After 30 AI Apps, Building Is Only Part of the Work

After 30 AI Apps, Building Is Only Part of the Work

AI can make a finished product feel one prompt away. Then I take that step and discover the next one. And the next.

That is the mirage I talk about in this video. After four years of using AI every day and building more than 30 side apps, I keep coming back to the same friction: getting something built is only one part of getting it into somebody's hands.

The small app that keeps growing

A meeting notes taker sounds like a contained project. I wanted one, built it, and had that initial feeling of “this is amazing.”

Then came the fixes. In the video, I describe getting into something like 100 or 200 prompts. The user experience needs work. A button is in the wrong place, or does not work. The app uses the wrong model. The design needs another pass.

There is also everything around the app: an installation process, a website, a download and a certificate. Each item looks like another small task. Together, they make the distance between an exciting prototype and a usable product much more visible.

I do not think a prompt count is a useful target. What interests me is what those repeated prompts reveal. The first working version has left a lot of decisions unresolved.

A download is a different problem

I built a free, local voice-to-text app. I wrote a blog for it, did LinkedIn marketing and made videos. The result I describe here was almost zero downloads and zero feedback.

That is my experience with that effort, rather than a forecast for anyone else's project. I still find it frustrating. Having an app and having people who want to try it are separate things to work through.

I used to think development was the blocker. With AI helping me build, the other parts become harder to ignore: finding the first users, getting discovered and giving someone a reason to care.

Even the marketing creates its own loop. Make an ad, improve it, see whether it worked, try to understand what happened, then make another version. Generating the next attempt does not settle the question of whether it helped.

The success story needs its surrounding context

In the video I push back on the online stories about building an app and making $50,000 a month. My reaction is skeptical. I want to know what sits around the headline.

Was there already an audience? An early client? Money available for marketing? Those are possibilities I raise in the video, not facts I know about a particular person. A revenue claim on its own does not tell me enough about how a product found its users.

The lesson I take from this is that implementation, execution and going to market deserve much more attention than the initial idea.

Taking the pressure down a little

I am sharing this partly to take some of the FOMO out of building with AI. A tool can help me move through a task while leaving plenty of other variables unresolved.

For my next experiment, I want to keep the whole path in view: what I am making, what it takes to use it and how someone will find it. The next prompt can still be useful. I just want to be clearer about which problem it is actually solving.

Watch the original video on YouTube.