liveBook · Manning

Coordinator agent

As explained in Build a Multi-Agent System (from Scratch)

A coordinator LLM agent manages subagents or shared resources in a multi-agent system.

A coordinator agent assigns subtasks to specialized agents and combines their results into an overall task outcome. It may dispatch local subagents or remote agents, treating remote A2A agents as black boxes. section 1.6 section 10.2.3

A coordinator can also manage execution and memory:

  • Oversee subagent execution, with a human supervising the coordinator rather than each agent’s processing loop. section 8.5
  • Manage shared memory, while accounting for concurrent writes that could overwrite or omit episode records. section 7.6

Resuming a remote task

A coordinator sends a CrewAI Hailstone agent a task phrased as “starting at the number four.” When the remote agent requests a numerical starting value, the coordinator sends “4” and receives the completed sequence. section 10.2.3

Figure 10.11 The sequence of interactions for dispatching an A2A agent during a task execution

The interaction sequence shows how a coordinator dispatches an A2A agent during task execution. section 10.2.3

How Build a Multi-Agent System (from Scratch) covers it

How deep it goes
Traces task dispatch, remote input requests, and resumed execution, and discusses memory and oversight responsibilities.
Examples
Uses croissant and pizza subtasks, shared writer-reviewer memory, and a CrewAI Hailstone task that requests clarification.
What you can do after
Apply it by dispatching subtasks to local or remote agents and handling a remote agent’s request for more input.
How it connects
Connects multi-agent systems with subagents, A2A agents, shared memory, and human-in-the-loop oversight.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build a Multi-Agent System (from Scratch). The section links point to the parts of the book it draws on.