Description
Browse Stanford's online courses in Agentic AI: https://stanford.io/4AIRb2z
Lecture Overview:
What do you need to know about how LLMs work to build better AI agents? In this Stanford CS329Z: Engineering AI Agents lecture, Stanford PhD student and DSPy contributor Michael Ryan covers decoding strategies (sampling and temperature), transformer attention (including linear and hybrid attention) and context length, the pretraining-to-post-training pipeline (SFT, RLHF, RLVR), inference (prefill vs. decode and the KV cache) and test-time scaling, structured I/O and constrained decoding, and context engineering, including why stuffing everything into the context window backfires.
Speaker: Michael Ryan
Michael Ryan is a CS PhD student at Stanford University developing human-centered, open-source AI systems that augment human intelligence and reflect the plurality of human values. His contributions to the DSPy open-source project enable developers to optimize multi-stage AI systems, and his algorithms are used for automatic AI prompt optimization at Meta, Google, Dropbox, Amazon, and other organizations.
Chapters:
0:00 Introduction: the LLM stack
3:57 What is a language model?
7:46 Sampling: greedy, top-p, top-k & temperature
16:03 Transformer architecture & self-attention
28:27 Attention vs. recurrence & linear attention
36:01 Pretraining & mid-training
44:21 Post-training: SFT, RLHF & RLVR
47:33 Training LLMs for coding agents
49:11 Q&A: long-horizon data
52:30 Inference: prefill, decode & KV cache
59:56 Reasoning models & test-time scaling
1:04:58 Structured outputs & constrained decoding
1:10:12 Context engineering & context rot
1:16:03 Recursive language models
Course website: https://cs329z.stanford.edu/
Watch the full CS329Z playlist: https://youtube.com/playlist?list=PLPRcF5HR02_U&si=3Z4OYphngcmQ0GIy
Recorded 9/28/2026 in CS329Z: Engineering AI Agents, a graduate course at Stanford University taught by Diyi Yang, Michael Ryan, and John Yang.
#AIAgents #LLM #ContextEngineering