Meta 工程团队复盘了用小团队借助 AI Agent 从零开发 Facebook 小组独立应用 Forum 的过程,核心是让 Agent 自主完成规划、实现、测试和修正,工程师负责设定目标与上线批准。
Meta 如何用 AI 构建 Forum:让 Agent 不只写代码,还能验证自己是否做完
Forum 是 Meta 面向 Facebook 小组推出的独立应用,不是工程平台。@Meta_Engineers 这篇博客复盘了一个小团队如何借助 Agent 从零开发产品。核心不再是生成更多代码,是让规划、实现、测试和修正持续运转,人负责目标、关键判断与上线批准。
https://engineering.fb.com/2026/10/15/ml-applications/how-meta-built-forum-ai-native-engineering/
从逐步指挥,转向监督完整执行过程
团队追求的是:Agent 自己完成多轮开发,工程师在偏离目标时介入,而不是每一步都下指令。但作者明确说,这仍是推进中的方向,并非所有阶段都已实现。
关键是把“做完”的标准接进执行过程:需求写成精确、机器可读的规格;构建结果、单元测试和界面对照检查持续反馈给 Agent;导航 Agent 操作正在运行的应用,执行回归验证。提交审查时,还要附上构建、测试结果和验证录像。能编译并不代表完成,每项上线仍需工程师批准。
收益不只来自模型,更来自代码和工具改造
团队把反复出现的纠正写进 Rules,用 Skills 固化操作,通过 Commands 串联复现、修复和审查,再用 Hooks 自动触发检查。
更值得注意的是,他们认为最大收益来自改造代码和工具,而非反复调整提示词:让模块结构更清楚,必要时减少复用;让 Agent 直接读取模拟器的实时界面层级,而不是依赖文字描述。
团队用八周实现了与 Facebook Groups 的核心功能对等;开发阶段的压力测试还帮助发现内存问题,修复后内存占用和缺页次数约减半。但文章未披露团队人数、完整功能范围及测试口径,也没有宣布这套工程系统开源。
🧠Forum, Meta’s new standalone Groups app, was built by a small team that brought an AI agent into the development cycle as a participant — planning, building, testing, and verifying. Building Forum we moved from human-in-the-loop — engineer drives, AI assists inside each step — to human-on-the-loop, where the agent runs the iteration (plan, implement, build, test, correct, repeat) and the engineer sets the goal, watches the run, and owns the outcome. The key decision: success wasn’t measured in code volume. It was measured in changes that build, pass tests, and survive review. ✅ Reached core feature parity with Facebook Groups in eight weeks 📈 An autonomous stress-testing loop caught a memory issue during development, roughly halving memory use and page faults 🔗 Read how our engineers went AI-native to build Forum: https://engineering.fb.com/2026/10/15/ml-applications/how-meta-built-forum-ai-native-engineering/ #AI #SoftwareEngineering #DeveloperTools #MobileEngineering #Engineering在 X 查看被引用的帖子
来源:meng shao · x.com