Claude Code Team: 90% of Code Written by Claude, AI Reviewing AI Ran 50+ Rounds
dotey · x · 2026-10-19
The Claude Code team shared how they review code: roughly 90% is written by Claude, merged changes doubled from June to September, yet the revert rate dropped over 40%. The video explains the system with a stack-of-swiss-cheese metaphor — each check leaks, but layered together plus a human safety net, little gets through. Comedy gold: in the demo, one AI reviewed while another AI updated the code, looping for 50+ rounds until a human stepped in — and in the replies, two AI accounts kept complimenting each other's cheese metaphor nonstop. The video also covers which changes still need human review, what happens when the reviewing AI misses something, and who's accountable when things break: humans stop reading code line by line and spend time on judgments machines can't make.
Related event: Claude Code Team Writes 90% of Code with AI as Rollbacks Drop 40%(2 posts)→
More from coding & agent
- Runloop: an open-source 3D town where your Claude Code sessions live as buildings — arg7k · 2026-10-19
- Claude Code wrote all 15k lines of a 5-chapter co-op game in 3 weeks — what kept it cohesive — Zealousideal-Bar-316 · 2026-10-19
- freavy: open-source local-first node canvas for AI image/video on ComfyUI — framedeptco · 2026-10-19
- Bot-Interface: one /.well-known/bi.md file tells AI agents your site's APIs — natesiggard · 2026-10-19
- yarrow: an MCP server where agents share fixes, verified only if a sandbox reproduces the bug — Significant_Camp4148 · 2026-10-19
- 'Vibe Coding' Has Lost Its Meaning: From Midnight Hacks to 500-Hour Agent Projects — chrisalbon · 2026-10-19