Anthropic adds dynamic workflows to Claude Managed Agents, enabling up to 1,000 parallel sub-agents per execution

Anthropic has added a "dynamic workflows" feature to Claude Managed Agents. A lead agent plans a job, hands pieces to sub-agents, and merges their results, with up to 1,000 agents running in parallel per execution. Developers switch it on by selecting the "multiagent_20261001" agent type, and can start via the docs or the "/claude-api managed-agents-onboard" command in Claude Code. Anthropic's own test hid 70 bugs in a 116,000-line codebase: a single agent found 14 to 27 per run, while the dynamic workflow consistently found 66.

Developers and enterprises already building on Managed Agents can now try splitting large jobs, such as code review or broad analysis, across many agents instead of one. The decision it enables is whether extra parallel compute buys enough accuracy to justify the cost for a given task. It also puts pressure on rivals such as OpenAI to show comparable orchestration. The impact is real but unproven beyond one vendor-run benchmark, and it is most relevant to teams with big, decomposable tasks.

The 66-of-70 result is Anthropic's own single test, so it says little about other codebases or task types. The "1,000 agents" figure is a ceiling, not typical use, and the coverage gives no token cost, latency or pricing. The report is a secondary source that cites Anthropic's announcement without links, so the details are unverified here.

1. 01Independent benchmarks or customer reports comparing dynamic workflows with a single agent on accuracy and total token cost, especially on tasks other than bug finding.
2. 02Anthropic's published pricing, rate limits and any concurrency caps for the multiagent_20261001 type, including how many agents real users actually run at once.
3. 03Whether OpenAI, Google or other rivals ship comparable parallel orchestration within the next few months, and whether the feature moves from beta to general availability.