Anthropic has announced the launch of dynamic workflows for the Claude management agent, capable of coordinating up to 1000 AI agents in parallel during a single execution. This mode involves a primary agent planning and assigning subtasks, then aggregating results from sub-agents, making it suitable for scenarios such as breaking down large tasks for parallel processing. Relevant tests show that in vulnerability detection within a 116,000-line codebase, the dynamic workflows can stably identify 66 hidden vulnerabilities, far exceeding the 14 to 27 detected by a single agent.
