Multi-Agent Systems | Beyond the Prompt

Lesson 16: Multi-Agent Systems

Blackboard

A team of agents, not one super-agent

10 min

Concept 1

Why multiple agents?

Two main reasons: parallelisation (independent tasks run simultaneously) and reusability (a design agent can serve both the marketing team and the product team). Debugging is also easier — you fix one specialised agent instead of untangling a monolithic system.

Concept 2

The smart home example

Climate control, lighting, security, energy management, grocery tracking, and an orchestrator. The security agent determines who enters and sets permissions. The climate agent reads weather APIs and controls temperature. The orchestrator routes user requests to the right specialist.

Concept 3

Agent communication

Agents communicate with each other using the same protocols they use with tools — including MCP. The climate agent might ask the energy agent about current usage before adjusting the thermostat. Each agent is treated like a tool by the others.

Concept 4

Hierarchy and orchestration

Most multi-agent systems have a hierarchy. The orchestrator agent handles user communication and delegates to specialists. Some systems allow all-to-all communication between certain agents (like climate and energy) while keeping others isolated. The structure depends on the task.

Pause and Think

Think about a hospital. Which departments could be handled by specialised AI agents, and which ones would need to share information with each other?

Flashcards

Key Terms

Click or press Enter to flip a card. Review these before the practice question.

Additional Resources

Paper — Creativity in Multi-Agent Systems (EMNLP)
EMNLP 2025 survey on what happens when you let multiple LLM agents collaborate creatively — divergent exploration, iterative refinement, that kind of thing. More academic than practical, but interesting.
Paper — Multi-Agent Architectures
Survey of multi-agent architectures — communication protocols, role assignment, coordination. Gives the theoretical backdrop to the orchestration patterns covered in this lesson.
Glossary — Multi-agent system
A short definition.
LangChain vs LlamaIndex vs LangGraph
The main frameworks for building these systems, compared.

Practice Question

In a smart home multi-agent system, the climate agent and the energy agent need to coordinate. What is the simplest way to enable this?

Does not count for certificate

Choose one answer, use the hint if needed, then check your answer.

Use the hint before checking if you need a nudge.

Choose one answer, then click Check Answer.