This white paper from Google explores Generative AI agents, defining them as applications that achieve goals by interacting with the world using tools. It details the key components: a model (often a large language model), tools (including Extensions, Functions, and Data Stores), and an orchestration layer that manages the agent's reasoning and actions. Different cognitive architectures and reasoning frameworks, such as ReAct and Chain-of-Thought, are examined, showing how agents learn to select and use appropriate tools. The paper also discusses practical applications and implementation using LangChain and Google's Vertex AI platform, highlighting techniques like targeted learning to improve agent performance.

Hosted on Acast. See acast.com/privacy for more information.