Google evolves Stitch into an AI-native design canvas with DESIGN.md and multi-threaded agents

What Google highlighted on X

On March 18, 2026, Google AI said it is evolving Stitch into an AI design canvas that transforms natural-language prompts into production-ready front-end code. The X post framed the update around three concrete additions: multimodal inputs such as text, images, and code; a context-aware design agent; and a new DESIGN.md file for moving design rules across tools.

That framing matters because it shifts Stitch from a one-shot mockup generator toward a more complete design workflow surface. Instead of using AI only to draft an initial layout, Google is positioning Stitch as a shared workspace where iteration, critique, prototyping, and design-system transfer all happen inside the same environment.

What the Google Labs post adds

Google's official blog says Stitch is becoming an AI-native software design canvas for creating, iterating, and collaborating on high-fidelity UI from natural language. The redesigned interface is built around an infinite canvas that can take images, text, or even code as context, which is a stronger workflow claim than a typical prompt-to-screen demo.

• The new design agent is described as reasoning across the entire project evolution rather than only the current screen.
• The new Agent manager is meant to track progress and help users explore multiple ideas in parallel.
• Google says DESIGN.md is an agent-friendly markdown file that can export or import design rules between design and coding tools.
• Stitch also now turns static screens into interactive prototypes instantly and supports voice-based design edits.

Why this matters

For product teams, the significance is not only UI generation speed. The more important change is that design intent, system rules, and iteration history are being made legible to agents. A file like DESIGN.md hints at a future where design systems become portable, machine-readable operating context rather than something locked inside screenshots, component libraries, or human memory.

It also pushes the design-to-code market toward longer workflows. If Stitch can hold multimodal context, critique designs, prototype instantly, and hand structured rules to coding tools, then the competition is no longer just who can generate the prettiest first screen. It becomes who can preserve intent and coordination from idea to prototype to implementation.

Sources: Google AI X post · Google Labs blog