Went through the EM loop at OpenAI a couple months back. Declined the offer for personal reasons but the process was genuinely interesting, sharing what I saw.
First: this is not a pure IC-plus-spreadsheet EM role. They want people who are deeply technical and who are thinking seriously about the implications of what they build. That shaped every round.
Process: Recruiter screen, hiring manager chat (more of a two-way conversation than a screen), 45-min technical deep-dive, then a full-day onsite. Four onsite rounds: technical execution, people management/leadership, cross-functional influence, and a values round with a senior leader.
Technical round: They still want to see that you can engage with the code. I got a system design prompt and was expected to go deep. Not just draw boxes, actually defend your choices, discuss consistency models, talk about where AI inference fits into the system. Know your fundamentals. If you've been pure people-manager for 3+ years and haven't touched a design problem, prep hard here.
People management round: Classic behavioral format but with OpenAI-specific framing. Calibration of talent in a fast-moving environment. How do you manage engineers who may know more than you about the domain. How do you handle disagreement about technical direction. Have specific examples ready, they'll probe multiple levels deep.
Cross-functional round: How do you work with research, product, policy. OpenAI has an unusual cross-functional surface area given they interface with policy/safety teams constantly. This came through in questions.
Values round: The most substantive values conversation I've had in any interview cycle. Be ready to talk seriously about AI risk, safety practices, deployment decisions. Not a gotcha, but they want to understand how you think about these things, not whether you parrot the right buzzwords.
For comp at the EM level in SF: if you're leveled as a senior EM, think top-of-market plus equity that could matter a lot depending on the exit. Do your homework.