Skip to main content
bash TV

Claude Writes 90% of Claude Code. How Does the Team Verify It?

TestMu AI (Formerly LambdaTest)

11 views20 Nov 2026

YouTube

Claude writes and reviews about 90% of Claude Code’s own code. The Claude Code team checks it by testing the real, running app end to end, using a separate grader, and keeping checks out of reach of the change. Without checks, you can run one or two AI coding agents. With them, dozens or hundreds. This short video sums up “How the Claude Code team thinks about verification”, the Nov 17, 2026 post by Addy Osmani, who works on Claude Code at Anthropic. These are Claude Code best practices any team can copy. It also shows how Kane CLI by TestMu AI covers the end-to-end testing and proof-on-every-run part, without building it yourself. Chapters 0:00 Verification is the bottleneck 0:15 3 problems and the 4 rules they became 0:33 How Kane CLI verifies AI-written code 0:44 Why most teams won’t build this themselves 0:54 What will you verify? Key takeaways - Without verification, you can run one or two agents. With it, dozens or hundreds. - Flaky tests teach agents to re-run instead of fix. Fix the test, or delete it with a written reason. - Unit tests can pass while the feature is broken. If you pick one kind of test, pick end-to-end. - A change should never be able to edit the check that judges it. - A separate grader judges the work, not the agent that wrote it. - People decide what “done” means. Every commit says how it was checked. How do you test AI-generated code? Run the real app end to end. Have a separate grader judge the result instead of the agent that wrote the code. Fix flaky tests or delete them with a reason, and never let a change edit the check that judges it. That is what the Claude Code team relies on most, alongside AI code review and reviews of their own reviewers. What is Kane CLI? Kane CLI is a terminal-based testing agent from TestMu AI, built for coding agents like Claude Code, Cursor, Codex and Gemini. You author a test once, then it replays in seconds with no LLM calls. Every run leaves an evidence file with steps, screenshots, network, console, and terminal logs, and a pass or fail verdict. 👉 Try Kane CLI: https://www.testmuai.com/kane-cli/?utm_source=youtube&utm_medium=organic&utm_term=2Ea5Y_PNUj4&utm_campaign=claudewriteswhoverifies Links Addy Osmani’s post: https://x.com/addyosmani/status/2122590683055993308 Kane CLI: https://testmuai.com/kane-cli Full write-up: https://medium.com/@mask.tech/verification-is-the-bottleneck-the-case-for-testmu-ai-94410ef10e47 #ClaudeCode #AITesting #KaneCLI

Join the discussion

Have something to add?

Sign in to reply, follow this thread, and earn XP when your answer helps someone.