AI Is Not a Coding Tool. It’s a New Operating System.
After 6 months of daily use, I've stopped thinking of AI as a code generator.
We still talk about tools like Claude Code as if they’re code generators. The name says it: code. Generative AI for producing code. But after 6 months of working with these tools daily, I’ve shifted how I think about them entirely. They’re not coding tools. They’re a new kind of computer operating system.
Think about what you do on a computer today. You open files. You edit files. You move things around. You search for things. With AI, you don’t do any of that anymore. Your interface is speech. The output is visual: text, interactive interfaces, whatever the task requires. The whole paradigm of pressing buttons and modifying files is going away.
One of the most interesting unlocks I’ve seen in the past 6 months is how AI operates your device on its own. It can install software. It knows how to use virtually every tool out there. You don’t need to learn any of it. This was a big deal for me personally. I was fluent at coding, but I’d taken close to a 20 year pause from active software development. Coming back, I realized I didn’t need to go study every new technology or read documentation on how to use specific tools. I just needed to know what a tool could achieve. Then I’d tell Claude Code: I want this, use this tool. And it would do it.
I’ve been editing video with Claude Code since November. I discovered FFMPEG, which is a command line video editing tool that can do basically anything. I’ve done green screen work with it. I take a video clip of myself on my phone, bring it to the computer, and ask Claude Code to put a black hole behind my head so I’m hovering in space. It does that.
This is what I mean by operating system. The change isn’t about producing code. It’s about how you interact with the computer.
The second thing I want to focus on here is QA. Quality assurance for games and mobile apps. This has been a really big unlock, and I cover it in my AI game dev course.
Here’s the typical workflow when you’re building a game with Claude Code. The code is ready. The default next step is: install it on your phone and try it out. Or have someone on your team, a QA person, test it. They file bug reports. There’s a whole rigmarole of humans interacting with the build. But now, the first steps of automating this process are at our fingertips.
I created a chess game. Then I asked Claude Code to play chess against the computer on the Android simulator. It can simulate taps on the screen at specific coordinates. It takes screenshots and determines where to tap next. It was fluent at playing chess. More importantly, it was testing that things work.
This is a simple game, sure. But the principle scales. How we produce bugs, how we reproduce bugs, how we navigate different screens and gameplay elements: all of that can be textually recorded, the same way we’ve done QA for years. The difference is that AI is now doing the work.
The challenges are still there. Think about a game like Hexa Sort, where hex tiles rapidly move around the screen from pile to pile. AI takes a screenshot every 5 seconds or so. That means it’s still limited to turn-based interactions. Fast animations are hard to evaluate from static screenshots.
But developments are happening. Android simulators now support video recording. So the AI plays the chess game, which has animations that need testing. It records a video, breaks it down frame by frame, maybe 5 frames per second, and analyzes each frame to see what the animation is doing.
We’re getting to a point where you don’t need to assume this is still a humans-only area. I think the only place humans will be needed soon is the creative work: having the taste to understand what works well in a game. All the non-creative work has now, or will in the next 12 months, be handled by AI.