What you will learn.
Agents are where the field is right now — and also where most of the bad code is. The gap between an agent that demos well and an agent that runs reliably in front of real users is, in our experience, the widest gap in modern AI engineering. This course is about closing it.
We spend seven weeks on agentic systems as engineering objects: planning architectures, the tradeoffs between linear, tree-search, and react-style control flow, tool design for long-horizon tasks, memory and state management, sandboxing and permission models, cost and latency control for multi-step systems, and — critically — how to debug an agent that does the wrong thing on step seven of a forty-step trajectory. Your capstone is an agent that performs a real task end-to-end with a human-in-the-loop boundary you design and defend.
This is our most demanding course. Students should arrive having shipped at least one LLM feature in production. The pace is fast and the reading list is non-trivial.