Why does the lesson forbid displaying an active 'Approve and Send' button the instant streamed tokens finish rendering in the Helpdesk Copilot UI?
Day 23: Reduce Latency Without Hiding Work
Reduce Latency Without Hiding Work
Loading lesson content...
In the retrieval phase of Helpdesk Copilot, an engineer refactors sequential context gathering to concurrent execution as shown above. What is the primary latency benefit of this refactoring?
A performance audit of Helpdesk Copilot shows that p95 end-to-end latency is 14.8 seconds, far exceeding the 7.0-second budget target. Profiling reveals that a secondary LLM call used as a 'safety judge' consumes 3.6 seconds, while generating 650 output tokens with Claude 3.5 Sonnet consumes another 5.8 seconds. Which optimization combination preserves all safety checks while meeting the latency budget?