How This Ex-Meta L8 Engineer Ships 40 PRs a Day with AI Agents
- Video: How This Ex-Meta L8 Engineer Ships 40 PRs a Day with AI Agents — Kun Chen, hosted by Peter Yang (56:18)
- Tags:
ai-agentsagentic-codingworkflowproductivity
Key Points
Kun Chen (ex-Meta/Microsoft L8, now solo AI builder) shows his live agent workflow.
- The scale: at least 5 active sessions, 20–30 agents running on average, 20–40 PRs every day.
- The bottleneck insight: “If you review every single line of code, you become the bottleneck.” He doesn’t review the agent’s first-pass code — and eventually found he “never catches anything the agents don’t catch” anyway.
- Workflow = Plan → Code → Validate (standard), but the time split is what’s different:
- Plan: most of his time; mostly him, assisted by agents. Clear requirements up front.
- Code: almost entirely the agents.
- Validate: agents do most of it, with his judgment only where things are ambiguous.
- The shift: team workflows were built for when humans spent most time coding. Writing 10x more PRs means you must move the human out of the loop as much as possible and redesign how work scales.
Related
- spectrum-of-agentic-coding — agentic coding landscape
- context-v15 — multi-agent orchestration
- agentic-ai-coding-codescene — CodeScene’s analysis of agentic coding