Expect: Finally, Coding Agents Can See What They Ship
Here is the biggest blind spot in the coding agent workflow: agents write code, but they cannot see the result. They generate a React component and have zero idea if the button renders in the right place. Until now.
Expect (expect.dev, github.com/millionco/expect) is a tool from Million Software (YC W24) that gives coding agents eyes. One command scans your git diff, generates a test plan via AI, and executes it against a live browser with Playwright. It works with Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot, and more through the Agent Client Protocol.
The killer feature is authentication handling. Expect extracts cookies from your local browser profiles — Chrome, Firefox, Safari — and injects them into Playwright sessions. No more setting up test accounts or mock auth. Your agent tests with your real login sessions, against your real staging environment.
Built by Aiden Bai, who created Million.js at 16 and later React Scan — projects used by Airbnb, Robinhood, Perplexity, and Shopify. The repo already has 2.8K stars on GitHub with 768 commits showing active development. TypeScript, FSL-1.1-MIT license.
Expect ships with an interactive TUI for reviewing test plans before execution, headless CI mode with proper exit codes, and rrweb session recordings for replay. This is the visual verification layer that coding agents have been missing — the difference between generating code and actually shipping working UI.
← Back to all articles
Expect (expect.dev, github.com/millionco/expect) is a tool from Million Software (YC W24) that gives coding agents eyes. One command scans your git diff, generates a test plan via AI, and executes it against a live browser with Playwright. It works with Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot, and more through the Agent Client Protocol.
The killer feature is authentication handling. Expect extracts cookies from your local browser profiles — Chrome, Firefox, Safari — and injects them into Playwright sessions. No more setting up test accounts or mock auth. Your agent tests with your real login sessions, against your real staging environment.
Built by Aiden Bai, who created Million.js at 16 and later React Scan — projects used by Airbnb, Robinhood, Perplexity, and Shopify. The repo already has 2.8K stars on GitHub with 768 commits showing active development. TypeScript, FSL-1.1-MIT license.
Expect ships with an interactive TUI for reviewing test plans before execution, headless CI mode with proper exit codes, and rrweb session recordings for replay. This is the visual verification layer that coding agents have been missing — the difference between generating code and actually shipping working UI.
Comments