What Are AI Agents? Agentic Systems Explained | Stanford CS329Z Lecture 1 with Diyi Yang

What Are AI Agents? Agentic Systems Explained | Stanford CS329Z Lecture 1 with Diyi Yang

Title history
  • What Are AI Agents? Agentic Systems Explained | Stanford CS329Z Lecture 1 with Diyi Yang — observed on 20261110-2124 UTC (YouTube)
  • Stanford CS329Z Introduction — What Are Agentic Systems? — observed on 20261109-2119 UTC (YouTube)
🎙 Diyi Yang 👥 1.3M 📅 November 9, 2026 ⏱ 69 min 👁 70K 📄 science communication 🧭 2026-11-10
Available in: English (current) Français

Keywords

AI agentsagent architectureReActmulti-agent systemstool use

Summary

This lecture from Stanford CS329Z introduces the concept of AI agents and agentic systems. Instructor Diyi Yang traces the history of agents from the 1973 actor model to modern LLM-based agents like ReAct and AutoGPT. She breaks down the anatomy of an LLM agent into five core components: an LLM core, planning and reasoning, memory, tools, and environment. The lecture covers key techniques such as chain-of-thought reasoning, ReAct, and Reflexion, as well as memory types (episodic, semantic, procedural). It also discusses tool use, including Toolformer and Gorilla, and explains compound AI systems, distinguishing between workflows and agents. The lecturer reviews model capability improvements and agent performance gains, noting that stronger base models lead to better agents. She highlights the importance of reliability, safety, privacy, and evaluation as key challenges for the field. The course logistics, including prerequisites and expectations, are also presented.

142 words

Critical Evaluation

Value of the Information & Strength of the Argument

The lecture provides a comprehensive and well-structured overview of AI agents, offering both historical context and a clear conceptual framework. The argumentation is solid, relying on established research such as ReAct, chain-of-thought prompting, and tool use. The speaker effectively demonstrates why reasoning, memory, and tools are essential for agent performance, using concrete examples and diagrams. The discussion on model capability improvements and agent benchmarks is supported by data trends, and the lecturer critically addresses the reliability-efficiency trade-off. The course encourages hands-on learning and emphasizes the importance of evaluation, making it a valuable resource for understanding modern agentic systems.

Scientific Rigor, Source Quality, Title Accuracy

The lecture is rigorous, referencing seminal works and presenting a balanced view of the field. The speaker is a Stanford professor with recognized expertise, ensuring high scientific credibility. Sources cited include the actor model, ReAct, AutoGPT, Toolformer, and Model Context Protocol, though specific paper references are not linked in the video description. The description provides links to the course website and related resources, which are authoritative. The title accurately matches the content, and the lecture stays focused on explaining agents and agentic systems. The structure is clear, with logical progression from history to components to challenges.

209 words

Title / Content Match

The title accurately reflects the content: it explains what AI agents are and introduces agentic systems. The lecture is the first in a Stanford course, matching the description.

Quality & Reliability

9/10

The lecture is delivered by a Stanford professor, covers well-established research, and provides accurate descriptions of agent architectures. It includes historical context and references to fundamental works.

Chapters

Cited Sources

Contribution & Novelties

This lecture offers a valuable synthesis of the field of AI agents, providing a clear taxonomy of components and a historical narrative that situates current developments. It bridges foundational concepts with modern practicalities, such as the use of MCP for tool integration. The contribution lies in its pedagogical clarity and its emphasis on reliability and evaluation as open challenges, which is crucial for practitioners and researchers.

66 words

Radar Profile

The radar profile is fairly balanced, with slightly higher scores in information quality and reliability, reflecting the academic rigor and depth of the lecture. The technical level is high but appropriate for a graduate audience, and the quantity of information is substantial.

Reliability 9/10