Single agent or multiple agents: test the handoff

What to inspect

Start from the work that must be done. A separate agent may help when it owns a distinct task with a clear interface. Compare the same task with and without the split, including missing context, validation effort, cost and recovery behavior.

Worked example

A research agent passes a summary to a writing agent but omits the source dates. The writer can produce a polished answer using stale material. A handoff contract should carry source references and constraints alongside prose.

Copy the full prompt

Adapt the inputs to your task. Remove private data before sending anything to a model. This page copies text locally; it does not run the prompt.

Evaluate whether [WORKFLOW] needs multiple agents. Define each proposed responsibility and the exact input/output contract at each handoff. List context that could be lost. Compare a single-agent baseline with the split using the same acceptance criteria. Leave cost and quality as unknown until measured. Include a failed-handoff recovery path.

Review checklist

1. Define a single-agent baseline.
2. Specify each handoff contract.
3. Carry evidence and constraints with the summary.
4. Measure coordination and recovery overhead.

Keep the resource

Keep exploring

• LLM evaluation: measure the usable result
• AI evaluation cases: turn a failure into a regression test

Documentation & scope

Anthropic: building effective agents ↗

This is an original workflow template with an illustrative example. It does not report completed model tests or measured performance. Provider APIs can change; check the linked documentation for your exact integration.