I was somewhat conflicted seeing the way the OSdev community responds to LLM usage.
I understand the conflict: the point of doing OSdev is that you're learning the low level hardware details and directly commanding the machine from the lowest level accessible. LLM usage is at odds with this.
But, on the other hand, the primary issue with OSdev on modern hardware is the complete lack of documentation for most hardware. A lone developer simply does not have enough time to reverse engineer from Linux or BSD drivers to achieve much hardware support. Plus, often you run into hardware quirks that are incredibly painful to debug and offer no real learning opportunity.
Recently I revived a ~8 year old modular kernel project I had written. It was at the point of being able to setup an x86_64 CPU, load some drivers and enter usermode. Using an ESP8266 board I setup a little debugging harness for a AtomicPi devboard I had laying around, and set it all up so Claude could control and upload code to it. It could read the serial port, ping the board over ethernet and power cycle it. From that point I let it work almost autonomously to implement a USB stack, finish the network stack, implement drivers etc.
It worked well, and for the most part I didn't care about the results. I don't have the time to commit to such a big project now and of course, the satisfaction of doing something myself (besides building the harness) was gone.
But a thing that stood out to me was, when trying to implement the ethernet driver, the chip in question did not have documentation, but it was known to be similar to another well documented chip. Claude was able to spend a few hours just trying different things, rebooting the board as needed until the differences were sorted out and it was able to ping the board. I feel this kind of application would be pretty relevant for even hobbyist communities like OSdev.
We all have limited time, and while LLM usage for everything certainly defeats the point, targeted usage for tedium that has very little value overall, should be acceptable, and I can't help but think lesser of devs I otherwise respect, when they insist on being dogmatic even when it doesn't contribute to the hobby.
Isn’t that the point of a hobby? To enjoy the path and not the result? To learn something new.
I understand using LLMs as tools to help you with your hobby, but I think when “Claude spends hours debugging it” it probably is no longer a hobby, but an outsourced job.