资深Builder手工川近日分享了利用AI大幅加速软件开发的实战经验。自2024年10月起，他将99%的编码工作交由AI完成，仅由人类负责项目的起始与结束阶段[1]。通过这种协作方式，他通常能在一周内交付最小可行产品（MVP），并在约一个月时间内将产品相对稳定地上线[1]。
在工具选择上，手工川采用多模型分工协作的策略，使用Claude Sonnet 3.5、Cursor、GPT系列、DeepSeek、Grok等多种模型[1]。为了统一管理多个AI Agent的协作，他开发了开源产品YODA，实现了便捷的Agent入口管理、自动化归档和自动收尾功能[1]。在代码部署方面，他采用独立分支开发的模式，完成后自动合并，将原本需要数十分钟的服务不可用时间压缩至几秒钟[1]。
关于成本投入，手工川提到Claude API的单日费用曾达到200至300美元，但他建议普通用户仅需准备一个Max账号即可满足需求[1]。他还强调，真实的产品需求来自于对生活的观察而非技术追求，认为"爱生活的人更容易发现大众需求"，产品设计不必过度追求复杂的技术实现[1]。
Shougong Chuan, an experienced builder, has revealed how leveraging multiple AI agents and artificial intelligence has dramatically lowered the software development threshold.[1] Starting from October 2024, he has delegated 99% of coding work to AI, with humans responsible for only 1% of the initial setup and final touches.[1] Through progressive development methodology, he typically delivers a minimum viable product within one week and achieves a relatively stable product launch in approximately one month.[1]
To optimize this workflow, Shougong Chuan employs a diverse toolkit of language models for collaborative task distribution, including Claude Sonnet 3.5, Cursor, GPT series, DeepSeek, and Grok.[1] He developed an open-source product called YODA to unify management of multiple AI agents, enabling efficient product research and development processes.[1] His development approach uses independent branches that automatically merge upon completion, compressing what previously required dozens of minutes of downtime to just a few seconds.[1] For API costs, he noted that Claude's daily expenses have reached 200-300 dollars, though he recommends that ordinary users prepare a Max account for their needs.[1]
Shougong Chuan emphasizes that genuine product needs stem from keen observation of everyday life rather than pursuit of overly complex technology.[1] He believes that people who love life more easily discover mass-market demands, suggesting that successful products need not be technologically elaborate.[1]