Some friends and I started working on a tool to help us deal with the increased output from AI. Instead of reviewing large PRs, we review architecture diagrams with proofs. Most of the companies we worked at had stopped reviewing code, and reading large output got exhausting.

We built https://draw.redcrayon.ai, inspired by tools we love like Excalidraw: a live architecture board you and your coding agent draw on together. The agent connects over MCP (Claude Code, Cursor, Codex, claude.ai) and draws what it changes as it works. The edges are meant to cite a source line (or fingerprint). You read the board instead of walls of text, and you can circle something in red crayon to get the agent to answer/fix it.

What we use it for:

- An incident scratch pad, where the agent lays out what it finds while debugging

- Seeing how a change affects the architecture before reading the diff

- Flagging the risky parts, like a DB migration, with every piece of code that reads that table

- Github App that generates a diagram per PR (something that is a work in progress)

- Ever wanted to create a branch of your existing diagram rather than completely clone/copy it? That's something we added to draw.

Docs: https://docs.redcrayon.ai

OSS (Apache-2.0): https://github.com/redcrayonai/draw