Design Tools Are Learning to Talk Back
Framer added a reasoning toggle, Paper raised $34M and open-sourced its shader library, Subframe and Pencil talk to coding agents via MCP, and Glaze builds Mac apps from plain-language descriptions. The design tool stack is bifurcating — here is what that looks like right now.
Framer↗ added a reasoning toggle to its AI agent this week. Light reasoning for quick edits, heavy reasoning when you want the model to plan first, fast mode when you are running Opus 5 and need it now. It is a small UX choice that makes a larger point: Framer↗ is treating its agent like a colleague whose thinking time you manage, not a feature you press.
Paper↗.design's MCP server exposes 24 bidirectional tools. Agents can read your design, modify it, sync tokens from Figma↗, pull live API data into UI components, and export directly to React or Tailwind from the canvas. Paper↗ raised $34M in July on 25x ARR growth since early 2026, and open-sourced Paper↗ Shaders↗ (their WebGL effects library) the same month. If you haven't looked at it recently, the pace of shipping there has picked up considerably.
Subframe↗ connects its visual editor to Claude Code↗ and Cursor↗ via MCP, so you stay in your IDE and pull design context without tab-switching. Pencil↗.dev goes further: it lives inside VS Code and Cursor↗, runs parallel sub-agents on different design scopes simultaneously (one on typography tokens, another on the hero section), and syncs bidirectionally to your codebase. Design tokens update in the editor when the agent updates them on the canvas.
Glaze↗ by Raycast runs it in reverse. You describe a Mac app in plain language, Glaze↗ uses Claude Code↗ or Codex↗ to write Swift, compiles it through Xcode, and installs the binary in your dock in 60 to 90 seconds. It has been public since July. The question it keeps raising: are "design tool" and "build tool" still usefully separate categories?
Figma↗ had a productive week too: agent skills (create reusable agent workflows from canvas context), text wrap improvements, Weave MCP. Expected updates from a well-run platform. The more interesting pressure on Figma↗ comes from tools rebuilding the canvas as agentic infrastructure.
The design tool stack is bifurcating. Some tools help humans design well, others help agents design at all. Paper↗, Subframe↗, Pencil↗, and Framer↗ are trying to do both — and that is the part worth watching.