Anthropic has introduced dynamic workflows to Claude Managed Agents, enabling the orchestration of up to 1,000 AI agents in parallel. This new infrastructure allows a lead agent to create a plan, distribute tasks to numerous sub-agents, and merge the results once tasks are completed.
In testing involving 70 hidden bugs within a 116,000-line codebase, Anthropic found that a single agent caught between 14 and 27 bugs per run, whereas the new dynamic workflow consistently identified 66 bugs. To activate these workflows, users can select the multiagent_20261001 agent type and begin via the documentation or Claude Code.
Because executing large-scale multi-agent workflows can consume a high volume of tokens, Anthropic advises users to start small and test with their own workloads. This capability significantly expands the platform's capacity for complex, automated parallel task execution.