Whether that scale pays off is contested. A senior OpenAI engineer has called agent swarms a massive waste of tokens, but Anthropic points to its own testing: the team hid 70 bugs in a 116,000-line codebase and found a single agent caught between 14 and 27 per run, while the dynamic workflow consistently found 66.

Those gains come from one bug-hunting test, so how they hold up across other task types is an open question. Anthropic itself warns the workflows can burn through a lot of tokens and recommends starting small. To try them, select the "multiagent_20261001" agent type, either via the documentation or by running "/claude-api managed-agents-onboard" in Claude Code.