Public course materials for a graduate course on building software with AI coding agents through specifications, repeatable workflows, review, and verification.
The course is still being developed: published materials may change, and files that explicitly identify themselves as drafts are not yet ready to assign without instructor review.
- Course overview — purpose, semester arc, and learning outcomes
- Weeks 1–3: Foundations — LLMs, agent architecture, prompting, context, and verification
- Software engineering for agentic development — classical software-engineering concepts applied to agent-based development (skeleton; content in development)
- Weeks 4–7: The Growing Tic-Tac-Toe — a staged project in specification-driven agentic development
- Technical concepts — topic index and further reading
- Prompting cheat sheet — compact prompting principles
Each unit contains instructor lecture outlines, student-facing lecture notes, Marp slide sources, assignments, and supporting student materials. The unit READMEs identify what is ready and how the pieces fit together.
The course exercises assume:
- Git and a command-line development environment
- Python 3.11 or newer for the foundations exercises and starter project
- Claude Code with an appropriate account for agent-based exercises
- Node.js and TypeScript for the Project 1 porting stage
Pandoc with XeLaTeX and Marp are optional; they are needed only to regenerate PDF handouts or rendered slide decks from the checked-in Markdown sources.
The named Tic-Tac-Toe, NautilusTRX, and lost-communities projects are historical course case studies. Required readings and exercises in this public edition use materials bundled here; the private prototype repositories are not prerequisites.
The example personal knowledge base intentionally contains one unresolved
/concepts/tool-schemas.md link. It demonstrates that Open Knowledge Format and
Obsidian vault-root links may point to knowledge that has not been written yet.
Course prose and media are licensed under CC BY 4.0; repository code and configuration are licensed under MIT. See LICENSING.md for the boundary, attribution, and third-party-material exceptions.