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.
Here are the three biggest things we learned:
1. Reducing the agent's scope of action improves outcomes (significantly)
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.
2. The first prompt matters. A lot.
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.
3. Most apps were small + personal + temporary.
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.
This led us to a big shift in our thinking:
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.
We think we're entering the age of disposable software, and AI app builders is where that shift comes to life.
Also happy to answer questions about what we learned from the first 10K apps AMA style.
kingkongjaffa•2mo ago
This is a fascinating thought. I wonder if there's some disconnect between good design and the immediacy of building something that solves exactly the thing you need to solve at the time.
What I mean is, when you first build something, it probably does what users need, but there's always some rough edges. Frankly out of 10,000 throwaway apps built, I'm going to guess probably less than 10 have been built with good design and taste.
It's like the difference between a TODO MVP toy app to track tasks, vs something like Linear which is beautifully designed.
Both probably have their place I think.
For my work I'm not sure I want my tools to be so discardable personally. I want to use predictable, well designed tools that have had their rough edges sanded down through iteratively reducing the micro-frictions I have in my day to day job. Behind every great product experience there's usually someone obsessing over a specific pain point and motivated to make something great.
Toy throwaway apps can't replace human thinking time and experience using a tool over months and years.
For personal and one time problems, toy apps can absolutely get the job done, and most people are willing to overlook the rough edges.