我花了 80% 的时间规划，20% 的时间用 AI 工具编写代码。
3 分•作者: cgvas•10 个月前
上个月用 AI 编码工具构建了三个应用程序。其中两个盈利了（总收入 120 美元），一个只是为了测试一些东西。
我想要测试一下，投入 80% 的时间用于规划和创建结构化文档的理论是否可行。
painpoint.space 总共花了 6 天时间。我花了 5 天时间只是……规划。实际编码可能只有 8 个小时。大约一周后，我迎来了第一个每月 20 美元的客户。
我过去常常直接跳到 Cursor，开始让它构建东西。浪费数小时反复解释相同的上下文。当它在不同会话之间忘记我们正在构建的内容时，我会感到沮丧。
因此，在打开任何 IDE 之前，我与 Claude 合作，创建了上下文文档，内容包括：
* 应用程序的确切功能
* 每个功能
* 数据库应该如何工作
* 甚至为我的 AI 助手定制规则
起初感觉像是在拖延。但当我最终开始编码时，一切都……奏效了。AI 立即理解了。没有困惑，无需重新解释。
为了测试这是否不是偶然，我尝试用 7 个提示构建 legaldiff.com。不是 7 个会话，而是从想法到运行应用程序总共 7 个提示。它之所以有效，是因为所有上下文都已存在。
三个应用程序的结果：
* painpoint.space：每月 40 美元经常性收入（从 Reddit 上寻找商业创意）
* renewpic.com：收入 130 美元，260 多个用户注册
* legaldiff.com：7 个提示即可获得可运行的 MVP
这种模式是一致的。更多的前期规划 = 与 AI 之间更少的反复沟通。
我的联合创始人现在正将此构建成一个可重复的系统 (precursor.tools)。仅仅谈论这种方法，就有 80 人加入了候补名单。
好奇其他人是否也发现了这一点。你的规划越好，你实际上使用这些 AI 工具“编码”的需求就越少。
查看原文
Built three apps last month using AI coding tools. Two are profitable ($120 total revenue), one was just to test something.<p>I wanted to test the theory of dedicating 80% of my time to planning and creating structured docs<p>painpoint.space took 6 days total. I spent 5 days just... planning. Maybe 8 hours actually coding. Got my first $20/month customer about a week later.<p>I used to jump straight into Cursor and start asking it to build stuff. Waste hours explaining the same context over and over. Get frustrated when it forgets what we're building between sessions.<p>So before opening any IDE, I workshopped with Claude to create context documents on
- What exactly the app does
- Every single feature
- How the database should work
- Even custom rules for my AI assistant<p>Felt like procrastinating at first. But when I finally started coding, everything just... worked. The AI understood immediately. No confusion, no re-explaining.<p>To test this wasn't a fluke, I tried building legaldiff.com with just 7 prompts. Not 7 sessions, 7 total prompts from idea to working app. It worked because all the context was already there.<p>Results from three apps
- painpoint.space: $40/month recurring (finds business ideas from Reddit)
- renewpic.com: $130 made, 260+ users signed up
- legaldiff.com: working MVP in 7 prompts<p>The pattern is consistent. More upfront planning = way less back-and-forth with AI.<p>My co-founder and I are now building this into a repeatable system (precursor.tools). Got 80 people on the waitlist just from talking about this approach.<p>Curious if others have stumbled into this. The better your planning, the less you actually need to "code" with these AI tools.