I went off the deep end with AI

What happens when code becomes the easy part? Excitement, dread, and what this means for my solo-run business.

For the past few weeks I ran an experiment: build something real, almost entirely with AI. Here’s what I learned, what’s shifting in how I think about code, and what I’m honestly terrified about.

I’m Joe Masilotti, author of Hotwire Native for Rails Developers. This newsletter is where I share weekly thoughts on Hotwire Native, indie consulting, and doing great work while staying present at home.

---

Building a real product with AI

I just launched PurchaseKit, my take on in-app purchases for iOS and Android apps. And I built more than 90% of it with AI.

This isn’t a trivial product. It has multiple moving parts: an iOS package, an Android package, a Ruby gem, and a hosted SaaS that coordinates between all the pieces. I built out the original iOS and Android proof-of-concepts, then let AI rip on pretty much everything else. I took a very iterative approach, having it build small features one at a time. Not too different from how I would build it by hand.

And to my shock (and horror) it did so with amazing ease.

I had to hold its hand a bunch, sure. There were a few hallucinations here and there, mostly related to Hotwire Native (which makes sense since there probably isn’t a lot of content on that out there, besides from yours truly!). But the Rails stuff? The boring CRUD of managing users, teams, billing, even webhooks… it churned those out without much editing or even review on my end. It was pretty darn close to perfect.

There have been bugs, sure. I only launched this earlier this month. But the fixes aren’t too far from copying a Honeybadger stack trace and dropping it into Claude Code. I can have a fix done in a few minutes without ever needing to know the context (which maybe isn’t actually a good thing).

I make sure a regression test is created first, then the AI can verify the fix by making sure that test passes. I like this approach because I can practically guarantee the issue will be fixed, instead of false positive test runs.

Code is no longer the hard part

Folks are always saying that code was never the hard part. And, as a developer, I always dismissed that. “They don’t know what they’re talking about, of course code is hard!”

Well, I think I’ve flipped on that.

Writing code is no longer the bottleneck. Coming up with what to code and how to get it into the right people’s hands is relatively harder than ever.

My 2026 mindset is to value code a little less. Like giving more stuff away for free. The iOS, Android, and gem for PurchaseKit are all MIT licensed, even though I originally planned a custom license that wouldn’t allow you to use it outside of a PurchaseKit integration.

Code is cheap now. Knowing what to build, how to piece it all together, where to host it reliably, how to provide support when things eventually break… all the business stuff is the hard part. It always has been, but code just took up so much of my time that it was hard to realize that.

What this means for my consulting business

For the past 10+ years I’ve helped businesses bring their Rails apps to iOS and Android with Hotwire Native. I’ve shipped 25+ apps to the App Store and Google Play this way.

So what does this mean for a consultant like myself? Where I “sell” code to businesses for money?

Honestly, I’m not quite sure.

I’m still going to offer “Zero to App Store” services. I think there are lots of businesses out there that don’t want to do this on their own, even with the help of AI. They value someone with a track record. They know what they get will be top quality, will work, will be reliable… and most importantly, if something goes wrong they know exactly who to call: me.

I’d be lying to myself if I said that this will continue to be 100% of my consulting business forever.

I’m going to need to shift to something that vibes better with agent-first coding. Am I offering a “Joe AI” that has all of my 10+ years of experience distilled into a GPT? Or a starter kit that is so well documented that AI can’t possibly mess something up? Or just offering more advisory contracts where teams continue to build and I watch over their shoulder, review PRs, and make sure they don’t shoot themselves in the foot?

I’m not sure yet.

But I’m excited, and a bit terrified, to experiment a bunch this year.

What happens when I’m not typing

Here’s something I didn’t expect: I’m slowly becoming re-addicted to social media.

Every time Claude does work I have the urge, and usually give in, to checking on something. Sometimes I fool myself and pretend that checking email for the 100th time is “productive”. But who am I kidding? And don’t get me started on how often I’ve refreshed Twitter…

I need to figure out something to do in the “down time” when an agent works. I’ve been experimenting with kicking off a new agent in another terminal tab. But I find that I lose context too quickly if the projects aren’t closely related. I also absolutely cannot write anything of substance or value in 30 second increments so that’s off the table.

Maybe I’ll just bring a book to my coworking space… and be that psycho who reads with his laptop open right behind the pages.

I’m not sure if this is how I’ll work forever. This was an experiment, and I’m still processing what it means.

---

Are you feeling this weird mix of excitement and dread too? Leave a comment, I’d love to hear what you’re grappling with.

And if you want to try AI-assisted coding yourself, here are a few free Claude Code invites.