2025AI智能体上手指南（英）

Introduction to Agents Authors: Alan Blount, Antonio Gulli, Shubham Saboo, Michael Zimmermann, and Vladimir VuskovicIntroduction to Agents and Agent architecturesNovember 20252AcknowledgementsContent contributorsEnrique ChanMike ClarkDerek EganAnant NawalgariaKanchana PatlollaJulia WiesingerCurators and editorsAnant NawalgariaKanchana PatlollaDesignerMichael Lanning From Predictive AI to Autonomous Agents 6Introduction to AI Agents 8The Agentic Problem-Solving Process 10A Taxonomy of Agentic Systems 14Level 0: The Core Reasoning System 15Level 1: The Connected Problem-Solver 15Level 2: The Strategic Problem-Solver 16Level 3: The Collaborative Multi-Agent System 17Level 4: The Self-Evolving System 18Core Agent Architecture: Model, Tools, and Orchestration 19Model: The “Brain” of your AI Agent 19Tools: The "Hands" of your AI Agent 20Retrieving Information: Grounding in Reality 21Executing Actions: Changing the World 21Function Calling: Connecting Tools to your Agent 22Table of contentsThe Orchestration Layer 22Core Design Choices 23Instruct with Domain Knowledge and Persona 23Augment with Context 24Multi-Agent Systems and Design Patterns 24Agent Deployment and Services 26Agent Ops: A Structured Approach to the Unpredictable 27Measure What Matters: Instrumenting Success Like an A/B Experiment 29Quality Instead of Pass/Fail: Using a LM Judge 29Metrics-Driven Development: Your Go/No-Go for Deployment 30Debug with OpenTelemetry Traces: Answering "Why?" 30Cherish Human Feedback: Guiding Your Automation 31Agent Interoperability 31Agents and Humans 32Agents and Agents 33Agents and Money 34Table of contentsSecuring a Single Agent: The Trust Trade-Off 34Agent Identity: A New Class of Principal 35Policies to Constrain Access 37Securing an ADK Agent 37Scaling Up from a Single Agent to an Enterprise Fleet 39Security and Privacy: Hardening the Agentic Frontier 40Agent Governance: A Control Plane instead of Sprawl 40How agents evolve and learn 42How agents learn and self evolve 43Simulation and Agent Gym - the next frontier 46Examples of advanced agents 47Google Co-Scientist 47AlphaEvolve Agent 49Conclusion 51Endnotes 52Table of contentsIntroduction to Agents and Agent architecturesNovember 20256From Predictive AI to Autonomous AgentsArtificial intelligence is changing. For years, the focus has been on models that excel at passive, discrete tasks: answering a question, translating text, or generating an image from a prompt. This paradigm, while powerful, requires constant human direction for every step. We're now seeing a paradigm shift, moving from AI that just predicts or creates content to a new class of software capable of autonomous problem-solving and task execution.This new frontier is built around AI agents. An agent is not simply an AI model in a static workflow; it's a complete application, making plans and taking actions to achieve goals. It combines a Language Model's (LM) ability to reason with the practical ability to act, allowing Agents are t