Industry News
The course materials for Stanford’s CS329Z course on Intelligent Agent Engineering have been made public. Andrew Ng, as well as the authors of SWE-bench, will also be participating in this initiative.
If you want to systematically learn how to create intelligent agents, you can access the course materials and reading list directly on the course website without the need to register.
The course is titled “Engineering AI Agents” and is taught by Diyi Yang, Michael Ryan, and John Yang. The curriculum begins with learning how to manually implement RAG systems, tool calls, and agent loops from scratch, and then moves on to exploring how frameworks like DSPy abstract these concepts. Subsequent topics include agent memory, multi-agent systems, and the reasons why they may fail. Finally, the course covers optimization techniques, data analysis, and evaluation methods.
The final section is of greatest value to those working on product development: how to determine whether an intelligent system is actually useful. The course places equal emphasis on questions such as “which problems should be chosen for testing,” “how to break down the system into components for evaluation,” and “how to develop appropriate assessment methods” – all these aspects are considered just as important as writing code itself.
The reading list is also quite good – it includes classics like “Attention Is All You Need”, as well as articles on contextual engineering from Anthropic and the book “AI Agents That Matter”.
On the schedule, on November 9th, there will be a chat session led by Andrew Ng. On November 16th, the authors of SWE-bench will be invited to speak; they are currently working at Anthropic. The materials will be updated throughout the course until the final presentation in December. The lecture notes are available publicly, but the assignments and project feedback are still intended for the students within the university community only.