Multi-Agent Systems and Collaboration
For engineers who have built a single AI agent and now need several agents to work together reliably. You will design how agents communicate, divide work, delegate and agree, orchestrate them with LangGraph and task queues, and test, observe and evaluate the whole system in production.
Course Content
When splitting one agent into several pays off, and how agents talk (messages, events, queues) and coordinate (centralised or distributed). 3 lessons, about 40 minutes.
The three ways agents share work — supervisor and workers, delegation with handoff briefs, and negotiation or voting — with the guards that keep each one from looping or starving. 3 lessons, about 45 minutes.
The tools that run multi-agent systems in production: LangGraph for stateful agent graphs, RQ and Celery for queued agent work, and observability that measures the handoffs between agents. 3 lessons, about 40 minutes.
A five-agent research assistant built end to end, then tested with scripted models, guarded with safety boundaries, and evaluated for emergent behaviour, cost and coordination. 3 lessons, about 45 minutes.