TL;DR
OpenAI product lead shares concrete Codex workflows achieving 70% higher PR output, emphasizing context, automated code review, and structured planning over raw code generation.
Key Points
- Codex users at OpenAI produce ~70% more PRs than non-users; now 95%+ of technical staff use it constantly
- Automated code review enabled on nearly every OpenAI repo with high confidence filtering to protect human attention
- Plans.md structured planning template critical for complex projects; Sora Android app built in 28 days using this approach
- GPT-5.2 solves previously unsolvable bugs with extended reasoning; human direction now the limiting factor, not code generation
Why It Matters
This reveals the actual production workflows for AI-assisted development at scale, moving beyond hype to concrete engineering practices. For developers, the key insight is that context and structured planning matter more than prompting cleverness—and that the bottleneck has shifted from code generation to human judgment and product direction.
Source: www.lennysnewsletter.com