请问HN：您更喜欢哪种代理框架/工具，为什么？
对于那些在B2B/B2C环境中实施了AI代理工作流程的朋友们：
你们选择了哪个框架（如SmolAgents、OpenAI Agents SDK、CrewAI、LangGraph等），为什么选择它？
你们权衡了哪些权衡（例如，简单性与控制、单代理与多代理需求）？
我特别想了解一些现实世界的见解，比如：
- SmolAgents的简约性（适合轻量级任务）是否足够，还是需要CrewAI的多代理协作（基于角色的工作流程）？
- 对于OpenAI SDK的用户：紧密的生态系统集成是否值得承担供应商锁定的风险？
- 状态管理有多重要（例如，LangGraph的有向无环图与AutoGen的异步聊天）？
感谢分享你们的经验！
查看原文
For those who've implemented AI agentic workflows in B2B/B2C environments:<p>Which framework did you choose (SmolAgents, OpenAI Agents SDK, CrewAI, LangGraph, etc.) and why?<p>What tradeoffs did you weigh (e.g., simplicity vs. control, single vs. multi-agent needs)?<p>I’m especially curious about real-world insights like:<p>- Did SmolAgents’ minimalism (ideal for lightweight tasks) suffice, or did you need CrewAI’s multi-agent collaboration (role-based workflows)?<p>- For OpenAI’s SDK users: Was tight ecosystem integration worth vendor lock-in risks?<p>- How critical was state management (e.g., LangGraph’s DAGs vs. AutoGen’s async chat)?<p>Thanks for sharing your experiences!