Did the full Anthropic loop earlier this year for a research engineer position, mid-senior level. Want to give an honest breakdown because a lot of what I read online was either vague or clearly outdated.
Rounds in order: Recruiter screen (30 min, standard) Technical phone screen with a current RE: Python, some ML fundamentals, one small coding problem Take-home project: they gave me ~5 days to complete a short research task. This was the most unusual part. Not a LeetCode-style thing. An actual mini problem where you write up an approach, implement something, and explain your reasoning. Onsite (virtual): two engineering rounds, one systems design, one "research taste" conversation Values/behavioral loop with two different interviewers
The take-home was the thing that surprised me most. It's not just a coding screen, it's a taste-of-work thing. They want to see how you think through ambiguity, not just execution. I spent more time on the writeup than the code and I think that was right.
The behavioral stuff was genuine. Not "tell me about a time you failed" boilerplate. More like: "how do you think about a situation where your technical judgment conflicts with what your team wants to ship?" They probed for actual positions, not platitudes.
Total timeline was 7 weeks start to finish. Got an offer. Joined. Happy to answer specifics.