Should I build my own app with AI or buy an existing one?

Build when you have a specific workflow or preference that off-the-shelf apps only half meet and the stakes of a bug are low. Buy when the tool is mission-critical, needs support or security guarantees, or already does what you need. A custom app only handles the situations you built it for, and if it breaks, fixing it is on you.

Can a non-developer build a custom app with AI?

Yes. I am not a trained developer. I describe what I want in plain language to Claude in voice mode, ask it for a build plan, hand that plan to Claude Code, and refine the result through trial and error. It worked for a book tracker, and a document scanner is about halfway there.

What is the best way to use AI when building software?

As a strategic partner. You supply the big picture of what you want, and the AI surfaces obstacles you would not think of. In my book tracker, Claude anticipated that an API search could time out and built around it before I ran into the error.

What questions should I ask before building my own version of an app?

Two: which features would I actually use, and which features do I wish it had? Those answers become the specifics you give the AI to brainstorm a better version and write a build plan.

What are the downsides of building your own AI-built app?

If something breaks, you have to fix it, and your version only covers the situations you designed it for. That is the cost of software made for one person.

What does it mean that AI is moving us from apps built for everyone to experiences built for one person?

Getting software shaped around your own workflow used to require hiring a developer or learning to code, so most people adapted their habits to the closest app. With AI, the app can adapt to you instead, including small preferences a mass-market tool has to guess at.

I Used to Look for the Closest App. Now I Try to Build My Own.

If you have a specific workflow that off-the-shelf apps only half fit, AI now makes it realistic to build your own version instead of searching for the closest app. I describe what I want to Claude, turn that into a build plan, and hand it to Claude Code.

I've reached the point where my first thought when I come across a new app is "can I build my own version of this?"

The book tracker

One recent example: a book tracker app.

I came across an ad for an app that cataloged your books by scanning the barcodes with a little device. I had already tried to start putting together a list of books I owned, but there was no organization to it. And by list I mean a collection of titles spread across my Notes app, Evernote, and screenshots of book covers in my photo library.

How I decide

So when I come across something like that I can really use, I ask myself two questions:

Which features would I actually use?
Are there features that I wish it had?

Then I open Claude in voice mode, tell it what I want the app to do, and ask it to brainstorm a better version with me. When I've got the specifics nailed down, I ask it for a build plan, hand that to Claude Code, and see what comes back.

For a book tracker app, I wanted the ability to scan a barcode with my phone, keep a catalog, import what I already had, and get a "you already own this" warning when I'm standing in a bookstore.

I'm not a trained developer, so I'm not writing a formal spec. I'm describing what I want in plain language and letting the back and forth with Claude sharpen it.

After a little bit of trial and error, I ended up with exactly what I had described. The trial and error came down to three things:

Barcodes. Connecting the right APIs so the app could read as many barcodes as possible.
Control. Making sure a scanned book isn't added to my library automatically unless I toggle that feature on.
Search. Being able to search what I already had by title or author.

One thing stood out during this process. Claude knew that a search using an API could time out and give me an error message, so it foresaw the problem and built around it before I ever ran into it.

The sweet spot for using AI is as a strategic partner. I'm giving it the big picture of what I want to build, and it's giving me the potential obstacles that I either don't see or would never think about.

The document scanner

I'm doing the same thing with Genius Scan, the document scanning app. There have been too many times that I've asked someone to send me a copy of a power of attorney or their LLC articles of organization, and what I get back is a group of photos of the pages lying in their lap, with part of their leg that I wanted to crop out.

I wanted a tool that not only converts and crops pictures of documents into a decent PDF, but also integrates with my existing workflows, knows where the documents need to go, and has a link that can be emailed or texted to make it as easy as possible for someone to use to send me copies with their phone.

It's not ready for primetime yet, but it's about halfway there. The cropping and converting to PDF works. The parsing feature needs more fine-tuning. Because I want it to organize and file the scans automatically, it needs to be able to read the documents accurately.

The tradeoffs

This process certainly has its tradeoffs. If something breaks, it's on me to fix it, and my version only handles the situations I built it for. That's the cost of an app made for one person.

The bigger shift

Until recently, getting software built around the way you work meant hiring a developer or learning to code yourself. For most of us that wasn't realistic, so we picked the app that came closest and adjusted our habits to fit it.

Now the app can adjust to me. A book tracker that waits for my okay before adding a scan to my library. A scanning app that knows where a power of attorney needs to go and gives the person sending it a link they can use from their phone. Those are small preferences, and they are the kind of thing a tool built for everyone may not include.

AI has let me flip my thinking. I used to look for an app that came the closest to doing what I needed. Now I see if I can build a personalized version.

And I think that fits into a bigger trend: AI is moving us from apps and tools built for everyone to experiences built for one person.