Back to Q&A board 
While other companies focus on "Product Sense" (can you build an app?) and "Execution" (can you move a metric?), Anthropic is essentially looking for a Safety-Conscious Systems Architect.
Instead of testing how you’d grow a user base, Anthropic tests how you’d govern a model. They care less about "Growth Loops" and much more about "Redlines," meaning they want PMs who can define exactly when a model becomes too dangerous to ship, even if it’s profitable.
The Technical "Why": You aren't just managing APIs; you’re managing Evals. They’ll grill you on how to measure "harmlessness" versus "helpfulness" without neutering the model’s intelligence.
Constitutional Intuition: They test your ability to translate human values into model constraints. It’s less about "User Experience" and more about "System Alignment."
Agentic Friction: With their focus on "Computer Use," they want to know if you can design the "Human-in-the-loop" guardrails that prevent an autonomous agent from going rogue on a user’s desktop.