Today, it is hard to escape LLM bots and the endless slop they emit, but the Linux kernel might be largely safe... for now.

Linus Torvalds has spoken up on the contentious topic of LLM-assisted software development. Despite his previous guardedly positive stance, for now, he seems to have come out strongly against it in the context he cares about the most – developing code for the Linux kernel. But he also doesn't want the documentation to become a political battlefield over this point.

He was responding to a message from Oracle-affiliated kernel developer Lorenzo Stokes, which seems to us to be guardedly anti-LLM: "Thinking LLMs are 'just another tool' is to say effectively that the kernel is immune from this. Which seems to me a silly position."

Torvalds replied:

Rather than try to paraphrase the great man, we thought we'd just give you his own words. However, we do also face the slight issue that it's not entirely clear to us what Torvalds's position here really is.

Stokes was replying to an email from Dave Hansen. As Linux benchmark and commentary site Phoronix reported in November, a team is working on a set of clear, unambiguous guidelines concerning LLM-bot-assisted contributions to the kernel.

It's a pressing issue. People are already using LLM coding assistants to work on kernel code: for instance, last year, Wikimedia developer Dmitry Brant blogged about Using Claude Code to modernize a 25-year-old kernel driver.

The Register has reported on Torvalds's various comments on LLM-bot-assisted coding several times in recent years. In 2024, he said that 90 percent of AI marketing is hype. (To be honest, The Reg FOSS desk thinks that's generous: we'd be happy to learn it was as low as 90 per cent.) A year later, he commented that he was OK with vibe coding as long as it's not used for anything that matters, which was a little unexpected. The Reg's own Rupert Goodwins begged to differ, writing "Vibe coding: What is it good for? Absolutely nothing (Sorry, Linus)".

• What the Linux desktop really needs to challenge Windows
• What if Linux ran Windows… and meant it? Meet Loss32
• GNOME dev gives fans of Linux's middle-click paste the middle finger
• Finally – a terminal solution to the browser wars

For now, LLM coding assistants are so popular with so many that Torvalds is right: if the kernel flatly prohibits their use, then they'll get used anyway. They can and do emit code-like text, and lots of it, and currently they're cheap. Ban them, and they'll still be used – the users will just deny it.

Of course, they might not remain cheap. This vulture strongly suspects that another AI winter is coming. The first one was circa 1984, and the second about a decade later, with the end of the Fifth Generation Computer Systems project sponsored by Japan's Ministry of International Trade and Industry (MITI). Now, all that is left of the FGCS is a web museum.

We suspect that the third and biggest AI winter is imminent. Some analysts think so too, while others feel that what's left behind will remain valuable. Right now, the "generative AI" industry is spending vast amounts to subsidize these models and their use, propping up the US economy in the hope of future trillion-dollar-scale profitability. If that doesn't happen and the industry collapses, the coding-assistant advocates may suddenly find that plain old human brainpower is much cheaper than hundreds of billions of dollars' worth of datacentres.

Some large model prices are already increasing sharply. Put another few zeros on the end of the pricing and we suspect people may get much less keen on using them. This may yet prove to be a self-correcting issue, given a few years. ®