Claude Managed Agents Add Dynamic Workflows for Up to 1,000 Parallel AI Agents

Anthropic is adding dynamic workflows to Claude Managed Agents, bringing multi-agent orchestration to the platform. A lead agent can create a plan, distribute tasks to sub-agents, and merge their results when the work is complete. The system supports up to 1,000 agents running in parallel per execution.

Anthropic’s testing focused on a codebase with 70 hidden bugs across 116,000 lines. According to the article, a single agent found between 14 and 27 bugs per run, while the dynamic workflow consistently found 66. That suggests a strong result for this specific bug-finding task, but the article notes that performance across other task types still needs to be tested.

Large multi-agent workflows can consume many tokens, so cost-effectiveness remains a key issue for teams evaluating the feature. The article notes that a senior OpenAI engineer recently criticized agent swarms as wasteful, while Anthropic points to its internal results as evidence of value. The practical takeaway: test dynamic workflows on your own workloads before assuming the gains will transfer.

To activate dynamic workflows, users can select the multiagent_20261001 agent type. Anthropic recommends starting small because these workflows can use a lot of tokens. The article also says users can get started through the documentation or by running /claude-api managed-agents-onboard in Claude Code.