Philip Su Gave Up Coding. Now He Manages AI Agents Instead.

Philip Su has a resume most engineers would kill for. Distinguished Engineer at Meta, stints at Microsoft and OpenAI. Now he's building Superphonic, an AI-powered podcast player, and he hasn't personally written a line of code in four months. Instead, he spends his days like a classic engineering manager, mostly setting priorities for AI agents and resolving conflicts between their recommendations. The actual review happens at the aggregate level, not line by line. His argument is blunt. When developers produce 5x to 20x more code using AI, there's no time for traditional code reviews. And more fundamentally, at some point human intervention becomes a liability, not a safeguard.

Su calls this the "lights-out codebase," borrowing from data centers that run with zero humans on site. Code where no person ever sees or edits the source. He draws a parallel to chess engines: nobody would suggest having a human grandmaster review every Stockfish move just to be safe. We've accepted that machines play better chess. Su believes we'll reach the same place with code, and when we do, requiring human review will seem just as ridiculous.

The Hacker News crowd is split. Some compare it to trusting code from external teams in large organizations, where you rely on testing strategies rather than line-by-line auditing. Others point out that current AI models still make enough errors that fully lights-out engineering feels premature. Su acknowledges he's arguing about direction, not current reality. But his practical advice cuts through: stop reading about AI and start building with it. Pay for the tokens and see what actually works.

The structural question is what happens to engineering teams. If five engineers become functionally 100 through AI amplification, but you keep the same PM and designer, the bottleneck shifts hard toward product judgment and coordination. Su thinks most CEOs are getting this wrong by cutting developers to save money. The smarter play is using that surplus capacity to tackle the next three items on your strategic roadmap. Despite concerns about AI eliminating jobs, many experts note that AI is transforming rather than eliminating software engineering roles, with job postings up 11% annually as companies need engineers to oversee AI-powered code-writing agents and design software architecture.