[{"data":1,"prerenderedAt":600},["ShallowReactive",2],{"blog:\u002Fblog\u002F2026\u002F10\u002F08\u002Fai-assisted-dev-workflow\u002F":3,"blog-next:\u002Fblog\u002F2026\u002F10\u002F08\u002Fai-assisted-dev-workflow\u002F":425},{"id":4,"title":5,"body":6,"categories":408,"comments":410,"copyrightNotice":411,"date":412,"description":413,"draft":411,"extension":414,"legacySlug":415,"meta":416,"navigation":410,"path":417,"seo":418,"stem":419,"tags":420,"updated":423,"__hash__":424},"blog\u002Fblog\u002F2026\u002F10\u002F08\u002Fai-assisted-dev-workflow.md","从技术方案设计到开发自测上线的AI实践",{"type":7,"value":8,"toc":382},"minimark",[9,14,22,30,34,39,58,62,82,93,98,104,114,120,125,131,135,152,156,160,171,175,180,186,191,197,202,205,210,216,220,224,235,239,247,251,255,263,267,281,337,343,347,351,362,366,376],[10,11,13],"h2",{"id":12},"一后端研发工作概览","一、后端研发工作概览",[15,16,17],"p",{},[18,19],"img",{"alt":20,"src":21},"AI 辅助后端研发的六个环节与分工","\u002Fblog-assets\u002Fai-assisted-dev-workflow\u002Fworkflow-overview.svg",[15,23,24,25,29],{},"后端开发工作主要包括",[26,27,28],"strong",{},"技术方案设计、开发、自测、联调、CR、上线","，后边章节主要从这些具体的工作来描述下 AI 工具的使用以及一些个人的小技巧。",[10,31,33],{"id":32},"二技术方案设计","二、技术方案设计",[35,36,38],"h3",{"id":37},"_21常见问题","2.1、常见问题",[40,41,42,46,49,52,55],"ul",{},[43,44,45],"li",{},"缺少技术相关的约束信息，比如技术选型、业务约束信息",[43,47,48],{},"AI 生成的技术方案过于发散，过度设计",[43,50,51],{},"缺少关键模块，遗漏重要信息",[43,53,54],{},"遗漏异常处理、幂等、并发、事务、权限和兼容性。",[43,56,57],{},"技术方案和接口文档分别生成，最终出现遗漏或口径冲突。",[35,59,61],{"id":60},"_22具体实践","2.2、具体实践",[40,63,64,71,74],{},[43,65,66,67,70],{},"通过 ",[26,68,69],{},"AGENTS.md"," 固化团队约束，让 AI 在进入项目时自动读取。",[43,72,73],{},"补充项目特有约束，例如技术选型、模块边界、状态流转、外部依赖和兼容要求。",[43,75,76,77],{},"提供",[78,79,81],"a",{"href":80},"\u002Fblog-assets\u002Fai-assisted-dev-workflow\u002Ftemplates\u002Ftech-design-template.md","技术方案模板",[83,84,89],"pre",{"className":85,"code":87,"language":88},[86],"language-text","我：根据需求文档XXX，阅读现有工程代码，按照技术方案模板，生成技术方案。其中重要约束：XXX\n","text",[90,91,87],"code",{"__ignoreMap":92},"",[40,94,95],{},[43,96,97],{},"关键处理流程先对齐，再让 AI 画 plantUML 图",[83,99,102],{"className":100,"code":101,"language":88},[86],"我：先对齐下创建任务的处理流程\nAI：XXXXXX\n我：XXX会不会有并发的问题，是不是XXX处理会更好些？\nAI：XXXXXX\n......\n我：基于以上对话，生成创建任务的时序图，使用plantUML\n",[90,103,101],{"__ignoreMap":92},[40,105,106],{},[43,107,108,109,113],{},"根据",[78,110,112],{"href":111},"\u002Fblog-assets\u002Fai-assisted-dev-workflow\u002Ftemplates\u002Fapi-doc-template.md","接口文档模板","生成接口文档",[83,115,118],{"className":116,"code":117,"language":88},[86],"我：根据需求文档XXX、技术文档XXX，按照接口文档模板，生成接口文档。其中重要约束：XXX\n",[90,119,117],{"__ignoreMap":92},[40,121,122],{},[43,123,124],{},"对齐需求文档、技术文档、接口文档，确保无遗漏项、冲突项",[83,126,129],{"className":127,"code":128,"language":88},[86],"我：对齐需求文档XXX、技术文档XXX、接口文档XXX，确保无遗漏项、冲突项、待解决项\n",[90,130,128],{"__ignoreMap":92},[35,132,134],{"id":133},"_23人工检查点","2.3、人工检查点",[40,136,137,140,143,146,149],{},[43,138,139],{},"方案是否满足真实业务目标，而不只是技术上可实现。",[43,141,142],{},"是否复用了现有能力，避免重复建设和过度抽象。",[43,144,145],{},"关键状态是否具备合法流转、幂等和并发控制。",[43,147,148],{},"接口、数据和配置变更是否兼容已有调用方。",[43,150,151],{},"异常是否有明确返回、日志、监控",[10,153,155],{"id":154},"三开发自测","三、开发自测",[35,157,159],{"id":158},"_31常见问题","3.1、常见问题",[40,161,162,165,168],{},[43,163,164],{},"技术方案过于笼统，AI 在缺少上下文时自行补全实现细节。",[43,166,167],{},"只覆盖正常路径，遗漏空值、极值、重复请求、并发和下游异常。",[43,169,170],{},"修改新功能时破坏旧接口、旧字段或历史数据兼容性。",[35,172,174],{"id":173},"_32具体实践","3.2、具体实践",[40,176,177],{},[43,178,179],{},"实际开发前，先让 AI 输出开发规划",[83,181,184],{"className":182,"code":183,"language":88},[86],"依据需求文档、技术方案、接口文档和项目 AGENTS.md 制定开发规划。\n",[90,185,183],{"__ignoreMap":92},[40,187,188],{},[43,189,190],{},"按步骤开发，完成一个步骤，同时进行相关测试",[83,192,195],{"className":193,"code":194,"language":88},[86],"完成步骤XXX，开发时同步开发单测\n",[90,196,194],{"__ignoreMap":92},[40,198,199],{},[43,200,201],{},"使用 curl 模拟接口请求进行集成测试",[15,203,204],{},"完成代码不等于完成任务。至少需要给出编译结果、测试结果、关键差异和未覆盖风险。无法执行的验证必须明确说明原因。",[15,206,207],{},[26,208,209],{},"完成前检查",[83,211,214],{"className":212,"code":213,"language":88},[86],"使用curl模拟接口请求进行集成测试，定时任务通过临时脚本触发。\n最后汇总：修改文件、核心行为变化、验证接口及结果、兼容性结论、剩余风险。\n如果有测试未执行，请说明具体原因。\n",[90,215,213],{"__ignoreMap":92},[10,217,219],{"id":218},"四联调","四、联调",[35,221,223],{"id":222},"_41常见问题","4.1、常见问题",[40,225,226,229,232],{},[43,227,228],{},"前端请求缺字段、字段类型错误，或实际请求与接口文档不一致。",[43,230,231],{},"后端响应结构、字段名或错误码与约定不同。",[43,233,234],{},"只看到最终报错，需要按调用链定位问题发生在哪一步。",[35,236,238],{"id":237},"_42具体实践","4.2、具体实践",[40,240,241,244],{},[43,242,243],{},"复制前端入参，直接发给 AI 进行入参诊断",[43,245,246],{},"复制日志平台出入参，发给 AI，让 AI 对比接口文档进行分析",[10,248,250],{"id":249},"五cr","五、CR",[35,252,254],{"id":253},"_51常见问题","5.1、常见问题",[40,256,257,260],{},[43,258,259],{},"边界条件、异常路径被遗漏。",[43,261,262],{},"AI 增加了很多业务上不存在场景的校验和处理",[35,264,266],{"id":265},"_52具体实践","5.2、具体实践",[40,268,269,272,275,278],{},[43,270,271],{},"先执行 AI Review，再由人工结合需求和项目背景复核。",[43,273,274],{},"按“必须修复、建议优化、待确认”分级，避免所有意见权重相同。",[43,276,277],{},"每个问题必须包含文件位置、触发条件、影响和修改建议。",[43,279,280],{},"对于 AI 判断错误的意见，记录误报原因，持续完善项目约束。",[282,283,284,300],"table",{},[285,286,287],"thead",{},[288,289,290,294,297],"tr",{},[291,292,293],"th",{},"环节",[291,295,296],{},"更适合 AI",[291,298,299],{},"必须人工确认",[301,302,303,315,326],"tbody",{},[288,304,305,309,312],{},[306,307,308],"td",{},"代码检查",[306,310,311],{},"异常处理、空值、重复代码、规范和测试遗漏",[306,313,314],{},"业务语义、方案合理性和隐性需求",[288,316,317,320,323],{},[306,318,319],{},"风险分析",[306,321,322],{},"并发、事务、幂等、兼容性和安全检查清单",[306,324,325],{},"风险是否真实、影响是否可接受",[288,327,328,331,334],{},[306,329,330],{},"修改建议",[306,332,333],{},"给出代码位置、触发条件和可选修复方案",[306,335,336],{},"选择最终方案并判断维护成本",[83,338,341],{"className":339,"code":340,"language":88},[86],"对比当前分支与 master，结合需求、技术方案和项目规范进行 Review。\n重点检查业务正确性、并发、事务、幂等、兼容性、安全和测试覆盖。\n按“必须修复、建议优化、待确认”分级。\n每项提供文件位置、触发条件、影响和建议；没有证据的问题不要下结论。\n",[90,342,340],{"__ignoreMap":92},[10,344,346],{"id":345},"六上线","六、上线",[35,348,350],{"id":349},"_61常见问题","6.1、常见问题",[40,352,353,356,359],{},[43,354,355],{},"配置项遗漏",[43,357,358],{},"发布顺序错误，导致线上故障",[43,360,361],{},"无回滚方案，导致线上问题时无法快速回滚",[35,363,365],{"id":364},"_62具体实践","6.2、具体实践",[40,367,368],{},[43,369,370,371,375],{},"按照",[78,372,374],{"href":373},"\u002Fblog-assets\u002Fai-assisted-dev-workflow\u002Ftemplates\u002Frelease-checklist-template.md","上线 checkList 模板","生成上线 checkList，避免上述问题",[83,377,380],{"className":378,"code":379,"language":88},[86],"对比当前开发分支和 master，并结合技术方案与上线 Checklist 模板生成检查清单。\n",[90,381,379],{"__ignoreMap":92},{"title":92,"searchDepth":383,"depth":383,"links":384},2,[385,386,392,396,400,404],{"id":12,"depth":383,"text":13},{"id":32,"depth":383,"text":33,"children":387},[388,390,391],{"id":37,"depth":389,"text":38},3,{"id":60,"depth":389,"text":61},{"id":133,"depth":389,"text":134},{"id":154,"depth":383,"text":155,"children":393},[394,395],{"id":158,"depth":389,"text":159},{"id":173,"depth":389,"text":174},{"id":218,"depth":383,"text":219,"children":397},[398,399],{"id":222,"depth":389,"text":223},{"id":237,"depth":389,"text":238},{"id":249,"depth":383,"text":250,"children":401},[402,403],{"id":253,"depth":389,"text":254},{"id":265,"depth":389,"text":266},{"id":345,"depth":383,"text":346,"children":405},[406,407],{"id":349,"depth":389,"text":350},{"id":364,"depth":389,"text":365},[409],"AI",true,false,"2026-10-08 16:40:00","后端开发工作主要包括技术方案设计、开发、自测、联调、CR、上线。本文主要从这些具体的工作出发，描述 AI 工具的使用以及一些个人的小技巧。","md","ai-assisted-dev-workflow",{},"\u002Fblog\u002F2026\u002F10\u002F08\u002Fai-assisted-dev-workflow",{"title":5,"description":413},"blog\u002F2026\u002F10\u002F08\u002Fai-assisted-dev-workflow",[409,421,422],"研发流程","工程实践",null,"17fGO7bzwMwF9WzAXBem-PPLAIo2HeGYj12pJfYqESU",{"id":426,"title":427,"body":428,"categories":587,"comments":410,"copyrightNotice":411,"date":589,"description":590,"draft":411,"extension":414,"legacySlug":591,"meta":592,"navigation":410,"path":593,"seo":594,"stem":595,"tags":596,"updated":423,"__hash__":599},"blog\u002Fblog\u002F2026\u002F06\u002F09\u002Fai-startup-validation-before-mvp.md","AI 时代创业最容易被忽略的一步：大多数人死在验证前",{"type":7,"value":429,"toc":580},[430,437,441,444,451,454,457,461,468,471,474,478,481,491,494,497,500,503,536,539,542,549,558,561,567,570],[15,431,432,433,436],{},"最近在 Anthropic 官方博客读了一份 AI 时代的创业指南，叫 ",[26,434,435],{},"《The Founder's Playbook: Building an AI-Native Startup》","，其中的很多观点读起来让人如梦初醒。",[10,438,440],{"id":439},"ai-让做出来变得前所未有的容易","AI 让「做出来」变得前所未有的容易",[15,442,443],{},"2026 年，AI 已经可以写生产级代码、做市场调研、分析竞品格局、起草融资材料、自动化运营流程。一个从来没写过一行代码的创始人，今天就能把应用跑起来；一个不懂商业的技术人，也能轻松生成一套完整的商业策略和融资 PPT。",[15,445,446,447,450],{},"这本指南把 AI 时代的创业分成四个阶段：",[26,448,449],{},"创意阶段（Idea）→ MVP 阶段 → Launch阶段 → Scale 阶段","。每个阶段都有对应的核心问题、行动建议和退出条件。",[15,452,453],{},"听起来很美好对不对？",[15,455,456],{},"但问题来了——",[10,458,460],{"id":459},"能跑起来-有人在用","能跑起来 ≠ 有人在用",[15,462,463,464,467],{},"很多人在 AI 的加持下，反而更容易跳过创业旅途中",[26,465,466],{},"最重要的一步","：验证自己的想法是不是真的解决了人们的需求。",[15,469,470],{},"一个能跑的原型太容易让人产生错觉——「我已经做出了东西，说明我正在解决一个真问题」。但实际上，从「有个想法」到「有人愿意付钱」，中间隔着一整个验证过程，而这个过程是 AI 加速不了的。",[15,472,473],{},"错误的努力方向，消耗的不仅是你燃烧的 token，还有你最宝贵的时间和精力。",[10,475,477],{"id":476},"大多数创意压根不应该走到-mvp","大多数创意压根不应该走到 MVP",[15,479,480],{},"指南里有句话特别扎心：",[482,483,484],"blockquote",{},[15,485,486,487,490],{},"大多数创意是没有经过验证和调研的，",[26,488,489],{},"根本不应该走到 MVP 阶段","。",[15,492,493],{},"但 AI 让「走到 MVP」变得前所未有的容易，于是这个最该被认真对待的阶段，反而成了最容易被跳过的阶段。",[15,495,496],{},"关于这一点，我深有感触。",[10,498,499],{"id":499},"创意阶段的关键问题",[15,501,502],{},"指南的创意阶段给了几个关键问题，帮助你判断自己的创意是否值得继续：",[504,505,506,512,518,524,530],"ol",{},[43,507,508,511],{},[26,509,510],{},"这个问题真实且具体吗？"," —— 不是「我想做一个帮助人们学习的工具」，而是「高三学生每次数学考试前两周平均会花 X 小时复习，但市面上没有专门针对这个场景的工具」",[43,513,514,517],{},[26,515,516],{},"出现得足够高频吗？"," —— 是每天、每周，还是每年才一次？",[43,519,520,523],{},[26,521,522],{},"你真的和目标用户聊过吗？"," —— 不是问卷，不是推测，是真实的对话",[43,525,526,529],{},[26,527,528],{},"有人做类似的产品吗？他们做得怎么样？"," —— 如果已经有人在做了，你凭什么能差异化？",[43,531,532,535],{},[26,533,534],{},"你的方案能完美解决这个问题吗？"," —— 还是只能部分解决？",[15,537,538],{},"这 5 个问题，每个都值得认真花时间回答。但现实是，大多数人在「想清楚」之前就已经开始「做」了。",[10,540,541],{"id":541},"我做了一个验证工具",[15,543,544,545,548],{},"为了让更多人在正式动手之前先过一遍这 5 个关键问题，我根据这份指南做了一个「",[26,546,547],{},"创意小黄鸭","」诊断工具：",[15,550,551,552],{},"👉 ",[78,553,557],{"href":554,"rel":555},"https:\u002F\u002Fwww.zhizhi123.com\u002Fcreative-duck\u002F",[556],"nofollow","点击体验：zhizhi123.com\u002Fcreative-duck",[15,559,560],{},"只需要 5 道简单的选择题，就能快速诊断你的创意成熟度，并给出一份针对性的行动建议。",[15,562,563,564],{},"如果诊断结果不理想——这恰恰是最好的结果。",[26,565,566],{},"在错误的方向上停下来，比在错误的路上狂奔重要得多。",[568,569],"hr",{},[482,571,572],{},[15,573,574,575],{},"原文：《The Founder's Playbook: Building an AI-Native Startup》— Anthropic 官方博客\n地址：",[78,576,579],{"href":577,"rel":578},"https:\u002F\u002Fclaude.com\u002Fblog\u002Fthe-founders-playbook",[556],"The Founder's Playbook",{"title":92,"searchDepth":383,"depth":383,"links":581},[582,583,584,585,586],{"id":439,"depth":383,"text":440},{"id":459,"depth":383,"text":460},{"id":476,"depth":383,"text":477},{"id":499,"depth":383,"text":499},{"id":541,"depth":383,"text":541},[588],"随笔","2026-06-09 21:00:00","最近在 Anthropic 官方博客读了一份 AI 时代的创业指南，叫 《The Founder's Playbook: Building an AI-Native Startup》，其中的很多观点读起来让人如梦初醒。 AI 让「做出来」变得前所未有的容易 2026 年，AI 已经可以写生产级代码、","ai-startup-validation-before-mvp",{},"\u002Fblog\u002F2026\u002F06\u002F09\u002Fai-startup-validation-before-mvp",{"title":427,"description":590},"blog\u002F2026\u002F06\u002F09\u002Fai-startup-validation-before-mvp",[597,409,598],"创业","创意验证","O749L2PuQisNDEEu4no83m7H9Hi6vTEtWosbzvvb_lI",1792885118840]