Course 4 · Unit 3

AI agents

Topics covered

Tool calling and structured outputs 90 min
The model doesn't execute anything; it proposes calls. You define the tools, validate their arguments and return results or errors it can understand.
The agent loop 65 min
An agent is a while loop with a model inside. Calling the model, executing tools, returning results and deciding when to stop.
Memory and context management 85 min
Context is a budget. A sliding window that doesn't break the structure, summarising when the token limit is exceeded and long-term memory retrieved by similarity.
Workflows, orchestration and MCP 90 min
Not everything has to be an autonomous agent. Task graphs with topological execution, fan-out/fan-in, orchestrator-workers and how MCP standardises access to tools.
Agent evaluation and security 95 min
Measuring an agent means measuring outcome, trajectory, cost and latency. Protecting it means assuming that what its tools read may be hostile and limiting what it can do.
Multi-agent systems — coordinating without losing control 85 min
Several agents can research in parallel, isolate contexts and review each other, but they can also get stuck in loops, multiply the cost and propagate errors. You learn the patterns, their failure modes and how to build an orchestrator with budgets and stopping conditions.

Before this unit it helps to have done:

The full explanations, auto-graded exercises, exams and projects are inside the academy.

Every lesson you complete gives you 10 yang, the academy's currency, and every unit exam you pass, 50.

Create your account and activate your subscription: you get the whole syllabus from day one.