How to Become a Model PM: A 6-Level Roadmap — Only 7 of 316 Have a PhD
aakashgupta · x · 2026-10-20
Aakash Gupta distills interviews with model PMs at OpenAI, Anthropic, Google and Meta's Llama team into a six-level roadmap.
- Level 1: how models are made (Karpathy's LLM deep dive, Stanford CS336)
- Level 2: behavior shaping — RLHF Book, OpenAI Model Spec vs Claude's constitution; build a one-page behavior spec
- Level 3: evals — the core skill; build an eval card from 30-40 failure cases; OpenAI's Core Models posting asks for distinguishing useful vs convenient metrics
- Level 4: follow the frontier via system cards; pick three numbers to watch
- Level 5: ship to developers — release notes and migration guides
- Level 6: land the job with one portfolio piece and a referral
Key data: none of 46 open model PM roles requires hands-on ML research; only 7 of 316 PMs at model-training companies have a PhD. Already a PM? Start at Level 3.
More from coding & agent
- Personal software factories are a waste of time, argues developer — rseroter · 2026-10-20
- Keyfleet lets personal AI agents join permanent fleets via 10–100 USDC onchain Keys — seanwbren · 2026-10-20
- Claude SEO, the 15k-star toolkit, now runs natively in Hermes with 45 SEO skills — MicahBerkley · 2026-10-20
- Testing EverOS: AI Agents Need Experience, Not Just Memory — Aiden_Tech_Ai · 2026-10-20
- Investor shares his internal agent stack: Mastra, Trigger.dev, Context and more — ryanbed · 2026-10-20
- 'Suddenly everyone is building Sandboxes' — agent sandboxes become the new gold rush — mgill25 · 2026-10-20