What is Supervisor Agent?
Also called orchestrator agent, router agent.
A supervisor agent is an agent whose job is to coordinate other agents rather than to do the work itself. It interprets the goal, decides which specialist should act next, passes along the necessary context, and decides when the overall task is complete. It is the most widely used arrangement in multi-agent systems.
In the usual implementation, specialists are exposed to the supervisor as callable tools or as named routes. After each specialist returns, control comes back to the supervisor, which reads the result and chooses the next move. That return path is what distinguishes the pattern from a chain, and it is what allows the supervisor to react when a specialist fails or returns something unusable.
The supervisor holds the overall goal, the running summary, and the completion criteria. Specialists hold only their own instructions and the brief they were given. Keeping the shared picture in one place avoids the situation where every agent carries a slightly different understanding of the objective, which is how multi-agent runs drift apart.
The pattern's weakness is that the supervisor becomes a bottleneck and a single point of failure. Every result passes through it, its context grows with each round, and a mistaken judgment about completion ends the run early. Mitigations include summarizing specialist output before it enters the supervisor's context and setting explicit, checkable completion criteria rather than leaving them to judgment.
Alternatives exist. A fixed pipeline removes the supervisor entirely when the order is known. A peer network lets agents pass work directly, trading control for flexibility. The supervisor should not be confused with orchestration infrastructure: the supervisor reasons about what should happen next, while the orchestrator reliably executes and records it.
Key points
- Coordinates specialists and holds the overall goal and criteria.
- Control returns to the supervisor after every specialist step.
- Becomes a bottleneck, and its context grows each round.
- A reasoning coordinator, distinct from orchestration infrastructure.
In practice
A hiring assistant receives a request to summarize a candidate pool. The supervisor asks a parsing agent to extract structured fields from twelve resumes, then asks a scoring agent to rank them against the role criteria, then asks a writing agent for a one-page summary. It reads each result, notices two resumes failed to parse, and sends those back before continuing.
Supervisor Agent FAQs
What does a supervisor agent do?
A supervisor agent coordinates specialists instead of completing every task itself. It interprets the overall goal, chooses who should act next, sends the needed context, reviews results, and decides when the work is complete.
How does a supervisor agent coordinate other agents?
Specialists are exposed as callable tools or named routes. After each specialist returns, the supervisor reads the result, updates its understanding of the goal, and selects the next action or sends work back for correction.
What are the limits of a supervisor agent pattern?
The supervisor can become a bottleneck because every result passes through it and its working context grows over time. Clear completion criteria, compact summaries, and bounded specialist choices reduce those risks.
What is the difference between a supervisor agent and orchestration?
A supervisor agent makes reasoning decisions about the next step. Orchestration infrastructure executes and records those decisions reliably, including state persistence, retries, and operational limits.