Config week: the canvas ate everything
Figma Config 2026 shipped Code Layers, Motion, Shaders, and AI Connectors. Framer countered with its own agents and shaders. Both tools are absorbing what the other used to own.
At Config this week, Figma↗ shipped something I've been waiting for someone to do: Code Layers, which turns any design layer into running, interactive code without leaving the file. You flip a rectangle into a component and iterate on it with your designer in the same canvas, commenting and running variants side by side.
Config 2026 went further. Figma↗ Motion is now a native timeline animation tool — keyframes, physics presets, built directly in. Figma↗ has been watching Framer↗ define what good web animation looks like for years. Shaders↗ let you describe a visual effect and the agent builds it. Generative Plugins work the same way: describe what you need, it arrives without a dev environment. Weave Tools — style transfer, image generation, asset extraction — are now on the canvas rather than in a separate product. And Agent Skills plus Connectors mean your Figma↗ agent can now reach Notion, Slack, Granola, GitHub, and Atlassian from inside the file.
Meanwhile, Framer↗ shipped interactive shaders↗ the same week (the Ripple Shader: WebGL-powered, cursor↗-following image effects). Framer↗ 3.0, earlier this month, brought AI Agents and Branching: Git's branch-and-merge model applied to website editing, letting teams review agent-driven changes before they hit production. Framer↗ is moving toward structured team workflow features. Figma↗ is moving toward production code. Both are absorbing what the other used to own.
Code Layers matters because it treats code as a first-class material on the canvas rather than an artifact you generate after the design work is done. Motion matters because animation has always been the handoff problem nobody solved cleanly: the Figma↗ file says one thing, the shipped product does something slightly different. Native motion closes that loop.
What I'm less sure about is how much complexity a single canvas can carry before it starts working against the people using it. Figma↗'s answer is that agents do the heavy lifting. Framer↗'s answer is essentially the same. Both bets depend on the agents being genuinely useful rather than impressively scoped.
One smaller thing worth watching: onBeacon launched this week with AI UX audits across 120+ behavioural science principles. Automated design review has been promised for years; onBeacon is the most credible version of it I've seen.
The canvas is absorbing everything. Whether that's good or a lot to deal with probably depends on how well those agents hold up.