How Stanford Teaches AI-Native Engineering — Mihail Eric, Monaco

Judgment

Mihail Eric’s decision for CS146S is to trade prompt-friendly exercises for maintainer-reviewed open-source patches, accepting slower, coordination-heavy grading because a maintainer—not the student or instructor—decides whether the work belongs.

Basis

Mihail Eric said the first CS146S run had nine assignments and a capstone. By the next run, he figured students could one-shot a lot of those exercises, and when he surveyed a room of Stanford sophomores and juniors, everyone was already using an agent. So part of the homework became contributing to an open-source project, with the repo maintainer reviewing the student’s patch. The mechanism is the constraint: an agent can implement a feature in an afternoon, but the student still has to learn whether the maintainer wants it, why the code behaves as it does, and what else the patch might break. That conversation is also how a junior engineer learns from a senior one. At Monaco, Eric said the team evaluated code review products and kept using Devin, but built its own agent workspaces because a background agent needs somewhere to run the product, reproduce a failure, and test the patch; his rule is “You buy to learn, and you build to keep.”