The shelf life of an up-to-date course

I am quite impressed by how recent and practical Stanford’s CS 329Z, Engineering AI Agents, is. The course covers compound AI systems, retrieval, tool use, agent loops, evaluation, and the design choices that make these systems work. These are problems that people are dealing with in industry now.

This is a common problem with courses that are products of their time. Data science degrees are an obvious example. A curriculum can be respectable when it is created, yet contain methods and assumptions that already feel dated by the time students graduate. Fast-moving fields make this especially visible.

What I find impressive about Stanford here is the precision of the choices. The instructors are not trying to cover every fashionable word. They have chosen a set of ideas that fit together and connect to actual engineering work. That is a useful lesson in curriculum design.

It seems possible, then, to teach genuinely current material at undergraduate level, even in a field that changes this quickly. But there is an uncomfortable question underneath it. If some of these ideas are outdated in a few months, what exactly should students take away from them?

Perhaps the answer is not the individual framework or API. It is learning how to decompose a system, choose components, collect useful data, evaluate results, and recognise trade-offs. Those habits may survive longer than the tools. The course is still a snapshot, but it is a thoughtful one.