Stanford’s updated Modern Software Developer course now requires students to contribute to unfamiliar open-source projects and defend their patches to the people who maintain them. The change responds to a simple problem: coding agents can complete much of last year’s homework in one prompt.
Instructor Mihail Eric said every sophomore and junior he surveyed was already using a coding agent. Instead of grading isolated exercises, the course puts students in larger codebases where they must learn why the software works as it does, determine whether a proposed feature belongs, and explain what an AI-generated change could break.
The ten-week syllabus still covers how agent harnesses work, context engineering, tools, security, code review and background agents. Students build a basic coding agent and inspect the requests behind a production tool, but the course treats implementation speed as only one part of engineering. Human decisions about architecture, intent and acceptable risk remain central.
Eric applies a similar split at Monaco, where he leads AI. The company buys general tools such as code review software but builds agent workspaces around its own cloud and operational needs. The course’s new assignment makes the same point: generating a patch is increasingly cheap, while understanding the repository and earning a maintainer’s approval remain scarce skills.