Anthropic introduces dynamic workflows for Claude Managed Agents

Anthropic has introduced dynamic workflows for its Claude Managed Agents, enabling multi-agent orchestration. This new capability allows a lead AI agent to create a plan and distribute tasks to up to 1,000 sub-agents working in parallel, with the system automatically aggregating the final results.

Previously, Managed Agents relied on sequential delegation, which required the lead agent to manage every sub-task within its own context window. The new dynamic workflow offloads the orchestration logic to a backend process, allowing the lead agent to focus on high-level planning while the system handles task distribution and execution.

In a benchmark test involving a codebase of 116,000 lines containing 70 hidden errors, a single agent identified between 14 and 27 errors per run. In contrast, the multi-agent workflow consistently identified 66 errors.

Developers can enable this feature by setting the agent type to ‘multiagent_20261001’. Because these workflows consume significant amounts of tokens, experts recommend using them for naturally divisible tasks like code scanning or document reviews rather than highly interdependent serial processes. The feature is currently in beta and does not yet support zero data retention or HIPAA compliance.