Did my onsite six weeks ago. Going to write the most concrete account I can because the existing threads are too vague to actually prep from.
The onsite is fully remote. Five rounds back to back with a 30-minute break in the middle. Here's the breakdown:
Round 1: Coding (45 min) Algorithm-focused. Medium-hard. Similar to what I described in earlier posts from others: real problem, not textbook. I implemented a solution in Python and we spent 20 minutes after on time/space complexity and a follow-up variant.
Round 2: System Design (60 min) Two interviewers. Design prompt related to building or scaling some form of ML serving infrastructure. Very relevant to Anthropic's actual work. See my other post on this round.
Break (30 min). Actually take it. Do not prep. I made the mistake of skimming notes and came back too anxious.
Round 3: Coding (45 min) More implementation-focused, less pure algorithm. Build a working thing that meets a spec. Clean code mattered here.
Round 4: Behavioral (45 min) Covered: a time I disagreed with a direction and what I did; how I've handled ambiguous or high-stakes decisions; something I changed my mind on after getting new information. Standard Anthropic values territory.
Round 5: Research / Values (30 min) This one is distinctive. It was less structured. One very senior person asked me questions about how I think about the field broadly. Not testing knowledge. Testing intellectual engagement. I talked about tradeoffs in AI deployment timelines and they engaged seriously with what I said rather than just listening politely.
Total time: about 5.5 hours including break.
Debriefs took about two weeks. I got an offer. Total comp came in around $410k TC at senior level in SF, majority equity. Base was $195k.
Ask me anything.