What is Multi-Agent Team?
Also called agent crew, multi-agent system.
A multi-agent team is an arrangement in which several agents with different roles work on related tasks, passing work and context between them instead of one agent doing everything. Roles are typically split by function, such as research, drafting, and checking, or by domain. Coordination is handled either by a designated lead agent or by a predefined sequence.
The motivation is focus. An agent with one job, a short set of instructions, and a small tool list behaves more predictably than one holding every procedure at once. Splitting also creates natural review points, because a separate checking agent evaluates output it did not produce, which catches a different class of error than asking one agent to review itself.
Two arrangements are common. A pipeline hands work from each agent to the next in a fixed order. A supervisor pattern has a lead agent decide which specialist handles each request. Pipelines are easier to debug because the path is known in advance. Supervisor arrangements absorb more varied input and are correspondingly harder to reason about when something goes wrong.
Multi-agent setups add failure modes a single agent does not have: context dropped at a handoff, two agents writing to the same record, loops where work is passed back and forth, and cost multiplying with each additional step. Many problems solved by adding agents are solved more reliably with one agent and a better written procedure.
Multi-agent team is the technical arrangement, and AI workforce is the organizational framing usually placed on top of it. A team lead agent is the coordinator in the supervisor pattern. Handoff points are where standard operating procedures matter most, since that is where context is most often lost.
Key points
- Several agents with distinct roles passing work between them
- Coordinated by a lead agent or by a fixed sequence
- Smaller scope per agent makes behavior more predictable
- Adds handoff, looping, and cost failure modes
In practice
A newsletter operation runs three agents. The first collects and summarizes the week's sources, the second drafts the issue in the publication's established voice, and the third checks every claim against the collected sources and flags anything unsupported. The editor reads the flags first, then the draft. When the checker and the writer disagree, the item is cut rather than published.