Developing an entirely custom operating system using Claude Code

I've been writing toy kernels and working on operating system projects since my childhood, and it's partly how I learned C. That includes this project, MontaukOS, which I started early in 2025, where I wrote a lot of the fundamental kernel code by hand. Because other things came up, I eventually lost interest in this project for a while, and let it sit for a few months.

When Claude Opus 4.6 came out, I was introduced to development through Claude Code (and agentic AI more broadly) for the first time. I figured I'd pick up this project sitting dormant on my hard drive and try to use the model to see how good it would be at low-level code, and learn how it worked.

I did this for quite a while, and found it to be much, much better than I expected. Eventually, I found that I was writing very little code by hand.

Over the months went from a toy kernel (not doing very much) to something I was dual-booting and running day-to-day on my laptop. Claude Opus 4.6 (among other models) produced a networking stack, disk & filesystem drivers, a graphics driver for Intel, a desktop environment (which I could never do, because I'm terrible with graphics code), a PDF viewer (maybe the most surprising one for me) and much more.

With the recent release of Claude Fable, this has gone quite wild. I'd been iterating with Opus on the OS's Bluetooth audio stack for weeks to no avail - and within just a few prompts, Fable was able to get my Bluetooth headphones fully working! It's absolutely incredible what the model can do. It'll be a shame to lose the model soon.

Screenshot of MontaukOS running on my laptop, playing music via Bluetooth headphones, displaying a PDF, processes, filesystem (bare-metal).

Screenshot of MontaukOS in a VM running a \"Hello, World\" program using the experimental Lua port

ISO and docs, if anyone's interested: https://montaukos.org/downloads.html Git/source code: https://git.montaukos.org/daniel/MontaukOS