Stanford CS329Z releases Tool Use and Frameworks & Orchestration agent engineering lectures
stanfordnlp · x · 2026-10-12
Stanford's Fall 2026 CS329Z "Engineering AI Agents", taught by Diyi Yang, Michael Ryan and John Yang, teaches the engineering shift from monolithic models to compound AI systems and autonomous agents; all lecture materials are now centralized on the course site.
Newly released lectures:
- Lecture 4: Tool Use — why models need tools, the REST → function calling → MCP interface evolution, Toolformer/ToolkenGPT/Gorilla tool-learning routes, and tool-use quality evals covering failure modes like wrong tool choice, hallucinated arguments, ordering errors and unstable pass^k.
- Lecture 5: Frameworks & Orchestration — from DSPy signatures and modular optimizers to LangGraph's shared state and conditional edges, extending to temporal durable workflows, telemetry and failure recovery; managing long-running agents with traces, timeouts, retries and execution budgets.
The course's core claim: tool calls take LLMs out of the parameter world; frameworks and orchestration keep them recoverable, observable and evaluable in real multi-step systems. Students build RAG, tool use and agent loops from scratch, then learn frameworks like DSPy, with two homeworks and a quarter-long project.
More from coding & agent
- Open-source Anatomy skill turns one-line prompts into interactive 3D machine diagrams via Claude — blaizedsouza · 2026-10-12
- Open-source REA rockets from 30K to 120K stars in 3 days as people reverse-engineer AC protocols with Claude — anselm · 2026-10-12
- Agent Cards and MCP tool descriptions are unverified claims — a team built signed A2A tests — Suncly_com · 2026-10-12
- Voice agent fills CRM fields at 90% confidence — yet payments bounce on wrong account numbers — CoursePopular824 · 2026-10-12
- Dev keeps Codex as code reviewer on top of Claude-generated code — MartinGTobias · 2026-10-12
- How a broken customer ID let a support agent 'resolve' tickets it never read — aiagentguide · 2026-10-12