Dave, you should be sorry
David Heinemeier Hanssonâs most recent blog post, published around July 21, is called âWolves, sheep, and gypsies.â Iâll get into what it actually says in a second, because it matters. Hereâs how it became news first: DHH posted a screenshot of Claude declining to translate the post into Italian, along with the line âThis is why we need competition and open weights in AI. Imagine a world where only Anthropic sat as the moral arbiter of acceptable speech.â He followed it with a longer post called âIâm sorry, Daveâ - a HAL 9000 reference, casting himself as the reasonable astronaut and the AI as the paternalistic computer locking him out of the pod bay for no good reason. Anil Dash called it out on LinkedIn: DHH, he wrote, âis using the exact same dehumanizing rhetoric against the Romani people in his hateful essays that the literal Nazis did, to the point that even common LLMs like Claude wonât translate the content.â Dashâs conclusion: âI think calling for exterminating a people is still disqualifying and we should draw a line.â
Thatâs a serious claim, and I think it holds up. I want to do two things here. First, show you this isnât new, by walking backward through more than two years of what heâs actually written, under his own name, on his own blog. Second, talk about what itâs like to watch people in my own professional world get excited to work with him anyway, like none of it exists.
What he actually wrote
The post opens with Denmarkâs wolf population and the argument that wolves need to be culled when they threaten livestock and people. Then it extends that logic to Roma people occupying public space in Copenhagen, and lands on this: âWhen wolves get out of control, you shoot them. When gypsies take over public spaces, you deport them. This isnât hard, it isnât cruel. Itâs the basic logic of self-preservation.â (A quick note on language: âgypsyâ is widely considered a slur against Romani people, and Iâm only using it here when quoting DHHâs own words or his postâs title. Everywhere else in this piece, Iâm using âRomaâ or âRomani.â)
Iâm not quoting that for shock value. Iâm quoting it because the âIâm sorry, Daveâ post never actually engages with it. It talks about translation and censorship and AI monopolies, but it never sits with what the sentence says. So letâs sit with it for a second: a named ethnic group, framed as a population that needs âcontrol,â with removal by force presented as the natural, uncomplicated conclusion. Thatâs not a wildlife management post that got a little edgy. Thatâs the whole structure of the argument.
The case heâs actually making about AI
I want to give this part its due, because DHH isnât entirely wrong about the general shape of the problem. His argument, as I read it: translating a text isnât endorsing it, and âwhat if Microsoft Word started refusing to right-align paragraphs it disagreed with?â He points out that a Chinese model, Kimi K2.7, apparently discussed Tiananmen Square without issue while Claude balked at his translation request, and asks, pointedly, âso remind me again which side is supposed to be the authoritarian one?â His conclusion is that we âdesperately need strong open-weight models to protect ourselves against this kind of soft ideological tyranny, which can turn into hard repression real quick if a monopoly status is ever locked in.â
I donât think thatâs a crazy thing to worry about in the abstract. A handful of companies deciding, unilaterally, what billions of people can and canât get help with is a real concentration of power, and Iâd rather live in a world with more model diversity than less. I donât fully know where I land on how much discretion an AI company should have here, and Iâm not going to pretend Iâve got it all figured out. Thatâs a genuinely hard question.
But this isnât a hard case, and using it to make the hard-question argument is the sleight of hand.
Where it falls apart
Translation is not neutral formatting. Right-aligning a paragraph doesnât change what the paragraph argues or who it can reach. Producing a fluent, persuasive Italian version of an essay that concludes ethnic deportation is âthe basic logic of self-preservationâ is not a mechanical task. Itâs extending that argumentâs reach to a new audience who couldnât read it before. Thatâs a meaningfully different ask than âprocess this document I disagree with.â Refusing it isnât an AI company deciding it doesnât like DHHâs politics. Itâs declining to help construct and distribute a specific, well-worn piece of rhetoric.
Thereâs a version of this argument that sounds sharper than it is: what an upside-down world, when a Chinese open-weight model will tell you the facts of Tiananmen Square, but an American frontier model wonât translate a blog post. Itâs a good line. Itâs also comparing two different kinds of act as if they were one kind. Telling someone the truth about a massacre that already happened is disclosure - it adds a fact to the world that some governments would rather stayed buried. Producing a polished, persuasive new translation of an essay arguing an ethnic group should be deported is construction - it adds a fresh, more legible version of an argument that didnât exist in that language before. A model describing 1989 isnât drafting propaganda for a second one. A model translating âthe basic logic of self-preservationâ into Italian is doing rhetorical labor on behalf of an argument that hasnât finished happening yet. Those arenât the same category of ârestriction,â and stacking them side by side to ask âwhich side is authoritarianâ treats disclosure and construction as interchangeable, when the entire ethical weight of the question sits in the difference between them.
Hereâs why I donât think âpopulation that needs controlling, removal is the logical conclusionâ is just one edgy take among many. It has a documented history, and itâs not subtle. In December 1938, Heinrich Himmler issued a decree to âtackle the Gypsy question on the basis of race,â seeking what internal documents literally called a âfinal solution to the Gypsy question.â Thatâs the same phrase later applied to the genocide of Jews. The office coordinating this was formally named the Reich Central Office for Combatting the Gypsy Nuisance. Roma were classified as âasocialâ and âwork-shyâ to justify internment, classifications later used to deny survivors reparations. By 1941, mobile killing units were reporting the âGypsy question resolvedâ after mass shootings. By 1943, roughly 23,000 Roma had been deported to a dedicated âGypsy campâ at Auschwitz-Birkenau. Between 250,000 and 500,000 Roma and Sinti people were murdered. Historians call it the Porajmos, meaning âthe devouring.â
Thatâs the history, but itâs not ancient or abstract. The European Roma Rights Centre documents the same move happening in contemporary European politics, including a Romanian mayor who said putting Roma in a zoo next to monkeys wouldnât produce a difference kids could spot. The ERRCâs word for this pattern is âescalationâ: rhetoric normalizing a population as a problem to be managed isnât separate from the violence that follows, itâs the documented on-ramp to it. DHHâs line follows that exact structure: name the group, frame it as a population problem, present forced removal as the obvious, non-cruel conclusion. Iâm not calling him a Nazi. Iâm telling you the sentence he wrote uses the specific rhetorical technique that, historically, comes right before the thing itâs rehearsing. And a company declining to help spread that technique into a new language is not the same as a company deciding it doesnât like your opinion on remote work.
Working backwards
Wind back about ten months from âwolves, sheep, and gypsies,â to September 15, 2025, and you land on âAs I remember London.â In it, he writes that âLondon is no longer full of native Brits. In 2000, more than sixty percent of the city were native Brits. By 2024, that had dropped to about a third,â and that Copenhagen undergoing the same shift âwould feel completely foreign, of course. Alien, even.â He references âthe Pakistani rape gangs that were left free to terrorize cities like Rotherham and Rochdaleâ and ârampant street thefts,â and says heâs âunwilling to resign the rest of the country to the kind of demographic replacement that befell London.â Iâll let you sit with that phrase, âdemographic replacement,â on your own - itâs not one Iâm putting in his mouth. The same post calls Tommy Robinsonâs march âheartwarmingâ and describes the far-right activistâs supporters as âperfectly normal, peaceful Brits whoâve had enough.â Tommy Robinson, for anyone who doesnât follow British politics closely, founded the anti-Islam English Defence League and has served five prison terms, including for contempt of court.
Wind back another three months, to June 3, 2025, and you land on âa post about his kidsâ school,â where he objected to eight-year-olds attending Gender and Sexuality Alliance meetings with staff during lunch and recess, called it âovert indoctrination,â and campaigned until the school canceled its planned Pride Week materials.
Wind back further still, to March 31, 2024, and you land on âa review of Abigail Shrierâs âBad Therapy,ââ where he approvingly cites her earlier work on âthe dramatic surge in transgender identification, and the medical interventions and experiments that have come with it,â holding it up as an example of someone willing to buck orthodoxy. Somewhere in that same stretch, per an article by writer David Celis arguing Rails needs new governance, DHH also lent his public support to Graham Linehan, one of the more prominent and explicit anti-trans campaigners in the UK.
Thatâs more than two years, read backward in about the time it takes to finish this section. None of it is buried. Itâs on his own blog, under his own name, and other people in Ruby and Rails circles - po-ru.com, tekin.co.uk, and Celis among them - have been documenting the pattern since well before this latest post. âWolves, sheep, and gypsiesâ isnât a departure. Itâs the next sentence in an argument heâs been building for a while, and every sentence before it is dated and public.
Where Omarchy fits
This is exactly why it matters that Omarchy specifically is where a lot of new people are meeting him. Omarchy isnât tied to Rails or Basecamp. Itâs reaching people who have never heard of any of this, who are excited about a nice keyboard-driven desktop and have no reason to have gone looking for what its author writes on his other blog. One is a genuinely well-made piece of software with no politics attached to the install script. The other is the two-plus years above. Nobody has to reckon with the second one to post a screenshot of the first, especially if they donât know it exists.
That gap isnât hypothetical, either: Tedium covered the backlash when Frameworkâs CEO contributed code to the project, with critics arguing that software âhas to be considered in terms of its ethical linesâ no matter how well it solves a real problem. I donât think that argument is fully settled. But itâs the same shape as the one Iâm making here: at some point a personâs actual, stated positions have to factor into whether you keep reaching for the thing they built. You canât make that call honestly without knowing what the positions are, which is the whole reason I wrote the backward walk above.
Iâve made a version of this argument before. When I wrote about why I finally left X, the core of it was that participation is endorsement - that posting on a platform whose owner had gutted its safety infrastructure on purpose and personally platformed Tommy Robinson wasnât a neutral technical choice, no matter how good the app felt to use. I donât think that argument stops being true just because the person doing the platforming built a Linux distro instead of buying a social network. And I donât think itâs a coincidence that Tommy Robinsonâs name shows up in both stories. Some things really do rhyme.
I also wrote a while back about whether company culture can survive capitalism, about how the pull toward treating culture, or values, or people, as just another input to optimize isnât usually a moral failure so much as a structural one, once a business relationship enters the room. I think thatâs what I watched happen in my own feed this week: someone I respect professionally posted, genuinely excited, about working with DHH directly - the kind of clean, congratulatory business update that reads like nothing above it exists. Thatâs not a neutral business update to me. Itâs someone deciding, whether theyâd ever frame it this consciously or not, that a partnership is worth more than the rights of people like me.
Iâm transgender. The âBad Therapyâ post, the one where DHH cites âthe dramatic surge in transgender identification, and the medical interventions and experiments that have come with itâ as evidence of someone worth listening to, is the one I felt specifically. I feel a version of it again every time a post like the one I saw this week gets treated as good news, full stop, no asterisk.
What Anil Dash got right
I think Dashâs post is going to read as hyperbole to some people, and I want to be direct about why I donât think it is. Heâs not calling DHH a Nazi as an insult. Heâs naming a specific, documented rhetorical structure - population framing plus forced removal presented as logical necessity - and pointing out that it has a body count attached to it, one thatâs well within living memory for people whose grandparents survived it. âThis would have been a career-ending scandal a few years ago, and it still should beâ isnât an overreaction. Itâs a description of how much ground has shifted, and how quietly, when writing like this gets treated as just another hot take from a guy with strong opinions.
Dave, you should be sorry
DHH built the âIâm sorry, Daveâ post around the idea that heâs the wrongly-detained astronaut and Claude is the cold, paternalistic machine denying him something reasonable. I donât buy the casting. Heâs not the guy locked out of the airlock for no reason. Heâs the guy who wrote a paragraph applying wildlife-culling logic to a named ethnic group, got annoyed that a tool wouldnât help him make that argument more legible to a new audience, and then built a whole case about ideological tyranny and AI monopolies to avoid ever mentioning what the paragraph actually said. Heâs also the guy whoâs been writing some version of this for more than two years, in public, under his own name.
I donât know exactly where the line should sit for how much discretion any single AI company should have over the worldâs speech. Thatâs a real question and Iâm not going to pretend Iâve resolved it. What I do know is that âa population is out of control, and removal is the basic logic of self-preservationâ isnât a fresh provocation. Itâs the oldest trick in the book, and it has a documented ending. Enjoy Omarchy if you want to. Just donât let a nice tiling window manager be the reason you didnât know the rest.