1 min lesson
Streaming and backpressure
Use "Under a burst, something has to give" to describe the practical rule.
Step 1 of 2
Streaming and backpressurethe editor is watching
Users expect tokens to appear as the model produces them. That means a long-lived connection per active run and a plan for when the consumer is slower than the producer. If the client can't keep up, buffer with a bound and shed or pause rather than blowing memory on the server.
Under a burst, something has to give. Decide explicitly: queue depth limits that reject new runs early with a clear signal, bounded streaming buffers that pause the producer and admission control at intake. Saying "we'd add more workers" without a saturation plan is the answer that fails the screen.
Open by naming the three constraints out loud - long-running, stateful, bursty - then say cost is a first-class SLO alongside latency and reliability. Most candidates optimize for speed and correctness and forget that every step is metered. Tying the design back to "this is Cursor's defining feature surface, so its reliability is strategic" shows you grasp why this team exists.
Learn more
Optional practice
Practice: Streaming and backpressure
QAn agent run fails on step 6 of 10 because the model provider returns a 429. Your current design restarts the whole run. What is the core fix and why does it matter for Cursor specifically?