宝玉 (@dotey)T2·Agents
71 pts
Lauren Tan, who leads the Grok Bot desktop app at SpaceXAI, shared her team's AI coding playbook at the AI Engineer conference: encode every correction as a lint rule instead of prose in agent docs, keep the codebase clean so AI does not copy bad patterns, ban error-prone APIs and 99% of comments, and have AI run and self-verify its own work with recordings. The team merged 11,000 PRs last month with roughly 0.5% reverted, and the app is faster than at launch.