Notes on making small apps with AI
About eight weeks ago, I was very excited about improving my bootstrapper and imagining a future involving a lot of personal, bespoke app development. I was optimistic about making lots of small, useful apps ("many of my spreadsheets should be custom webapps, and many things that are not spreadsheets should also be custom webapps").
I suspect this was wrong, for a few reasons:
- Narrow subject matter doesn't mean that the app will be simple or easy to make. Email clients and Letterboxd are good examples of how apps with narrow breadth nevertheless need to handle a lot of complexity, because they deal with real-world things, and all real-world things have a lot of complexity you need to model if you want to handle them usefully in software. Even with huge efficiency gains from AI, there's still a lot of work to do, not least because discovering user questions and finding good answers to them will involve a lot of irreducibly human work for the foreseeable future.
- If the app doesn't need to reflect some slice of reality with high fidelity, or for whatever other reason is even more limited in scope, some general-purpose app is probably better. So, for example, a sufficiently simple note-taking app will probably lose out to Drafts or Apple's Notes app. Very low latency, robust integrations, and similar benefits are usually more important than mini-app-specific tweaks in these cases.
- If you want to use many such apps programmatically, you'll run into the delegation problem.
I do use a lot of my own software every day (for flashcarding, tracking my reading, and managing recipes), and I love being able to prototype quickly, but I'm no longer optimistic that my digital future will involve so many of my own apps.