上个月有1万款应用在我们的平台上构建——以下是我们的收获。
大家好，我是Jonathan，Fine.dev的联合创始人。
在过去的一个月里，我们的平台——一个基于人工智能的应用创建平台，已经见证了超过10,000个应用的诞生。这让我们从一个独特的角度理解了人们如何实际使用人工智能来构建软件。我们原以为自己已经掌握了所有，但所学到的内容完全改变了我们的思维。
以下是我们学到的三大关键点：
1. 限制代理的操作范围可以显著改善结果。
起初，我们认为“人工智能能做的越多越好”。结果发现……并非如此。当代理拥有过多自由时，用户得到的结果往往模糊、冗长或不相关。但当我们缩小范围时，结果却惊人地改善。我们甚至几乎停止使用工具调用。我们从未预料到会发生这样的事情，但事实就是如此。总结一下——小而专注的提示 → 更简洁、更有用的应用。
2. 第一个提示非常重要。
我们发现提示的质量差异巨大。“给我一个生产力工具”和“给我一个包含3个可勾选并每天重置的早晨清单”的区别是天壤之别。实际上，应用的成功往往取决于第一个提示的详细程度。如果提示足够好，用户可以轻松在其基础上进行迭代，直到得到理想的结果。如果提示不够好，迭代的结果往往不太有用。总结一下——确保在你的第一个请求上投入精力，这将为后续过程定下基调。
3. 大多数应用都是小型、个人化和临时的。
这让我们大吃一惊的是：人们并不是在创建初创公司或商业，而是在为自己构建工具。为了这一周。为了这一刻。一个仅用于今年假期的礼物追踪器，一个为周末的团体旅行规划的工具，一个帮助孩子晨间日常的快速仪表板，一个一次性活动的RSVP方式。这些应用大多数并不是为了长久存在。这正是它们的价值所在。
这使我们在思维上发生了重大转变：
我们一直将软件视为产品或基础设施。但在看到10,000个应用诞生后，我们坚信它也正在成为内容：快速创建，易于丢弃，且极具个人化。实际上，我们甚至发布了一个Feed，每个帖子都是一个可以重混、重建或丢弃的工作应用。
我们认为，我们正进入一个一次性软件的时代，而人工智能应用构建者正是这一转变的体现。
如果大家对我们从前10,000个应用中学到的内容有任何问题，我很乐意以问答的形式进行解答。
查看原文
Hey all, Jonathan here, cofounder of Fine.dev<p>Over the last month alone, we've seen more than 10,000 apps built on our product, an AI-powered app creation platform. That gave us a pretty unique vantage point to understand how people actually use AI to build software. We thought we had it pretty much figured out, but what we learned changed our thinking completely.<p>Here are the three biggest things we learned:<p>1. Reducing the agent's scope of action improves outcomes (significantly)<p>At first, we thought “the more the AI can do, the better.” Turns out… not really. When the agent had too much freedom, users got vague, bloated, or irrelevant results. But when we narrowed the scope the results got shockingly better. We even stopped using tool calls almost all together. We never expected this to happen, but here we are. Bottom line - small, focused prompts → cleaner, more useful apps.<p>2. The first prompt matters. A lot.<p>We’ve seen prompt quality vary wildly. The difference between "make me a productivity tool" and "give me a morning checklist with 3 fields I can check off and reset each day" is everything. In fact, the success of the app often came down to just how detailed was that first prompt. If it was good enough - users could easily make iterations on top of it until they got their perfect result. If it wasn't good enough, the iterations weren't really useful. Bottom line - make sure to invest in your first request, it will set the tone for the rest of the process.<p>3. Most apps were small + personal + temporary.<p>Here’s what really blew our minds: People weren't building startups / businesses. They were building tools for themselves. For this week. For this moment. A gift tracker just for this year's holidays, a group trip planner for the weekend, a quick dashboard to help their kid with morning routines, a way to RSVP for a one-time event. Most of these apps weren’t meant to last. And that's what made them valuable.<p>This led us to a big shift in our thinking:<p>We’ve always thought of software as product or infrastructure. But after watching 10,000 apps come to life, we’re convinced it’s also becoming content: fast to create, easy to discard, and deeply personal. In fact, we even released a Feed where every post is a working app you can remix, rebuild, or discard.<p>We think we're entering the age of disposable software, and AI app builders is where that shift comes to life.<p>Also happy to answer questions about what we learned from the first 10K apps AMA style.