The models are good now - try again.

Hi friend!

Today I have a little nudge for you:

If you've been using AI models in your work for more than 6 months, you likely have a stack of things you tried that didn't quite work.

Maybe it’s an app you tried to build, a presentation or image you tried to generate, or an entire workflow for automating your tax filings.

My nudge is simple: Try again.
​
Use one of the newest models and harnesses.

I had this experience myself last week. As you might know, one of my side projects is Readbetter.io: a service that lets users deliver web content to their Kindles as native EPUBs.

For years, I’d wanted to build an iOS companion app for it, but I never got around to it.

Then, six months ago, while testing Opus 4.6 in Claude Code, I gave it the task: Build a simple iOS app connected to my existing stack, with a limited set of features.

It took a whole lot of back-and-forth and re-steering. The result kind of worked, but it wasn’t close to production-ready. The entire codebase was a mess, so I abandoned the idea again.

Last Friday, just as I was about to finish my workday, something brought the idea back to mind.

I selected Grok 4.5 High (Fast) in Cursor—a very capable, fast, and inexpensive model with a Cursor subscription—and sent it a simple prompt:

I had it ask me a few questions, after which it generated a solid technical plan for a native Swift app.
​
I let it build - and about 30 minutes later, I had a functioning app with exactly the feature set I wanted.

During those same 30 minutes, the model also walked me through every step required to set up a new app in the Apple Developer portal. Figuring that out on my own would easily have taken me two hours in the pre-AI days.

It almost feels boring to write that an AI model built a fully functional native iOS app in 30 minutes.
​
We take these capabilities for granted so quickly. Just six months ago, this was barely possible at this level.

And what’s true for an iOS app is true for many other things. Over the next few weeks, I’ll certainly be pulling several projects out of my “tried but failed” stack.

Try it for yourself, and let me know how it goes. I’d be thrilled to hear about your experience.

​
Thanks for reading,
Robert