On 29 September 2025, the CEO of Vercel posted this:
Enjoyed my discussion with PM Netanyahu on how AI education and literacy will keep our free societies ahead.
We spoke about AI empowering everyone to build software and the importance of ensuring it serves quality and progress.
Optimistic for peace, safety, and greatness for Israel and its neighbors.
Thirty-three million views. It is not an obscure remark, and it was not misconstrued. It says what it says.
Read it twice. The problem isn’t the politics. The problem is the grammar.
Notice what has agency in those sentences. AI will keep societies ahead. AI empowers everyone. It serves progress. The author is a witness to something already in motion — a weather report, delivered by someone who arranged to be photographed standing in the weather. There is no actor who chooses, and therefore no actor who could be wrong.
That is not sloppy writing. It is the characteristic voice of a person who has decided in advance that whatever happens next is not their responsibility, and who has found a vocabulary that makes the decision invisible — including to themselves.
I want to take that seriously, because I think it names the central ethical problem of our profession, and because three thinkers diagnosed it long before anyone had a phrase like “AI literacy.”
I. Hume: you cannot get an ought from an is
Start with the oldest objection, and the one everything else rests on.
In A Treatise of Human Nature (1739), David Hume noticed that moral arguments perform a silent substitution. An author spends pages establishing what is the case — how things are, how they came to be, what follows from what — and then, without warning, the copula changes. Suddenly every proposition is joined by an ought.
Hume’s complaint wasn’t that the conclusion is wrong. It was that a step has been skipped and the author hasn’t noticed. You cannot derive a value from a fact. Some further premise is doing the work, and if it isn’t stated, it isn’t being examined.
Apply that to the post.
- Is: AI capability is advancing rapidly.
- Is: More people can produce software than before.
- Ought: Therefore this constitutes progress, progress ought to be accelerated, and being “ahead” is the thing to want.
The missing premise is enormous. It contains an entire theory of what human flourishing consists of, who counts as a beneficiary, what “ahead” means and ahead of whom, and — crucially — what may be traded away to get there. None of it is argued. It is smuggled in as though it were a fact about the world rather than a contested claim about value.
This is the foundational move, and every other trick in the genre depends on it. Once “progress” is established as a fact rather than a choice, the person making the choice disappears, and with them anyone you might hold to account.
Hume’s other contribution matters here too. He argued that reason alone never motivates action — that “reason is, and ought only to be the slave of the passions.” He did not mean be irrational. He meant that the engine of moral life is what you actually care about, and intellect is the instrument that serves it. A profession that trains people to be superb instruments while declining to ask what they are an instrument of produces exactly what Hume would predict: extraordinary means, unexamined ends.
II. Arendt: the problem was never that he was a monster
In 1961 Hannah Arendt went to Jerusalem to cover the trial of Adolf Eichmann for The New Yorker. She expected a monster. What she found unsettled her more.
Eichmann was a functionary. He was diligent. He was preoccupied with his career and mildly aggrieved about promotions he felt he’d been denied. He spoke in the clichés of his organization and appeared to hold no thought that had not been issued to him. In Eichmann in Jerusalem: A Report on the Banality of Evil (1963), Arendt reached the conclusion that made the book notorious: the man was not a fanatic or a sadist. He was thoughtless. He had simply never occupied the position of the people his work processed.
Her argument gets flattened into “evil is ordinary” or “anyone could do it,” which turns it into fatalism. The actual claim is more specific and far more demanding. She identified a faculty — thinking, meaning the internal dialogue in which you examine what you are doing — and observed that its absence is what makes industrial-scale atrocity administratively possible.
Not hatred. Hatred doesn’t scale. It burns out, it’s inefficient, it requires everyone in the chain to be personally invested.
Thoughtlessness scales beautifully. It requires only that each participant attend to their own segment, execute it well, and decline to ask what the segment is part of.
That is a description of a modern engineering organization. I don’t say that to provoke. I say it because the org chart, the sprint, the ticket, and the abstraction layer are specifically designed to let each person reason about their component without reasoning about the system. That is what makes them powerful. It is also what makes them dangerous — and the danger isn’t a defect in the method. It’s the method working exactly as intended, pointed somewhere nobody checked.
Why Arendt in particular
There’s a reason to reach for Arendt here rather than any other moral philosopher, and it isn’t rhetorical convenience.
Arendt was a German Jew who fled the Nazis, was interned at Gurs, escaped, and arrived in the United States stateless in 1941. She had standing.
She was also, from early on, a critic of the political direction of the state whose court she was reporting from. In “Zionism Reconsidered” (1945) she warned that establishing a Jewish state without resolving the standing of the Arab population would produce permanent conflict and morally deform the project itself. She argued for a binational federated arrangement and was largely dismissed for it.
Three years later, on 4 December 1948, The New York Times published a letter signed by Arendt, Albert Einstein, Sidney Hook and others protesting Menachem Begin’s American tour. They described his Herut party as “closely akin in its organization, methods, political philosophy and social appeal to the Nazi and Fascist parties,” and noted that the leadership of American Zionism had declined to expose the danger to its own constituents.
Hold those two facts together. The woman who gave us “the banality of evil” — writing about a trial held in Jerusalem — had already spent years warning that a state can be founded by survivors of atrocity and still commit grave wrongs, and that the people best positioned to say so are the ones most strongly discouraged from saying it.
This is why the reflexive dismissal fails. Applying Arendt’s framework to the conduct of the Israeli state is not a hostile appropriation of a Jewish thinker’s work against her own people. It is the use she made of it herself, at a cost that ended friendships which never recovered.
III. Durkheim: prosperity is a destabilizer
The third piece is the one people find most surprising, and the most underrated.
In Suicide (1897), Émile Durkheim identified a category he called anomic suicide — arising not from misery but from normlessness: the condition in which the rules calibrating a person’s expectations to their circumstances stop functioning, so desire has nothing to measure itself against.
Here is the part left out of the summary. Durkheim’s central claim is that anomie is produced by rapid change in either direction. He was explicit that sudden prosperity deregulates as thoroughly as sudden collapse. A boom shatters the scale by which people had measured a sufficient life, and appetite, having no natural ceiling, expands to fill whatever space it is given. It leaves people with an infinity of desires and no means of satisfying any of them — and that condition, not poverty, is what he found correlated with self-destruction.
Read that as a description of an industry.
We have built a culture in which the reference point resets every eighteen months, in which “what is sufficient” is definitionally unanswerable because the frontier moves before you finish the sentence, and in which the governing verb is keep up. We then observe, with apparent puzzlement, that the people inside it are exhausted, that meaning is scarce, and that a workforce selected for capability reports historically poor mental health.
Durkheim would not have been puzzled. Deregulated ascent is not the opposite of crisis. It is a form of crisis, and its symptom is precisely the inability to say what would be enough.
So when someone says “progress” as though naming an unambiguous good, the honest follow-up is Durkheim’s: progress toward what, measured against what, and at what point are we permitted to stop? A conception of progress that cannot answer those questions is not a goal. It is an appetite that has learned to describe itself in the language of virtue.
IV. Progress is an externality, not a goal
Here is the claim I want to defend.
Progress is not something you can aim at. It is a word applied retroactively to what happened when determined people refused to accept a broken system.
Look at anything you would genuinely call progress and you find that the people who produced it were not pursuing progress. They were pursuing a specific grievance. They wanted this factory to stop maiming workers. They wanted this law repealed, this water clean, this ward staffed, this person to stop being treated as property. “Progress” is the word historians attach to the aggregate afterwards. It was never in the room.
The distinction isn’t academic, because the two orientations behave differently under pressure.
If your goal is a specific injustice, you can tell whether you are winning. The grievance is either addressed or it isn’t. You have a stopping condition, a measurable, and a constituency who will tell you bluntly when you’ve failed them.
If your goal is “progress,” you can never fail — because whatever occurred is, by construction, what progress turned out to look like. Harm becomes a transitional cost. Displacement becomes disruption. The people who bore it become an unfortunate feature of the adjustment period. There is no state of the world that could falsify the claim that you were making things better, which is another way of saying the claim is empty.
This is exactly where Arendt’s thoughtlessness and Hume’s missing premise meet. “Progress” is a thought-terminating concept. It delivers the feeling of having justified something without the labour of having done so.
Not geniuses
A corollary that matters more than it looks.
The mythology of this industry is that progress issues from exceptional individuals — that history is a relay race between geniuses and the rest of us are spectators. This is false as history and corrosive as ethics.
False, because the record doesn’t support it. The eight-hour day. The weekend. Workplace safety law. The extension of the franchise. The dismantling of formal segregation. None came from a visionary. They came from enormous numbers of ordinary people, most of whose names are gone, who were stubborn for longer than the systems opposing them could sustain. The mechanism was not brilliance. It was courage, humility, and persistence, distributed across people of every background and every colour, most of whom never saw the outcome.
Corrosive, because the genius myth is load-bearing for abdication. If progress comes from exceptional people, then you — an ordinary engineer with a mortgage and a standup at 9:15 — have no role and no responsibility. You’re the audience. The myth is comfortable precisely because it excuses you.
The historically accurate version is far more demanding. It says the outcome is the sum of ordinary refusals, which means it depends on yours.
V. Your dependency graph has a politics
Here is where this stops being philosophy and starts being a Tuesday decision.
The man in that post runs the company behind Next.js. An enormous share of the web’s front end is built on it. That is not incidental to the argument — it is the argument, arriving at your terminal.
We tell ourselves that tools are neutral, and at the level of the compiler
that’s true enough. But a framework is not only a compiler. It’s a company, a
funding structure, a hiring pipeline, a conference stage, a set of defaults, and
a person who gets to speak with the accumulated authority of everyone who
depends on it. When you npm install, some fraction of your professional
legitimacy is lent upward. That fraction is small. It is not zero. And it
aggregates.
I want to be careful, because the weak version of this argument is genuinely weak, and I’d rather state the objections than have them stated back at me:
- Maintainers are not executives. The people writing the code are mostly not the people posing for photographs, and punishing the former for the latter is neither fair nor effective.
- Switching costs are real. “Rewrite your stack” is cheap advice from someone not paying for it, and for most teams it is simply not available.
- Purity is unreachable. Every layer of this industry has a compromised actor somewhere in it. A standard that requires clean hands requires you to stop computing.
All true. And none of it establishes that the correct fraction of consideration is zero, which is what “tools are neutral” actually asserts.
What I’d say instead: you get to choose your defaults, and defaults compound. Not a purge — a preference. When two options are close, let this be the tiebreaker. When you’re starting something new, where the switching cost is nearly nothing, choose deliberately. When someone asks why, tell them, because the telling is most of the mechanism.
I moved this site off Next.js. I’m not going to pretend that inconvenienced anybody but me, or that a static site generator is a political act. It cost me a weekend and it bought me nothing except that I no longer have to think about it. That is a small thing. Small things are what aggregates are made of.
And if it ever costs more than a weekend — if it means a language, a stack, a job, a city — that’s a price I’ve decided in advance I’m willing to pay, so that I don’t have to relitigate it in the moment when it’s hardest to think clearly. Deciding in advance is the whole trick. Arendt’s Eichmann never decided anything; he just kept finding that the next step followed from the last one.
VI. What this actually requires of an engineer
Moral exhortation that terminates in a feeling is worth nothing. Concretely:
1. Reinstate the actor in your own sentences. Every time you write that a technology “will” do something, rewrite it with a subject. Not “AI will democratize software” but “we are building X, which we expect to have effect Y on population Z, and here is who bears the cost.” Most of the time you will find you cannot finish the sentence. That’s the point — it’s the missing premise becoming visible.
2. Know what your abstraction is abstracting. Not needing to know what’s underneath is a genuine professional virtue. It’s also the precise mechanism Arendt described. You don’t need to know everything. You do need to know, at least once, what the whole system does — because “I only worked on the scheduler” has a history.
3. Recognize that engineering leverage is unusually high and unusually early. This is the practical asymmetry. The person who says “that’s not technically feasible on that timeline” is believed in a way that the person saying “that’s wrong” is not. Whether that’s admirable is beside the point; it’s a real lever, disproportionately held by people who insist they have no power. Retention defaults, what gets logged, whether a record can identify an individual, whether a deletion is real — these are settled in implementation, by someone with a ticket, usually before any policy conversation happens. That person is frequently you, and the decision is frequently reversible only at enormous cost.
4. Refuse in ways that leave a record. Quiet non-participation is individually clean and collectively useless. A refusal that is written down — in a design doc, a review comment, a resignation letter that says why — changes what the next person believes is possible. Most of what makes a norm is people discovering they weren’t the only one.
5. Accept that you might be wrong, and act anyway. This is the humility half and it isn’t decoration. Arendt’s demand is that you think, not that you reach a particular conclusion. Someone who adopts a position because their timeline adopted it hasn’t thought; they’ve just changed which authority they defer to. Thinking means holding open the possibility that the thing you’re building is good, examining it anyway, and being willing to discover that you were the one who was wrong.
VII. On precision, which is a weapon and not a weakness
If you’re going to say hard things about powerful people, say them exactly.
The International Criminal Court issued arrest warrants for Benjamin Netanyahu and former defence minister Yoav Gallant on 21 November 2024. The Pre-Trial Chamber found reasonable grounds to believe they bear criminal responsibility for the war crimes of starvation as a method of warfare and intentionally directing an attack against a civilian population, and the crimes against humanity of murder, persecution, and other inhumane acts, over a period running from at least 8 October 2023 to at least 20 May 2024. The warrants remain outstanding and are enforceable in principle by all 125 states party to the Rome Statute.
That is the accurate formulation and every word of it is defensible.
He has not been tried. He has not been convicted. Writing “convicted war criminal” would be false, and would hand every reader looking for an exit the exact exit they wanted. Writing “subject to an outstanding ICC warrant for alleged war crimes and crimes against humanity” is unarguable — and it is not one degree less damning, because the thing being described is a sitting head of government formally accused by the world’s permanent criminal tribunal of starving a civilian population.
The same discipline applies to the man who posted the photograph. He is not accused of anything by anyone with jurisdiction. He is a chief executive who chose to publicise a friendly meeting, in a tone of untroubled optimism, with a person under that warrant — and then described the subject of the conversation as progress. That is the whole of the charge, and it is enough. It requires no embellishment, and embellishing it would only give the argument away.
Precision is not squeamishness. It’s what lets a claim survive contact with people who need it not to. The strongest version of an accusation is always the most exact one, and the people most eager for you to overstate are the ones planning to litigate the overstatement instead of the substance.
VIII. The objections, taken seriously
A piece like this is worth nothing if it only speaks to people who already agree.
“You’re demanding engineers become political.” They already are. Every default is a policy. Every retention window is a judgement about whose past can be used against them. Every ranking function is a distribution of attention. The choice isn’t between political and neutral engineering — it’s between politics you’ve examined and politics you’ve inherited. “Neutral” means “aligned with whoever set the defaults.”
“Individual refusal accomplishes nothing. Someone else will build it.” Sometimes true, and the argument proves too much: it’s an argument against ever being first to refuse anything, and it dissolves every norm that has ever formed. It also misdescribes how professional norms actually arise. They form when enough individuals make a thing costly and embarrassing to ask for. Nobody is the decisive refusal. Everybody is a term in the sum.
“AI literacy is genuinely good. You’re attacking a benign statement.” The statement isn’t the target. The structure is. Read it again: no subject who chooses, no population named, no cost acknowledged, and no condition under which the author would conclude he’d got it wrong. A statement engineered so that nothing could count against it isn’t a claim — it’s a mood, delivered in the vicinity of a man under indictment, carrying the implication of endorsement and the deniability of having said nothing.
“This is selective outrage. Why this and not everything else?” Selectivity is a condition of finite attention, not a character flaw, and this objection is almost always deployed to produce paralysis rather than consistency. The honest answer is: yes, and you should apply the same standard everywhere you have leverage. The person asking usually hasn’t.
“You’re asking people to risk their livelihoods.” Sometimes, yes, and it should be said plainly rather than dressed up. Not everyone can afford to refuse — visa status, dependents, health insurance and immigration all bind differently, and a movement that only admits people who can afford martyrdom isn’t a movement. But the corollary is load-bearing: those with the most slack owe the most refusal. If you’re senior, well paid and employable, the cost to you is a fraction of the cost to a junior engineer on a work visa. Obligation scales with capacity, and pretending otherwise is how people who could act cheaply hide behind people who can’t.
IX. Where this leaves me
The personal part, without hedging.
I don’t believe technical excellence absolves anyone of the question of what they’re excellent for. I don’t believe “progress” names a goal — I’ve argued it names an externality, the residue left by people who fixed a specific broken thing and let history do the summarizing. I don’t believe the people who produced the progress worth having were geniuses. They were ordinary, stubborn, frightened people who kept going, from every background and every colour, and most of them are unrecorded.
And I will not treat the conduct of a head of state under an outstanding ICC warrant for starvation as a method of warfare as neutral background for a conversation about developer tooling. Not because I’m certain of every fact in a contested and horrifying conflict, but because the posture — the studied incuriosity, the vocabulary built so that agency vanishes, the optimism issued from a position of total insulation — is the exact thing Arendt spent her life describing.
Stacks are interchangeable. Languages are interchangeable. Employers, cities and countries are, with difficulty and privilege, interchangeable. What isn’t interchangeable is the answer you give when someone finally asks what you were working on, and whether you thought about it.
Arendt’s demand was never that you be heroic. It was that you not be thoughtless. The bar is low. It’s available to everyone. And clearing it is still, somehow, unusual.
Sources
- David Hume, A Treatise of Human Nature (1739), Book III, Part I, Section I — the is–ought gap.
- Hannah Arendt, Eichmann in Jerusalem: A Report on the Banality of Evil (1963).
- Hannah Arendt, “Zionism Reconsidered,” Menorah Journal (1945).
- Einstein, Arendt, Hook et al., letter to The New York Times, 4 December 1948.
- Émile Durkheim, Le Suicide (1897) — anomie, including the anomie of prosperity.
- International Criminal Court — defendant page, Benjamin Netanyahu.
- UN record of the 21 November 2024 Pre-Trial Chamber decision.
- Just Security — mapping state reactions to the warrants.
