Back to catalog / Agent workflows
Plan research, then parallelize fact verification
Have a coordinator decompose a question, bounded workers verify authoritative sources, and synthesize one report.
Goal
Goal: Have a coordinator decompose a question, bounded workers verify authoritative sources, and synthesize one report.
Steps
- Give the coordinator a research question, output schema, and acceptance checklist; expose only
web_searchandweb_fetchto workers. [plan] - Split the question into non-overlapping fact checks; require each worker to return structured findings with URLs and evidence. [plan]
- Have the coordinator deduplicate, resolve conflicts, and synthesize; compare worker token/tool usage and cost with a single-agent baseline. [plan]
How to verify
Confirm each claim is traceable, subtopics cover the scope without duplication, and quality/cost is compared with the single-agent baseline.
Troubleshooting
If workers duplicate searches or conflict, narrow tasks and have the coordinator recheck primary sources; cap parallelism by budget.
Permissions
Prerequisites
- Python 3.11+, Anthropic API key, Managed Agents beta SDK, and a topic authorized for web research.
Permissions
- Workers may access only approved web sources; a human must verify citations and freshness.
Limitations
More workers are not always cheaper; model choice, task budgets, and current prices affect the tradeoff.
Pinned upstream source
- Repository
- anthropics/anthropic-cookbook
- File path
- managed_agents/CMA_plan_big_execute_small.ipynb
- Pinned commit
9c1e493ddd2b82df399d2a61b9c806c539e18be6- Source checked
- 2026-10-14 · Reviewed against the pinned upstream source
Repository metrics
These values describe the upstream repository, not popularity or likes for an individual skill or playbook.
- Stars
- Not collected
- Forks
- Not collected
- Observed at
- Not collected
No metrics snapshot is available yet.
License: MIT
Main topic
Turn ecommerce campaign materials into a reviewable weekly brief