Generating a patch is only one step in delivering working software. The useful lens for this curriculum is the feedback loop around the agent: clear specifications, a navigable repository, appropriate tools, and checks that make mistakes visible.
What Sets It Apart
- Building a small coding agent exposes its tool loop and system prompts, helping developers reason about failures beyond rewriting a prompt.
- The 2026 syllabus connects context engineering, MCP, skills, and repository readiness with review and security: agent customization becomes part of engineering quality.
- Open source contributions account for 30% of the 2026 grade and the final project for 50%, putting substantial weight on work that must survive integration.
Who Should Take It
A strong fit for programmers who want to supervise agents on existing codebases. It assumes software engineering foundations; the FAQ estimates 10–12 hours weekly, and some tools require subscriptions. Public materials support independent study, while auditing requests are open to Stanford students and staff.
