谷歌智能体白皮书_Agents_Companion-英文版

Agents Companion Authors: Antonio Gulli, Lavi Nigam, Julia Wiesinger, Vladimir Vuskovic, Irina Sigler, Ivan Nardini, Nicolas Stroppa, Sokratis Kartakis, Narek Saribekyan, and Alan BountAgents CompanionFebruary 20252AcknowledgementsEditors & curatorsAnant NawalgariaContent contributorsAnant NawalgariaSteven JohnsonHussain Chinoy DesignerMichael Lanning Introduction 6Agent Ops 8Agent Success Metrics 12Agent Evaluation 14Assessing Agent Capabilities 15Evaluating Trajectory and Tool Use 17Evaluating the Final Response 20Human-in-the-Loop Evaluation 21More about Agent Evaluation 22Multiple Agents & Their Evaluation 23Understanding Multi-Agent Architectures 24Multi-Agent Design Patterns and Their Business Impact 25Important components of Agents 28Challenges in Multi-Agent systems 31Multi-Agent Evaluation 32Table of contentsAgentic RAG: A Critical Evolution in Retrieval-Augmented Generation 33Agentic RAG and its Importance 34Better Search, Better RAG 36Agents in the enterprise 38Manager of agents 38Google Agentspace 40NotebookLM Enterprise 41Google AgentSpace Enterprise 43From agents to contractors 46Contracts 46Contract Lifecycle 49Contract execution 49Contract Negotiation 50Contract Feedback 51Subcontracts 51Automotive AI: Real World Use of Multi-Agent Architecture 54Specialized Agents 54Conversational Navigation Agent 54Conversational Media Search Agent 56Message Composition Agent 56Car Manual Agent 57General Knowledge Agent 58Patterns in Use 58Hierarchical Pattern 58Diamond Pattern 60Peer-to-Peer 62Collaborative Pattern 64Response Mixer Agent 66Adaptive Loop Pattern 67Advantages of Multi-Agent Architecture for Automotive AI 68Agent Builder 69Summary 70Endnotes 74Agents CompanionFebruary 20256IntroductionGenerative AI agents mark a leap forward from traditional, standalone language models, offering a dynamic approach to problem-solving and interaction. As defined in the original Agents paper, an agent is an application engineered to achieve specific objectives by perceiving its environment and strategically acting upon it using the tools at its disposal. The fundamental principle of an agent lies in its synthesis of reasoning, logic, and access to external information, enabling it to perform tasks and make decisions beyond the inherent capabilities of the underlying model. These agents possess the capacity for autonomous operation, independently pursuing their goals and proactively determining subsequent actions, often without explicit instructions.The future of AI is agentic.Agents CompanionFebruary 20257The architecture of an agent is composed of three essential elements that drive its behavior and decision-making:• Model: Within the agent's framework, the term "model" pertains to the language model (LM) that functions as the central decision-making unit, employing instruction-based reasoning and logical frameworks. The model can vary from general-purpose to multimodal or fine-tuned, depending on the agent's specific requirements.• Tools: Tool