Multi-Agent WorkflowL07 · Module I · Lesson 01 of 1
Lab · 90 min

Lab: Build a multi-agent workflow

Three agents, one task, one measured outcome.

Summary

Three agents, one task, one measured outcome. The lab that teaches whether multi-agent coordination is worth the coordination cost — and when a single agent is faster, cheaper, and more reliable.

Objectives
  • 01Design a task that plausibly benefits from multiple agents.
  • 02Implement three agents with defined roles and handoffs.
  • 03Instrument the coordination — who calls whom, why, how often.
  • 04Compare end-to-end against a single-agent baseline.
Key Ideas
  • Multi-agent is a design choice, not a virtue; the baseline is the test.
  • Every coordination message is a cost; count them.
Lab

Choose a task with defensible role separation (e.g., research → draft → critique; or plan → execute → verify). Implement three specialized agents with distinct tools and prompts. Ship a working orchestrator. Instrument every message between agents. Compare on a fixed test set against a single strong agent baseline. Report cost, latency, and quality. Write a decision memo: does multi-agent win here, or is the single-agent baseline strictly better?

Deliverables

  • Multi-agent system with three specialized agents.
  • Single-agent baseline for the same task.
  • Instrumentation of coordination messages and costs.
  • Comparison report with a defended recommendation.

Rubric

  • Role separation — the three agents do meaningfully different work.
  • Baseline — an honest single-agent comparison exists.
  • Instrumentation — you can answer 'where did coordination fail?'.
  • Judgment — the recommendation is defended, not asserted.