What I learned along the way

As I said earlier, I didn’t start this project to reinvent puzzle games. I started it to really understand where we are with agentic coding and what lessons I could take away from driving something all the way to production.

That experiment taught me a few lessons along the way. If you’re thinking about developing something similar, these are the takeaways I wish I had going in.

Start with a full vision

I built Puzoozle iteratively, adding features as ideas came up. One feature led to another. I started by letting users create puzzles and share them. Then I added leaderboards. Then I decided to add friend groups. Then someone asked me to add daily puzzles of the day. Then the drawing tools needed more features.

It worked. The app got built. But it wasn’t efficient. I spent a lot of time restructuring because I hadn’t planned the architecture up front. Friend groups were over here in the code, but leaderboards were structured differently over there. When I added a new feature, I’d realize it should integrate with something I’d built three features ago, so I’d have to go back and refactor.

If I had written a complete PRD (product requirements document) upfront and mapped the full architecture before building, I would have gotten to the finish line much faster. I would have seen how all the pieces fit together. I would have built the database schema right the first time. I would have structured the user flows correctly from day one.

AI executes incredibly quickly, and Claude Code can write in an hour what might have taken me a day or two. That speed is amazing, but it also means bad architectural decisions compound faster.

You can iterate your way to a finished product (like I did), but you’ll waste a lot of time. Planning still matters. It may matter more now because it’s so easy to build extra features; you will tend to build a lot of extra coolness and go down fun paths without a good plan.

Go all the way to production

I deliberately pushed this project through the App Store submission process. I wanted to find out where the process breaks down, where AI capabilities end, and what still needs development.

Many people are experimenting with agentic coding right now. I see it in the developer community online and amongst many of my friends. People are building impressive demos, cool prototypes, and neat features.

But the number of them that are finishing and shipping is much smaller. Like never before, you have this sense that anything is possible to build. You start a project, and before it’s complete, your mind is quickly on the idea for the next project.

It’s addictive.

As someone who just finished a product, I’m here to tell you there’s a lot more to learn by taking it all the way. The deployment requirements. Seeing users use your products. Reporting issues. Fixing bugs. All of these taught me additional lessons about using AI to develop.

Claude Code handled mobile deployment surprisingly well. But you don’t see the friction until you push all the way through it. You don’t find the limitations until you hit them.

Starting five projects is easy. Finishing one project teaches you the rest of the story.

So my advice is this: pick one thing and run with it. Don’t get distracted by the next shiny idea. Push through deployment. Submit it to the store. Deal with deployment, the user feedback, and the support of code while developing new features, all from an agentic coding world.

There’s a lot of education when you take it all the way.

Fundamentals still matter

I’ve watched teammates at Rōnin build genuinely impressive things with Claude Code despite having little to no coding background. Chuck Harris, our VP of Client Relations, has built multiple functional apps. Chris Bybee, our Director of Operations, who hasn’t coded seriously in years, jumped back in and has been building complex features. One of our Business Analysts, Marcia Clark, went from zero coding experience to building working applications.

That alone is remarkable. The fact that you can have a conversation with AI and produce functional software without typing syntax is amazing when you step back and think about it.

But there’s still a ceiling if you don’t understand the fundamentals.

I have had multiple sessions with Claude Code now, where we talked through big architectural decisions. I’ve noticed that, when talking with non-traditional agentic coders, the mentality is to simply ask Claude Code for a recommendation and run with it.

This can take you very far without really understanding how it’s solving the problem or the potential trade-offs. For complex enterprise solutions with security, governance, and integration concerns, there remains a need for an experienced development mind to conduct these interviews. Not to write the code, but to provide the right context to the agentic coding tool.

The roles are shifting. Development is becoming more about being the technical brain that guides the AI toward the right solution. The job is moving from “write code” to “ensure the right code gets written.”

Is It possible to build a mobile App without coding? You bet.

Puzoozle is live in the App Store right now. You can download it here and try it yourself.

Every feature inside it, every engagement, every line of code running on the backend, was generated through conversation. I described what I wanted. Claude Code built it.

Ten years ago, building Puzoozle would have required months of development time. Five years ago, maybe a month with modern frameworks and tools. Today, I built it in scattered evenings over a few weeks without writing any code myself.

The barrier between “I have an idea” and “I built it and shipped it” is thinner than it’s ever been in the history of software development. The gap between idea and execution isn’t just shrinking; it’s closing. It’s disappearing. So, the question isn’t whether you can build something anymore. The question isn’t even what you’re going to build.

The question is: how much longer are you going to wait?