google white paper ai agents 1736267517
PDF · 2 Mar 2025
This Google whitepaper (Sept 2024) explains Generative AI agents: applications that achieve goals by observing and acting on the world using external tools. It defines core components — model (LM as decision-maker), tools (Extensions, Functions, Data Stores), and an orchestration layer/cognitive architecture (reasoning frameworks like ReAct, Chain-of-Thought, Tree-of-Thoughts). The paper contrasts agents vs. models, describes how tools enable real-world interaction, and provides examples and code (Python snippets, LangChain, Vertex AI) illustrating Extensions, Function Calling, RAG with vector stores, and production agent architectures.
Questions this document answers
Julia Wiesinger; Patrick Marlow; Vladimir Vuskovic.
September 2024.