Bring your own AI
Design tools are competing to be the best canvas for whatever AI you're already using. Framer adds external agent support, Pencil doubles down on its MCP model, and Paper opens its shader library.
Bring your own AI
Three design tools shipped updates this week with the same argument.
Framer↗ added support for external AI agents. Claude Code↗, Cursor↗, Codex↗, and Gemini↗ CLI can now all connect directly to your Framer↗ project alongside the canvas-native agents that shipped in Framer↗ 3.0 in June. The practical upside is real: you can import existing Next.js codebases into Framer↗ and have your regular agent work on them inside the canvas. That sounds like a minor integration until you realise what it means: Framer↗ is betting that the best AI in your workflow is probably not theirs.
Pencil↗ has been making this bet since day one. The whole tool is an IDE-native canvas that offloads code generation to whatever agent you're running. Supply your own Claude API key, point it at Cursor↗, use Codex↗ — Pencil↗ provides the visual layer and leaves intelligence to you. The latest CLI update makes this scriptable: you can run its MCP tools from automated workflows without opening the UI.
Paper↗ went the other way on shaders↗ but made a similar call. Their WebGL effects library went Apache 2.0 open source at the start of the month. Thirty-plus effects — mesh gradients, noise textures, halftone, aurora backgrounds — all zero-dependency JavaScript, installable via npm, free for commercial use without attribution. Paper↗ keeps the paid canvas and opens up the visual primitives.
The tools doing well right now figured out how to be the best place to work when AI is already in the room. The ones that just bolted on a model didn't.
Elsewhere: Icon Animator launched for the specific job of animating icons for UI. Frame-by-frame Lottie export, browser-based, no install. A small tool with a clear job, which is more than most.
OpenPencil↗ continues picking up attention: open-source, reads .fig files natively, WebRTC p2p collaboration with no server, 7MB desktop app. Data stays local. Still early, but the GitHub star count is moving.