We've worked with dozens of teams building LLM agents across industries. Consistently, the most successful implementations use simple, composable patterns rather than complex frameworks. Over the past year, we've worked with dozens of teams building large language model (LLM) agents across industries. Consistently, the most successful implementations weren't using complex frameworks or specialized libraries. Instead, they were building with simple, composable patterns. In this post, we share what we’ve learned from working with our customers and building agents ourselves, and give practical advice for developers on building effective agents.
核心贡献 · Key contributions
倡导使用简单、可组合的模式而非复杂框架来构建 LLM 智能体。 Advocates simple, composable patterns over complex frameworks for building LLM agents.
明确定义了工作流（预定义代码路径）与智能体（动态 LLM 控制）之间的架构区别。 Defines clear architectural distinction between workflows (predefined code paths) and agents (dynamic LLM control).
识别出六种常见的智能体模式：增强型 LLM、提示链、路由、并行化、编排器-工作者、评估器-优化器。 Identifies six common agentic patterns: augmented LLM, prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer.
建议从简单的 LLM 调用开始，仅在能显著改善结果时才增加复杂性。 Recommends starting with simple LLM calls and only adding complexity when it demonstrably improves outcomes.
通过 ACI（智能体-计算机接口）和工具文档提供实用的工具设计指南。 Provides practical tool design guidelines via ACI (Agent-Computer Interface) and tool documentation.
局限 · Limitations
聚焦于 LLM 智能体，结论可能不适用于其他类型的 AI 智能体。 Focuses on LLM agents; findings may not generalize to other AI agent types.
依赖与客户合作中的轶事证据，而非控制实验。 Relies on anecdotal evidence from customer collaborations rather than controlled experiments.
未讨论极端延迟或成本约束下的可扩展性或性能。 Does not address scalability or performance under extreme latency or cost constraints.
模式假设 LLM 能可靠使用工具并恢复错误，未在不可靠模型上验证。 Patterns assume reliable LLM tool use and error recovery; not validated for unreliable models.
缺乏跨不同任务和领域对智能体模式的定量比较。 Lacks quantitative comparison of agentic patterns across diverse tasks and domains.
论文章节 · Sections（共 14）
构建高效智能体Building effective agents
什么是智能体？What are agents?
何时（以及何时不）使用智能体When (and when not) to use agents
何时以及如何使用框架When and how to use frameworks
构建模块、工作流与智能体Building blocks, workflows, and agents