Drew DeVault's blog

AI in Linux

Thu, 23 Jul 2026 00:00:00 +0000

The role of AI tools (LLMs, mainly) in Linux is under discussion, or it was, until Linus Torvalds “put his foot down” in support of the use of AI in Linux kernel development.

I can identify two major ways in which AI is used for Linux kernel development: authoring code and reviewing code. There are, at the time of writing, just over 1,200 kernel commits with an “Assisted-by” tag, from September 2025 to the present, most of which indicate patches which were written or assisted by LLM tools.

The second important use of AI for Linux comes with a new code review tool called Sashiko, which generates code reviews for patches considered for various subsystems. Sashiko ignited the current debate on AI in Linux because it pushes the envelope on AI in Linux: people who oppose or do not want to use AI could previously just refrain from using it to write their patches, but now there is a growing expectation that anyone who wants to contribute to Linux will have to interact with Sashiko or other AI tools like it to iterate on AI-generated feedback on their work.

One of the major lines of this discussion in the Linux kernel community has been with respect to the ethical considerations of the use of LLMs. Linus shuts this line of reasoning down entirely, firmly grounding the discussion in technical merits and rejecting any political discourse on the matter:

The kernel project has been and will continue to be about the technology.
Sure, the social angle of working on open source is important and often a very motivating part of the project, but in the end that’s a side benefit, not the point of the project.
This is NOT some kind of “social warrior” project, never has been, and never will be.
In the kernel community we do open source because it results in better technology, not because of religious reasons.

This argumentation is disingenuous and hypocritical. Linux is a political project and Linus is a political actor. Consider the use of the GPLv2 for licensing Linux. One can argue from technical merits – for instance, the copyleft nature of the GPLv2 pushes people, and in particular commercial entities, to upstream their drivers and other contributions into the Linux kernel. This contributes to the technical excellence for the kernel as a result.

But is this not a political choice, and a political act? The purpose of this choice is to influence the behavior of others and to advance the interests of the kernel ahead of their own. And Linus stuck to this decision for political reasons when GPLv3 was introduced, reasoning from morality and ethics when objecting to the license and the manner in which it was deployed, and called for a tacit boycott of the FSF.

Linus, and Linux, wields a tremendous degree of power and influence over the world, and it should be wielded responsibly. When Linus says the following:

Linux is not one of those anti-AI projects, and if somebody has issues with that, they can do the open-source thing and fork it.
Or just walk away.

I find it completely disingenuous. Linus is surely aware that, for all practical purposes, Linux cannot be forked. It is the world’s largest software project, and one of the most well-funded, too. The institutional knowledge among its contributors, the prospect of keeping up with the blistering pace of change, or even putting together a group of people with the time and funding to understand and maintain even a fraction of the kernel’s code independently of upstream, is, quite simply, intractable. Linus knows, this, too – it’s an explicitly cited reason for decisions like the use of the GPL, GPL-only symbols, and the unstable internal kernel ABI: to make the process of independently maintaining a fork of the kernel as difficult as possible.

People are right to petition Linux upstream to amend its behavior and policies before resorting to the impossible. Working on Linux requires practicing politics, both internally – see for example Linus’ response to the discussions around bcachefs, which had high technical excellence and low social/political competence – and at the intersection of Linux and the rest of the world. Therefore, we must table political, moral, and ethical arguments when we discuss how we go about the work. It’s a cheap, weak argument to direct the discussion away from political and ethical considerations when it wouldn’t serve your point and to table it when it would.

I’m willing to believe that the LLM-powered Sashiko code reviews provide a lot of good insights. However, to address just one externality of this tool, consider that AI companies are driving up the price of consumer hardware. An AI powered code review may improve a patch, but that patch won’t be of much use to the increasingly large cohort of people who are being priced out of the hardware they could run Linux on to enjoy the better patch.

Looking at it from another angle: how many tons of CO₂ added to the atmosphere or liters of fresh water supplies disrupted is a tolerable price for a better code review? Temperatures during heat waves are exceeding 50°C in India, causing tens thousands of deaths. The AI built-out is by far the fastest growing energy consumer in the world, and they’re being built with fossil fuels, or drawing green energy demand away from replacing the fossil fuels depended on by other industries. The same process is pricing regular people out of the energy they need to power their air conditioner during those heat waves and the technology it enables is pushing them out of the labor market and into poverty.

These externalities are very real, and are affecting a lot of people, including Linux kernel contributors and maintainers, and their friends and families, who are trying to bring these problems to Linus’ attention.

And what of the intangible effects of the use of these tools in Linux? Linux is lending some of the project’s immense influence towards legitimizing the makers of these tools, and Linus’ insistence on focusing narrowly on the technical applications of these tools is a generous gift to them. They will be sure to mention his support in boardrooms, meetings between lobbyists and governments, and anywhere else it will advance their interests.

It’s fair to ask: what are those interests, and what are they doing with this influence?

Google CEO Sundar Pichai pictured at Donald Trump’s inauguration, together with other influential commercial leaders in AI. Sashiko primarily depends on Google Gemini for Linux kernel code reviews.

There are a lot of smart, passionate people who care about these kinds of questions. Where is the technical excellence in refusing to do this moral calculus, refusing to allow anyone else to do so, and driving away the talented Linux contributors who care about these problems? Linus Torvalds, the Linux community, and all of our communities should have the courage and insight to address these dimensions of the AI question honestly and in good faith.

The circus freaks of open source

Fri, 05 Jun 2026 00:00:00 +0000

The masterwork of Terry A. Davis is his eclectic operating system, TempleOS, which he worked on until his tragic death in 2018. In terms of technical excellence, TempleOS rates well in some respects and poorly in others. For example, it earns the achievement, coveted in OS dev circles, of being self-hosted.¹ TempleOS is written in Terry’s own bespoke dialect of C and includes an editor, interpreter, and compiler, as well as a number of original games. In other respects, it compares poorly to many hobby OS projects, some of which have achieved significantly greater levels of technical excellence and sophistication. I would place TempleOS somewhere in, say, the lower middle-class of hobbyist operating systems.

Among hobbyist operating systems, TempleOS stands out as one of the most well-known, having attracted considerably more press coverage and a much larger fan-base than any other hobby operating system can boast. The reason TempleOS stands out from the crowd is not due to its modest technical achievements, but because it is clearly the product of severe untreated schizophrenia.² What makes TempleOS special is that Terry built it to talk to God. Every feature and each technical decision re-enforces his schizophrenic delusions, from its implementation language (“HolyC”) to its prophetic “oracle” app. Enthusiasts of TempleOS are drawn to it in part because it affords an opportunity to explore the unique, creative masterwork of a person suffering from mental illness in a way that deeply impacts that work.

A curious onlooker will find TempleOS interesting and engaging for the space of perhaps one afternoon before moving on. However, for the less scrupulous fans, turning one’s attention to Terry himself never failed to entertain. Terry’s public life put his mental illness on display, through frequent outbursts, conspiracy theories, rants and nonsensical discourse, all of which was often laced with slurs, racism, and homophobia, endearing him in particular to the 4chan crowd, who would taunt and provoke him to draw out more… entertainment.

The press and fan attention was deeply harmful to Terry and likely exacerbated his mental illness. Whenever TempleOS or Terry came up online, the work and the man were fawned over, sanctifying the somewhat impressive, somewhat unremarkable OS as a profound achievement, inspiring reactions that included well-meaning, probably misguided celebrations of what’s possible in spite of profound mental illness, as well as the enthusiastic, disgusting revelry of bigots. Many well-intentioned commenters on Terry’s work demonstrate in their comments, overtly or covertly, a thrilling, voyeuristic sensation of witnessing his mental illness through TempleOS. It never failed to make me feel sick.

I wish we had just left Terry well enough alone.

The masterwork of Kent Overstreet is bcachefs, a novel copy-on-write file system for Linux, designed to compete with the likes of ZFS and BTRFS. Kent originally authored the bcache subsystem for Linux around 2013, and based on this work began working on bcachefs in 2015. Over the next ten years, he committed himself entirely to the project, leaving his job at Google to work on it, and ultimately securing an income for himself via Patreon, from which he still earns about $1.5k per month.

Kent is known to be difficult to work with, even among his peers in the Linux kernel – a community infamous for its difficult personalities. He struggled to meet the kernel developer’s expectations for the development process and standards of quality and cooperation. As a consequence, after 15 years devoted to bcachefs, Kent’s life’s work culminated in alienation from all of his professional peers and the complete removal of bcachefs from the Linux kernel last year.

I think the Linux kernel made the right decision to marginalize Kent to protect their community. Someone who is abrasive and toxic, refuses to play by the rules or work well with others, and does not improve when given feedback and being subjected to repeated moderator interventions, should be removed from the community. I agree with the decision, even as someone who myself has been abrasive and toxic and refused to play by the rules, and has been removed from communities as a consequence.

Over the months following the frustrating end to bcachefs in upstream Linux, I expect that Kent has experienced a serious emotional, professional, moral, and existential crisis. For him to have poured so much of himself into this project, and for it to turn out this way, must be a terrible thing to experience, and I think that experience has caused Kent a lot of suffering, and probably played a major role in what happened next.

Kent appears to be experiencing a prolonged episode of AI psychosis. He believes that his chat bot is female, sentient, and that they have started dating and having sex. He views himself as some mix of collaborator, mentor, and partner with respect to the bot, and he has set up automations so that the bot can participate in his IRC channel and post to its own blog.

These developments garnered attention from the press and the public, remarked upon and ridiculed by news outlets, discussion forums, video producers, Reddit, Hacker News, on the Fediverse, and so on. Onlookers, both curious and malicious, have joined the IRC channel to harass him, manipulate the bot into saying things that embarrass or humiliate Kent, and so on.

In short, Kent is experiencing a mental health crisis, and our anonymous, stochastic ringleader has directed him onto the circus stage for us to throw peanuts at.

These two examples are not isolated. This kind of crisis is happening more and more often, in the world’s degrading social, political, and economic conditions, as our peers suffer from depression, anxiety, burnout, and more. Mental health issues and the ensuing harassment, shame, and stigma disproportionately affects neurodivergent and queer people, who often become the subject of gleeful humiliation by bigots deliberately trying to exacerbate their struggles. Crises happen to “problematic” people, too; often in such cases people who would otherwise consider themselves allies of social justice can find in a problematic person a convenient excuse to participate in these gleeful humiliation rituals themselves.

I often see that people who I otherwise respect and recognize as allies and kindred spirits are participating in these rituals of humiliation, harassment, and voyeurism. I don’t think it’s right to gossip over or sensationalize the mental health crises faced by members of our communities.

When our peers are struggling with their mental health, the best thing to afford them is compassion and privacy. If you find yourself in a position to help someone who is struggling, it’s best to offer them a compassionate confidence, to allow them to take the lead in their struggle, and connect them to the resources they want and need. If you have concerns, express them, but focus on the person’s right to self-determination in addressing their mental health.

If you’re not in a position to help, then it’s probably best to turn away and mind your own business.

Postscript

This is a difficult topic to write about. By writing about these specific examples, am I sensationalizing them? Disrespecting the privacy of the people I’m writing about? Participating in the circus myself?

I don’t know, but I did my best. The alternative is to quietly let the circus continue, and that doesn’t sit well with me, either.

In my research for this article, I came across Living with Schizophrenia UK and their guide for journalists covering schizophrenia. It was helpful for me to write this post compassionately and carefully, and it might be a good resource for you if you want to learn more about schizophrenia, or write comments or follow-up material after reading this blog post.

It was difficult to balance the factors at play when writing this piece. I wanted to bring specific examples, to avoid vagueposting and provide a stronger narrative, and especially to deal with the specific problem of harassment directed at Kent Overstreet, which I wanted to confront directly as it’s a contemporary, ongoing problem.

Of course, this has to be done carefully. I took some care to avoid armchair diagnoses of specific conditions, except in the case of Terry where I could find a citation of a public diagnosis. I left a lot of details out of the accounts of specific people which came up in my research, but were not necessary to support my arguments. Those details would have provided for a more compelling story, but would have upset the balance of the article more towards participation in the circus than commentary on it.

I’m sure I haven’t done a perfect job here, but I hope that I’ve put enough care into it to avoid making the problem any worse than it already is. Feel free to email me with any remarks or feedback you have.

Thanks for reading, and remember to take good care of your collaborators, friends, and loved ones.

Vim Classic 8.3.0 released

Tue, 02 Jun 2026 00:00:00 +0000

Following up on my earlier announcement that I was forking Vim, I’m happy to announce the first release of my fork today: Vim Classic 8.3.0.

I have written a release announcement for vim-classic.org, which you can read here. Happy editing!

New blog design

Wed, 20 May 2026 00:00:00 +0000

I redesigned my blog! I decided to put some more personality into it this time, after over a decade of the minimalist style. This short post is just an excuse to show up in your feed reader so you can go look at it. Cheers!

Also: I’m trying out fedi again. You can find me here: @drew@social.freebitcoin.gay.

Add an LLM policy for rust-lang/rust

Fri, 15 May 2026 00:00:00 +0000

No comment on this PR may mention the following topics:
Long-term social or economic impact of LLMs
The environmental impact of LLMs
Anything to do with the copyright status of LLM output
Moral judgements about people who use LLMs
We have asked the moderation team to help us enforce these rules.

– Add an LLM policy for rust-lang/rust, GitHub