9.7 Multi-agent systems
You can decompose a task across agents without chaos.
Before:08. Large Language Models
Multi-agent systems decompose work across specialised agents — supervisor, hierarchical or peer patterns — with message passing and shared state as connective tissue, and handoffs as the place where context quietly dies. It sits in the advanced run because single-agent rigour comes first. The theatre warning is earned: adding agents multiplies coordination overhead and emergent misbehaviour, and many celebrated multi-agent systems are one good agent wearing several hats.
Work through these
Supervisor, hierarchical and peer patterns
Patterns for organising several agents: one directing others, a tree of responsibility, or peers negotiating. Choosing a pattern is the main architectural decision here.
Message passing and shared state
Agents coordinate either by sending messages or by reading and writing shared state, and the two have different failure modes. Shared state is simpler and harder to reason about at scale.
Handoffs and context loss between agents
When one agent hands work to another, the context does not transfer completely, and what is lost is usually what mattered. Designing the handover explicitly is what prevents it.
Model Context Protocol — Introduction · DocsDebugging emergent behaviour
Behaviour arising from interaction between agents is difficult to reproduce and therefore difficult to fix. Recording complete runs is the only practical way to debug it.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.