Agent swarm
A mostly marketing name for many AI agents working in parallel on one task; in practice usually a coordinated multi-agent system.
Agent swarm is a name vendors use for many AI agents working on one task at the same time. The word suggests a leaderless group that organizes itself, like ants or bees. What sits behind it in production is usually a multi-agent system with a coordinator: an orchestrator or a script that splits the work, runs agents in parallel and merges the results.
OpenAI published an experimental, educational library called Swarm, built around agents and handoffs; its README now says it has been replaced by the OpenAI Agents SDK. A truly decentralized swarm, where agents negotiate work with no central control, is rare in production because it is hard to test, hard to debug and hard to bound in cost. Parallel agents multiply tokens, and without a coordinator nobody owns the final answer.
When a vendor says “swarm”, ask a few plain questions. How many agents run on a typical task, and who decides that? What coordinates them, and which step checks the merged result? What is the cost per task, and is there a cap? Has the system been evaluated against a single agent on the same tasks? Can you see a trace of what each agent did? If the answers describe a fixed pipeline, the honest name is an agentic workflow.
Sources
- GitHub: openai/swarm github.com
Related terms
- Multi-agent system (multi-agent orchestration): Several AI agents with separate roles and contexts working on one task, usually coordinated by a lead agent or a fixed workflow.
- Orchestrator-workers pattern (supervisor): One model plans the task, hands parts of it to worker models or subagents, and combines their results.
- Agentic workflow (workflow vs agent): A process where your code fixes the steps and a model does some of them, as opposed to an agent that decides its own steps.
- Agent handoff: One agent passes control of a conversation, together with its context, to a specialist agent or to a person.
- Cost per task (agent unit economics): What one finished agent task really costs: tokens, retries, tool calls and the human time to review it, compared with doing it another way.
- Dynamic workflow: An orchestration script the agent writes at run time. A runtime executes it, starting many subagents in phases and combining their results in code.
- Agent team: A lead agent coordinating several long-running agent sessions that share a task list and message each other directly.
- Evals: A repeatable test set that scores an AI feature, so changes can be measured instead of guessed.
