partnership was official. What does this partnership change?
Our users can now bring their work from Claude Code into Figma and get all the benefits of the design canvas: infinite space, exploration, side-by-side comparison, and the ability to zoom out and see the big picture. Many people work fast but may not be going in the right direction. This helps them avoid tunnel vision. Taste, craft, and visual thinking are differentiators. The round trip from code to design will only get better with more partners.
At the crux of the selloff in software is this idea: do you need that round trip? What are your views on AI replacing the whole design framework, rendering Figma pointless?
I have a different point of view. We're in an era where you can build things quickly, but you need to build the right thing. Good enough isn't enough. Engineers now have to consider what to build, which is the world of design. Companies need a point of view to differentiate. The stakes are higher, and more software will be created. It's an exciting time, and Figma will get better as models get better.
What's different about this moment in the last six weeks that's causing so much fear in the market? Why do you see it as an opportunity?
I've always expected model capabilities to improve. That's why Figma must get better as models get better. Every company needs to consider if they improve as AI improves. The world is moving from closed systems to open systems. In open systems, you can't simulate everything; you need a point of view. History shows that technology leaps require thinking about the bigger picture.
What do you need to understand about that? Embracing that open model.
In the past, closed systems allowed you to model all potential players. Now with open systems, you have to think about what's possible and develop a point of view. That critical thinking is needed now more than ever. When you can build quickly, you need to focus on the bigger picture and the design process. History informs the future.
There was a post arguing that companies need one of three things: proprietary data, regulatory lock-in, or being embedded in a transaction. What does Figma have that can't be replicated?
We have multiple modes, and everyone should think about building new modes. Don't lock users into one workflow. Our bet is that wherever you start—napkin sketch, shower thought, design, or code—you can go anywhere else. Embrace the Unix philosophy of starting anywhere and going everywhere.
What is the biggest misconception people have about designing products with AI today?
The biggest misconception is that the first prompt is the final output. Iteration is key. We bought a company called Weave, now Figma Weave, which uses node-based editing to create pipelines refining AI outputs. Think of AI output as clay to shape, not a one-shot final product. Don't assume a single prompt is the process.
What happens when the AI agent becomes the integration layer?
There are verifiable and non-verifiable tasks. Design and creation are inherently non-verifiable. Agents need auditability. If you trust agents today, you're brave; they're not flawless. You need human-readable auditing. Also, consider agents working alongside you, like multiplayer in Figma. Seeing their cursors in real time gives intuitive understanding.
We're seeing more powerful agents this year, like Claude running on a separate device. I don't have time to sit next to an agent. Security concerns are improving. Where does that leave the interface?
