For system design, budget out loud

An infra design round at Cursor turns on real constraints. Put numbers on it: a p99 latency budget per hop, the cost delta of active-active versus active-passive, what fails when a region goes dark and how you'd autoscale on the right signal. A latency-obsessed team wants to hear you think in budgets and error budgets.

In the past-work deep-dive, lead with a decision you'd defend and the tradeoff behind it. "We chose Karpenter over Cluster Autoscaler because our pod churn was bursty and bin-packing latency was hurting cold-start cost" shows ownership and judgment in one sentence. Save the architecture tour for after the interviewer asks.