The Cost & Latency Profiler

Help me profile this agent's cost and latency.

Agent: [DESCRIBE]
Typical run: [DESCRIBE THE STEPS]
Current measurements, if any: [PASTE]

Produce:
1. Where cost accumulates: which steps, what proportion
2. Where latency accumulates
3. What scales with run length and what is fixed
4. Which tool calls are on the critical path
5. What could run in parallel that currently runs serially
6. What is being recomputed that could be cached
7. What context is being resent that could be cached at the
   provider level
8. Where a smaller model would be sufficient
9. The single change with the best cost-to-effort ratio

Then:
10. What should I measure that I currently do not?
11. What is a reasonable target for this workload, and how far off
    am I?

Do not state specific model prices — I will look those up. Work in
relative terms and tell me what to measure.