Orchestration
Orchestration is the coordination layer of a multi-agent system: one orchestrator receives a goal, splits it into tasks, assigns them to specialised agents, collects their outputs, resolves conflicts and assembles the final result.
No jargon, no math
In plain words, what is it?
One agent doing everything is a generalist with too many tabs open. Orchestration is how you run a team instead. An orchestrator receives the goal, splits it into subtasks (research this, price that, draft the other) and routes each to the agent built for it: a data agent, a drafting agent, a checking agent. As results come back, the orchestrator compares them, spots contradictions, sends work back for another pass, and only then assembles the answer. It also owns the plumbing: retries, timeouts, budgets. Where it breaks: if the split is wrong, every agent executes the wrong task perfectly, so the quality of the decomposition is the quality of the whole system.
- 01
Receive. The orchestrator takes the goal and the budget for the run.
- 02
Split. It decomposes the goal into subtasks matched to specialised agents.
- 03
Dispatch. Subtasks run in parallel, each agent with its own tools and data.
- 04
Reconcile. The orchestrator compares outputs and sends contradictions back for rework.
- 05
Assemble. Consistent outputs merge into one final result.
Terms that live nearby.
Multi-Agent System
A multi-agent system is a team of specialised AI agents working on one goal: each agent has its own role and tools, and a coordinator divides the work, collects the results and resolves conflicts, instead of one model trying to do everything alone.
ReadAgentic Workflow
An agentic workflow is a business process whose steps are executed by AI agents: they plan the work, call tools, read results and decide the next move, instead of following a fixed script that breaks the first time reality differs.
ReadPlanner
A planner is the component of an AI agent that turns a goal into an ordered sequence of steps before any action is taken, deciding what to do in which order instead of improvising one move at a time.
Read§ 04
Want this running in your stack?
We design and ship AI systems with permissions, logs and human approval. It starts with a 30-minute scoping call.