The Parent-Child Model

Agent teams follow a parent-child model. One agent — the parent — acts as the coordinator. It understands the full objective, breaks it into subtasks, and delegates each subtask to a specialized child agent. The children execute their work in isolated contexts and return results to the parent. The parent synthesizes everything into a final deliverable.

Think of it like a consulting engagement. A project manager receives a client request: "We need a competitive analysis of the AI agent market." The PM does not write the analysis alone. He assigns one analyst to research competitors, another to pull market data, and a third to draft the narrative. Each specialist works in their domain, and the PM assembles the final report. The PM's value is coordination and quality control, not individual contribution.

1. 1. Parent Receives the Objective The parent agent gets the high-level task from the user. Its first job is to understand the full scope and identify what subtasks are needed.
2. 2. Parent Decomposes the Task The parent breaks the objective into discrete, well-scoped subtasks. Each subtask should be achievable by a single specialist without needing to coordinate with other specialists.
3. 3. Parent Delegates to Children Each subtask is assigned to a child agent with a clear prompt: what to do, what format to return, and what constraints to follow. The child works in its own isolated context.
4. 4. Children Execute and Return Each child agent completes its subtask and returns results to the parent. The child does not know about other children or the broader objective — it only knows its own assignment.
5. 5. Parent Synthesizes The parent collects all results, resolves conflicts, fills gaps, and assembles the final output. This synthesis step is where the team produces more than the sum of its parts.