Stanford's Computer Science Department has introduced a new course, CS329A, on self-improving AI agents. It will be offered in Fall 2025 as a graduate seminar.

The curriculum is quite rigorous: self-improvement techniques, large language models, multi-step reasoning and planning, and evaluation framework construction. Cutting-edge research like AB-MCTS and The AI Scientist are included in the syllabus. The two main instructors, Aakanksha Chowdhery and Azalia Mirhoseini, are well-known researchers in the field.

Five years ago, building AI agents required piecing together blog posts and trial-and-error. Now, there’s a systematic course—much more convenient.

The course website is cs329a.stanford.edu, with a striking red background.