AgentCanvas is a friendlier way to steer the app an AI is building for you. It turns your app into a plain-English map of what it does: user journeys shown as simple "When / Do / If" flows. You change how the app works by editing steps in your own words, and your AI coding agent (Claude Code, Codex, Cursor) builds the change.
How an AI agent uses AgentCanvas: run agentcanvas start --workspace ., open the local URL it prints, then watch .agentcanvas/pending for plain-English change requests, implement each one, run the relevant tests, and re-index with agentcanvas index --workspace ..
agentcanvas start --workspace .
.agentcanvas/pending
agentcanvas index --workspace .