They sent me an automated CodeSignal test before I ever spoke to a human, which consisted of 4 stages of the same problem with each stage getting progressively more complex. Did not progress beyond this stage.

The interview loop at Anthropic is completely different from the standard FAANG pipeline. They do not care about your ability to speed run algorithms. The entire process is built around "First Principles" thinking and writing extremely robust and safe code. After the recruiter screen, I had a deep dive pair programming session with an engineer. It felt much more like a collaborative work session than a test. They want to see how you handle edge cases and system failures in real time. I was asked to build a reliable message ingestion component that could handle streaming data with unpredictable latency spikes.