WORTH TESTING · 20 OCTOBER 2026
The 'Anthropic engineer builds a harness in 10 minutes' video is a 2025 Claude Code keynote
Watched in full (10 minutes). It's Anthropic's own keynote segment by Cat Wu, a product manager on Claude Code, from the Claude Opus 4 and Sonnet 4 launch, which dates it to May 2025. The post's quotes about '90% of engineers running self-improving loops' and 'harness engineering' are not in it, and nothing is built live: she shows a sped-up recording of Claude Code spending about 90 minutes adding a table tool to the open-source whiteboard app Excalidraw. Verdict: a clear, honest demo of an AI coding agent, but old and not about harnesses. Intermediate level, coding required. It only has five real things to test.
Try this
- Use the Claude Code extension for VS Code or JetBrains to review the AI's proposed changes side by side before accepting them. 01:44
- Write one detailed feature request and let the agent make its own to-do list, explore the files and build it. 04:13
- Turn on auto-accept edits (Shift+Tab) only where tests and linters exist, so the agent can catch and fix its own mistakes. 05:04
- Ask in plain English to 'push up current changes and create a PR' instead of typing Git commands. 06:57
- With Claude Code in GitHub Actions, mention @claude in a pull request comment to get chores like documentation done. 07:28
Timestamps point to the moment in the source video.
Cost
Free to watch on X. Our notes don't cover the cost of the tools shown.
Source
Hype or Fact?
Related AI tools
How we knowSteps to try, not a claim that we independently tested every result. Follow the linked sources and check current prices. Our methods & sources ↗