Stanford's new CS329Z course dissects agent architecture and real failure cases
dotey · x · 2026-11-12
Stanford's new CS329Z 'Engineering AI Agents' course is publishing lectures on Stanford Online. Lecture 1 by Diyi Yang breaks agents into a ReAct-style think-act-observe loop plus memory and tools, using a real OpenClaw failure — batch-deleting emails because the 'ask me first' instruction got compressed away during memory summarization — to show how each component can break. Co-instructor John Yang created SWE-bench.
More from coding & agent
- Riveter V2 launches: describe any dataset and it researches the live web to return structured data — ycombinator · 2026-11-12
- Nous hackathon entry Mote lets Hermes render, inspect, and fix its own documents — kristoph · 2026-11-12
- Hero's Journey: open-source six-step AI short-film workflow, free to fork — neuroform · 2026-11-12
- Mote: a single executable that turns Markdown into Word, PPT and PDF for agents — kristoph · 2026-11-12
- xAI ships Grok Bot 1.0: AI teammates that join meetings and work in Slack — soleio · 2026-11-12
- Ten gems from Diogo's Code Summit talk: RLHF is part of the problem, software is back — TheTuringPost · 2026-11-12