The Latency Optimisation Prompt

Reduce this agent's latency.

Current profile: [PASTE MEASUREMENTS, OR DESCRIBE]
Target: [DESCRIBE]
What the user experiences during the wait: [DESCRIBE]

Produce:
1. Where time is spent: model inference, tool calls, network,
   serialisation, queueing
2. What is on the critical path
3. What could run in parallel that runs serially today
4. What could start speculatively before it is known to be needed
5. What could be precomputed
6. Round trips that could be eliminated by composing tools
7. Where a faster model at lower quality is an acceptable trade
8. What could be moved out of the request path entirely

Then, separately from making it faster:
9. What could make the wait feel shorter — streaming, progress
   updates, partial results, optimistic UI
10. What the user should see at each stage

Perceived latency and actual latency are different problems. Give
me both.