The Bloom Filter: When Critiquing AI Means Following the Money, Not the Craft

Abner's essay opens with an unusual disclaimer, a stamp reading "a human wrote this essay. Proof of work," and that small gesture tells you almost everything about the anxiety running underneath the piece. When authorship itself has to be certified, the act of writing becomes a political statement, and the writer has already conceded that the ground beneath craft has shifted. What follows is less a manifesto against artificial intelligence than a confession about how it feels to argue against something while suspecting that your argument makes you insufferable.

The central conceit is a mirror. Harold Bloom, the Yale critic who spent his career defending a canon and savaging what he saw as decline, becomes Abner's uncomfortable reflection. Bloom famously dismissed Harry Potter as the product of a culture that had stopped reading seriously, and his contemporaries called him a snob, a reactionary, a relic. Abner watches the old interview footage and recognizes the posture in himself, and more uncomfortably, in the Lobsters community he frequents, a forum of software engineers who guard their standards with the same prickly conviction. The recognition is the essay's engine. He is not exempting himself from the charge of elitism. He is admitting he is guilty of it and then asking what to do with that guilt.

The argument hidden inside the self-deprecation

The thesis arrives quietly, almost as a tactical retreat. Abner concedes that monologues about software craft, about the dignity of the work, about whether code has gotten better or worse, land as snobbery. When he tells people that the state of GitHub proves the industry has decayed, he gets blank stares. The complaint sounds aesthetic, and aesthetic complaints from technical people read as gatekeeping. So he proposes a different rhetorical instrument entirely. Stop lamenting the craft. Point instead at the broken business case.

This is shrewder than it first appears. An argument about quality is unfalsifiable and self-flattering; it positions the speaker as the keeper of taste. An argument about return on investment is something a chief financial officer can verify or refute. By moving the debate from "is this work good" to "is this work paying for itself," Abner trades the high ground of the critic for the firmer ground of the accountant, and the accountant, in his telling, wins more converts.

The evidence he marshals is economic rather than artistic. He cites the broad repricing of AI services in 2026, the stalled and blocked data center projects, the investor nervousness, and a string of stories about organizations discovering that their spending had no floor. Uber's operating chief reportedly admitting that AI costs are harder to justify internally after blowing past budget. Copilot users startled, once subsidies thinned, to learn how much their token allotment had quietly been worth. An Axios report describing a company that burned through half a billion dollars in a single month because no one set usage limits. The detail he lingers on, the one he says made him blink, is a chief technology officer reporting that employees were spending model calls to check the weather. The absurdity is the point. When a tool is metered but feels free, people pour it onto trivial tasks, and the bill arrives later as a shock.

What the demand for proof actually asks

The most forceful passage in the essay is a plain request. Show me a company spending millions on AI tokens that can name a clear, indisputable return. Show me the processes automated and what automating them saves against the cost. This is not a rhetorical flourish dressed as a question; it is a genuine challenge, and its power comes from how rarely it is answered cleanly. The discourse around AI inevitability, Abner notes, never seems to terminate in a profit line. Vast claims about transformation coexist with an absence of the one number that would settle them.

There is a real economic logic underneath the frustration. The case for any productivity technology eventually has to close the loop between expenditure and output. Electrification, the spreadsheet, the relational database all eventually produced measurable gains that justified their cost, sometimes after a lag, but the gain materialized. The discomfort Abner names is that the current wave is being financed on faith in a future closing of that loop, while the present shows enormous compute bills and ambiguous results. He is not claiming the loop will never close. He is observing that the people loudest about its inevitability are also the least able to point to where it has closed already.

The phrase he coins for the failed strategy, "tokenmaxxing," captures a familiar institutional pattern: buy licenses widely, distribute access, and hope value emerges from saturation. The quoted verdict, that throwing AI licenses at the wall to see what sticks has not produced tangible returns, describes a procurement reflex more than a strategy. Organizations adopted the tools because not adopting them felt like falling behind, and the cost of that defensive posture is only now being tallied.

Implications and the limits of the move

Abner's rhetorical pivot has a cost of its own, and to his credit he half-acknowledges it. Lobsters is a computing forum where business rants are off topic, and he has to justify why an essay about runaway cloud spending belongs there at all. His answer is that when engineers are increasingly billed for the privilege of compute, the technical and the financial can no longer be cleanly separated. The cost of a model call is now a constraint on how software gets written, which means the business enterprise is no longer external to the craft. The economics have entered the editor.

The deeper question the essay raises but does not resolve is whether the financial critique is a more honest argument or merely a more persuasive one. Abner discovered that pointing at defective business cases recruits the neutral developers who roll their eyes at craft complaints. That is a fact about persuasion, not about truth. It is possible that the software really has stagnated and that the economic argument is simply the wrapper that gets the aesthetic grievance past people's defenses. If so, then the move from Bloom the critic to Abner the auditor is partly a disguise, a way to keep making a taste judgment while wearing the costume of fiscal prudence.

The counter-perspective worth holding

An honest reading should grant the other side. The absence of a clean return-on-investment number is not proof that no return exists; many genuinely useful technologies took years to show up in productivity statistics, and the measurement problem is real rather than evasive. The weather-checking anecdote proves that metering was sloppy, not that the underlying tool is worthless. And there is a version of Abner's own argument that turns back on him: the engineer who insists on indisputable proof before granting any value can sound exactly like Bloom insisting that nothing written after the canon could possibly matter. The demand for certainty can itself be a form of the elitism he is trying to escape.

What survives that scrutiny is the modest, defensible core. Costs that were hidden by subsidy are becoming visible, and visibility is forcing a reckoning that hype postponed. Whether or not the technology earns its keep, the era of pretending the meter does not exist is ending, and the conversations that follow will be conducted in the language of budgets. Abner's contribution is to notice that this language persuades where the language of craft only alienates, and to choose it deliberately. He cannot unsee the Bloom filter in himself or his community. His response is not to stop criticizing but to find a critique that does not require everyone else to share his taste, only to read the same invoice.