If you’ve ever used a 3D printer, you’ll remember that amazing feeling the first time you printed something you’d never built or made yourself. Download a model file, load some plastic filament, press a button, and almost like magic, a three-dimensional object appears. But the result isn’t refined and ready for mass production, and creating a new shape requires more skill than simply pressing a button. Interestingly, today’s AI coding agents feel much the same way.
Since November, I’ve used Cloud Codex and Cloud Opus 4.5 through a personal Cloud Max account to experiment extensively with AI-assisted software development (I’ve also used OpenAI’s Codex in a similar way, though not as often). After fifty projects, I’ll say the obvious: I haven’t had this much fun with computers since learning Basic on my Apple II Plus when I was 9 years old. This opinion comes not as an endorsement but from personal experience: I started this project voluntarily, and I paid out of my own pocket for both OpenAI and Anthropic’s premium AI plans.
Throughout my life, I have dabbled in programming as a utilitarian coder, writing small tools or scripts when needed. In my web development career, I wrote a few small tools from scratch, but I mainly modified other people’s code for my own needs. Since 1990, I have programmed in BASIC, C, Visual Basic, PHP, ASP, Perl, Python, Ruby, MuseCode, and a few others. I’m no expert in any of these languages—I’ve learned just enough to get the job done. I’ve developed my own hobby games over the years using BASIC, the Torque game engine, and Godot, so I have some idea of what makes a good architecture for a modular program that can be expanded over time.
In December, I used Cloud Code to create a multiplayer online clone Katamari Damacy Called “Christmas Roll-Up”.
In December, I used Cloud Code to create a multiplayer online clone Katamari Damacy Called “Christmas Roll-Up”.
Credit: Benj Edwards
Cloud Code, Codex, and Google’s Gemini CLI can do small-scale software wonders. They can create attractive prototypes of simple applications, user interfaces, and even games, but only as long as they borrow patterns from their training data. Like 3D printers, it takes a lot more work to do production-level work. Creating sustainable production code, managing a complex project, or actually designing something new still requires experience, patience, and skills that today’s AI agents themselves can provide.
<a href