代理 AI 设计模式

代理 AI 设计模式是执行的 AI 系统的构建块 — 这些系统中 LLM 不是产品，而是决策循环的引擎。当你停止问“模型应该输出什么？”而开始问“代理接下来应该做什么，使用哪个工具，在谁的授权下，预算是多少？”时，这些就是你会使用的模式。

下面的选择偏向于生产：增强型 LLM 原语（模型 + 检索 + 工具 + 内存），将其包裹起来的循环（ReAct，计划与执行），使其可扩展的协调形状（监督者，协调者-工人），使其扎根的检索形状（代理 RAG），以及保持其问责的治理形状（决策日志，步骤预算，沙箱隔离）。它们共同回答“一旦你交付，代理 AI 系统实际上是什么样子？”

每个模式都以 GoF 风格进行分解：意图、上下文、问题、力量、因此、解决方案、后果（好处 + 负担）、约束（在强大生成器下保持模式一致的严格禁止），以及相关[] — 与其他模式的类型边缘。请参阅下面的相关指南以获取相邻框架 — 架构、代理设计模式更广泛、多代理协调、RAG 和安全性。

Field-tested patterns to start with

• Augmented LLM — Build the foundational agent block as an LLM augmented with retrieval, tools, and memory that the model actively chooses to use, rather than a bare-model call.
• ReAct — Interleave a single thought, a single tool call, and a single observation per step so the agent reasons over fresh evidence.
• Tool Use — Let the LLM produce typed calls against an external toolkit instead of producing free-form text the surrounding system has to parse.
• Plan-and-Execute — Plan all the steps once with a strong model, then execute each step with a cheaper model under the plan.
• Agentic RAG — Replace static retrieve-then-generate with autonomous agents that plan, choose sources, retrieve iteratively, reflect, and re-query.
• Supervisor — Place a coordinating agent above a set of specialised agents and route work to them.
• Orchestrator-Workers — An orchestrator dynamically breaks a task into subtasks at runtime and delegates each to a worker LLM, then synthesises results.
• Decision Log — Persist the agent's reasoning trace alongside its actions so post-hoc review can explain why.
• Step Budget — Cap the number of tool calls or loop iterations the agent is allowed within a single request.
• Sandbox Isolation — Run agent-emitted code or actions in a contained environment with restricted filesystem, network, and process privileges.
• Approval Queue — Queue agent-proposed actions for asynchronous human review while the agent continues other work.
• Reflection — Have the model review its own output and produce a revised version in one or more passes.

Recommended reading

• Planning & Control Flow — 45 patterns
• Tool Use & Environment — 46 patterns
• Retrieval & RAG — 23 patterns
• Governance & Observability — 40 patterns
• Safety & Control — 66 patterns

Or open the full contents for all 531 patterns in 14 books.

Related guides

• AI Agents Patterns — AI agents patterns: named, reusable shapes for building AI agents that reason, use tools, coordinate, and stay safe — single-agent loops an…
• AI Agents Patterns Catalog — The AI agents patterns catalog: a complete, GoF-formal pattern language for AI agents across reasoning, planning, tool use, retrieval, memo…
• LLM Agent Design Patterns — A GoF-formal catalog of LLM agent design patterns: ReAct, tool use, plan-and-execute, reflection, step budget, and more. Each pattern decom…
• Agentic Design Patterns — A GoF-formal catalog of agentic design patterns — named, reusable shapes for building autonomous AI agents: agent loops, tool use, planning…
• AI Agent Design Patterns — How to build an AI agent: the named shapes you reach for during design and implementation — reasoning (ReAct, plan-and-execute, reflection)…
• Agent Design Patterns — Agent design patterns treat the agent loop as a software-engineering primitive: an observe→reason→act cycle wrapped in tools, memory, super…
• Agentic Patterns — A complete pattern language for agentic systems, organised in Alexander-style books across reasoning, planning, tool use, retrieval, verifi…
• Agentic AI Architecture — How to structure agentic AI: the architectural patterns that hold an LLM-powered system together. Supervisor, orchestrator-workers, augment…
• RAG Agent Patterns — Patterns for building retrieval-augmented generation agents: naive RAG, agentic RAG, hybrid search, cross-encoder reranking, contextual ret…
• Multi-Agent Patterns — Patterns for coordinating multiple LLM agents: supervisor, orchestrator-workers, handoff, debate, hierarchical agents, swarm, role assignme…
• AI Agent Safety Patterns — Safety patterns for LLM agents: step budget, kill switch, constitutional charter, approval queue, sandbox isolation, input/output guardrail…