advanced · agents

The Twenty-Second Agent

An AI agent takes 20 seconds per run and users are complaining. How do you make it faster?

Practical scenario

A support-triage agent averages 20 s per run. A trace shows five sequential model calls, three tool calls, and a context that grows from 2k to 14k tokens across the run.

What it tests

  • Whether the candidate traces before tuning
  • Understanding that step count usually dominates
  • Awareness of the latency/cost/quality trade

Lessons behind this question