Anthropic Enables Claude to Coordinate 1,000 AI Agents Simultaneously

Anthropic now offers dynamic workflows for its Managed Agents Claude, capable of orchestrating up to 1,000 agents in parallel. The company highlights an internal bug detection test while warning of costly token usage and advising users to evaluate on their own workloads. Activation steps are specified on the platform.

Token Cost and Recommended Caution

Anthropic indicates that using these dynamic workflows can lead to significant token consumption and suggests starting with limited deployments. A senior executive at OpenAI recently described swarms of agents as a major source of token waste. The observed benefits are not guaranteed for all types of tasks, leading to the recommendation to validate these methods on one's own workloads.

Up to 1,000 Agents Orchestrated by a Main Agent

Dynamic workflows allow for the execution of up to 1,000 agents in parallel. A main agent formulates a plan, distributes tasks among sub-agents, and merges their results. This feature builds on an existing infrastructure of managed agents, but the dynamic workflows are new. In an internal test, Anthropic concealed 70 bugs in a codebase of 116,000 lines: a single agent detected between 14 and 27 per execution, while the dynamic workflow consistently found 66.

Activating Multi-Agent Mode on the Claude Platform

To activate dynamic workflows, one must select the agent type "multiagent_20261001." The startup can be done by consulting the documentation or executing the command "/claude-api managed-agents-onboard" in Claude Code.