AI agents · Types
What is a manager agent?
The short answer
One agent receives the work, decides while the work is under way which specialist handles each part, and assembles the result. The specialists never deal with the user.
It is one of the 4 forms of multi-agent system, with the agent pipeline and the agent team and the guardian agent.
| What starts it | Any of the above |
|---|---|
| Who decides the next step | One agent, while the work is under way |
When should you use a manager agent?
Use it when requests vary widely and you want one point of control and one voice.
When should you not?
Not when the order is always the same. The manager then adds cost and one more thing that can fail.
What does it need to run safely?
It needs a limit on how often it may delegate, a clear description of each specialist, and a record of which agent did what.
An example
A network operations manager receives an alarm, hands diagnosis, customer impact and ticketing to three specialists, and writes one incident note.
What else is it called?
Orchestrator, coordinator, supervisor, hierarchical agents.
How do you build one on Katonic?
- New agent, Answer & assist, then Answer questions. Start blank and fill Identity and Behavior. In the system prompt, say which agent to use for what.
- Tools, scroll to the Agents card. The page describes it: "Agents this one can call as tools - it invokes them within a turn and composes their results. Different from Coordinate a team, which hands the whole turn to one specialist."
- Attach the agents it may call. The picker lists chat agents and workflows.
- Advanced. Leave the execution strategy on Tool-calling unless you have a reason.
- Test with a request that needs both agents, and read the trace to see which was called.
- Deploy, then Publish.
Questions people ask
What starts a manager agent?
Any of the above.
Who decides its next step?
One agent, while the work is under way.
What is a multi-agent system?
A multi-agent system is several agents, each with its own instructions and tools, working on one piece of work. The forms differ in who decides the order. OpenAI advises starting with one agent and moving to several only when needed. Microsoft's guidance is to start with several where the work crosses security boundaries or teams, and otherwise to test a single agent first. Gartner expects 70 percent of AI applications to use multi-agent systems by 2028.