[OpenAI] The SWE loop is longer than you expect: 8 rounds, ~10 weeks
For anyone in the OpenAI pipeline: budget the time and energy.
Mine ran: recruiter → a few-hour take-home (real-codebase flavored, not LeetCode) → two same-day phone rounds (coding + design) → onsite (coding, system design on Excalidraw, a deep-dive on my own past project, behavioral).
The coding bar felt closer to a research lab than standard SWE: speed and correctness while they question you. Biggest wildcard was interviewer variability; two coding rounds felt like different difficulty universes.
How much do you all think the take-home actually weighs vs the onsite?
Comments (3)
Sign in to comment.
From what I saw the take-home is mostly a filter; it gets you to the onsite but the onsite is where the decision really happens. Don't over-optimize it, just don't fail it.
10 weeks is wild. Mine was ~7 but the team-matching at the end added two more on its own. The opaque matching phase is the part nobody warns you about.
The matching phase got me too. Cleared the loop then sat for two weeks with no signal. Apparently normal.