The onsite project - the decision round

Cursor's onsite is famously a multi-day, in-person project with the core team. For this role, expect a realistic User Operations / quality problem and almost no scaffolding. This is the round where the offer is decided.

Four things are being graded under realistic, ambiguous conditions: autonomy, product sense, debugging depth and craft. The format rewards behaving like a teammate who joined this week, not a candidate performing for an exam.

1. 1Scope ruthlessly. Ask sharp clarifying questions early, then narrow to the slice you can actually ship in the time box. Stating what you're cutting is itself a signal.
2. 2Reproduce and root-cause for real. If the problem is a bug or a quality gap, go deep enough to name the actual cause, not a surface symptom.
3. 3Ship something real. A working script, a triage taxonomy, a repro harness, a draft brief - a concrete artifact beats a deck of intentions.
4. 4Frame it through the feedback loop. Show how your work turns user signal into a product decision, not just a one-off fix. That is the charter.
5. 5Present a crisp recommendation. Close with what you did, what you found, the call you'd make and what you'd do next with more time.

• Treat the team's channel as a real Slack: ask good questions, share progress and don't go dark for hours.
• Use agent-assisted tools openly to move fast and show your judgment when you reject their output.
• Manage the time box: a finished smaller thing beats an unfinished bigger thing.