CS329Z: Engineering AI Agents is a remarkably practical new course from Stanford. It’s entirely free, and the lectures are being published online as the course unfolds. I watched the two lectures released so far, which walk through building an agent harness from scratch. From the course design to the assignments, it’s a model of what a responsible university computer science program can do.

And yet, it left me with a deep chill and a sense of helplessness. From prompting to context engineering to harnesses, we all know that each stage seems to have a shelf life of about a year. Before the final exam even arrives, these concepts and techniques may already be obsolete. But if universities don’t teach even these, what should they teach?

When I studied computer science, learning operating systems, advanced algorithms, and data structures felt like learning to make fire by rubbing sticks together, only to enter the workforce and find that all I needed every day was an electronic lighter.

Now everyone is learning to use the electronic lighter. By the time they enter the job market, they may find that meals are served at the push of a button, with no need to light a fire at all. People might still show you some respect for knowing how to make fire with sticks. But tell them you know how to use an electronic lighter, and they’ll just shrug.

What should the role of the university be in this era?