Summary
As AI coding agents make it trivial to turn a prompt into a pull request, teams face a new problem: how to review and merge changes at a volume humans can’t read line by line. Theo of t3.gg works through a write-up on how the Claude Code team handles this, and compares it with his own experience on T3 Code, which has received as many as 800 pull requests in a single day with a three-person team.
The video covers the Claude Code approach of minimizing human review time without raising reverts or incidents, running an automated Claude review on every PR before a person looks, and using review files written by code owners to describe what failure looks like in each part of a codebase. Theo adds his own guidance on fixing problems at the highest-leverage layer first: improve the codebase itself, then add linters, type checks, and static analysis, and only then reach for Claude.md files, skills, or other prose instructions.
He also shares how he thinks about onboarding agents like a highly capable new teammate who doesn’t know the repo, and describes process lessons from scaling T3 Code with Opus 5.5. The video includes a sponsor segment for Browserbase.
📺 Source: Theo – t3․gg · Published November 06, 2026
🏷️ Format: Opinion Editorial
