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.
No jargon, no math
In plain words, what is it?
You decompose the goal into roles the way you would staff a project: a researcher that finds data, a writer that drafts, a reviewer that checks. Each role is an agent with its own instructions and tools. A coordinator (another agent, or an orchestration layer) splits the task, hands each piece to the right agent, collects the outputs and sends anything doubtful to review. Simple tasks stay single-agent; multi-agent pays off when the work has independent pieces that can run in parallel and a final step that needs cross-checking. It breaks three ways: coordination overhead (three agents can be slower than one good one), error cascades (one agent's mistake becomes the next agent's input), and cost (every agent burns tokens on every step).
- 01
Split. A planner breaks the goal into independent sub-tasks.
- 02
Assign. Each sub-task goes to the agent with the right role and tools.
- 03
Execute. Agents work in parallel, each on its piece.
- 04
Review. Outputs are checked against sources and against each other.
- 05
Assemble. The coordinator merges the verified pieces into the final result.
Terms that live nearby.
AI Agent
An AI agent is software that pursues a goal on your behalf: it reads the situation, plans the next step, acts through tools (email, calendars, databases, APIs) and checks the result before continuing, instead of answering one question and stopping.
ReadOrchestration
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.
ReadAgent Loop
The agent loop is the cycle an AI agent repeats while pursuing a goal: observe the current state, plan the next action, act through a tool, check the result, then continue, re-plan or stop.
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.