Knowledge work in the age of mechanical reproduction

Today, for a change, I’d like to stay away from the immediate news.

Partly because the news has become a cacophony of press releases at the expense of any analysis of real issues.

For example, U.S. commandos were ready to board a Chinese cargo ship when someone realized that the claim the ship was carrying components for the nuclear industry had been confabulated by a language model. The incident nearly triggered a military confrontation between the U.S. and China, but a week later, no one’s talking about it anymore. Because, hey, what’s the big deal?

Instead, today The Guardian runs the headline: “OpenAI says its review into hacks, including on Australian government sites, is costing $500,000 a day.”

To that “news” (can you hear the quotation marks?), my reaction is:

SO FUCKING WHAT?

I mean, good for you for conducting your own investigations, but one of your representatives should be doing them while you guys stay behind behind fucking bars. Becuse last time I checked, unauthorized access to a computer system, with or without damage, with or without data exfiltration, carries a prison sentence.

Because for me, the real news is: what are Altman, Amodei, and Pichai (all of whom have admitted that their systems "autonomously" went hacking) doing not only out of jail, but without even a criminal proceeding in sight?

Or again, what is Zuckerberg doing out of jail, when META is objectively working to steer the elections in Brazil in the direction preferred by the U.S. government?

What are all these assorted techbros doing out of jail, when Milei’s government in Argentina is bending over backward to allow individuals, companies, and foreign governments (who knows who they have in mind) to literally buy up entire swaths of the country for example, in Patagonia) for their own pleasure and that of their damn data centers?

But of course, no one talks about this, while the imaginary “AI rebellion”—which exists only in the late-adolescent, ketamine-fueled minds of the tech oligarchs—is served up for breakfast, lunch, and dinner, weekends included.

So, if you’ll allow me, let’s talk about something else.

In that little gem that is the Simpsons movie, our hero has a psychedelic experience that makes him “have a Christmas.” Obviously, what he had was an epiphany, but Homer Simpson is our portrait in these desperate yet frivolous times, and so he has a Christmas.

And I had a Christmas too, after years of explaining that no, it’s not intelligence, that there’s no one on the other side of the screen, and that those pushing this idea have troubling cultural and political affinities with people we hoped history had buried with the fall of the Third Reich... after all that, I realized that this so-called “artificial intelligence” is, after all, good for something.

This isn’t my Christmas.

This is: Artificial Intelligence is good for showing what people manage to pass off as “work” these days. And it’s disheartening.

An explanation is in order. Anyone who hasn’t been living under a rock in recent years must know (or at least have heard) that so-called “artificial intelligence” produces nothing but regurgitations of the material it was trained on.

Why this technology immediately piqued the interest of managers and entrepreneurs was obvious to me from the start: like every other form of automation before it, it sells the dream of being able to get work done without workers, which is the true capitalist ambition.

Instead, until recently, I was racking my brain trying to figure out why certain professions, seemingly above suspicion, appear not only open to the idea but even enthusiastic about introducing language models into their normal work processes.

Programmers, for example, or lawyers. Two professions I would have bet were impervious to replacing expertise and professionalism with language generators unconstrained by reality, that is, as I’ve been saying for four years, bullshit generators.

And yet.

And yet I’ve had to change my mind because, even though we can’t speak of uncritical, let alone universal, adoption, I haven’t seen the kind of staunch opposition I would have expected. Maybe lawyers and programmers have suddenly gone mad and think they can benefit from an ideology that only hopes to replace them?

No. Things, as always, are more complicated. I know almost no one in either profession who truly believes they can be replaced by technology.

But, equally, I know almost no one who rejects the use of so-called “Artificial Intelligence.” How can this be explained?

For now, let’s set aside software and focus on more traditional office work. Everyone I talk to tells me the same thing: language models give you a rough draft; you tweak it and check for obvious errors; and while you might not like the result, your boss does. And so does his.

This tells us something interesting: it tells us that we’re evaluated not for the work we do, but for the illusion of work we’re able to put on display.

If a text extruder can produce passable results, it means that when we draft business letters, contracts, memos, and company policies and procedures, we’re only a few edits away from a simple copy-and-paste job.

All of this has a name, coined by anthropologist David Graeber in 2018: Bullshit Jobs. According to Graeber, these are jobs that are all for show and have no real impact, mere shuffling of paperwork, busywork. Activities devoid of any real meaning that serve only to make us pretend we’re busy, while our souls are being eroded.

I fully agree with Graeber’s analysis, but I arrive at slightly less despairing conclusions.

What does it tell us about our work if a more or less significant portion of it can be produced by a bullshit generator?

If, in a letter, a report, a contract, or a policy generated using a language template, I only have to check that the names are correct and that the clauses and references to legal articles match, what does that tell us about that letter, that contract, or that policy?

To me, it suggests that all that was really needed was a document template, a form, where I could simply fill in the blanks.

And it also suggests that this form requires only those workers with the operational and process expertise to know what needs to be written in the variable parts of the template. The rest of the chain of command, however, is completely superfluous.

This, in and of itself, obviously means nothing. But I find it interesting that those who are pushing the hardest to force the adoption of so-called “Artificial Intelligence”, hoping for the resulting technological replacement of operational levels, fail to realize that they are, in fact, demonstrating their own utter uselessness.

Fate, as we know, does not lack a sense of irony.

I couldn’t care less about the fate of managers. Instead, what is the meaning of intellectual labor in the age of its mechanical reproduction, to paraphrase Walter Benjamin? We could discuss this and arrive at some interesting conclusions.

For example, the joke about the mechanic comes to mind. A shiny new Ferrari pulls into the shop; the usual boorish guy who gets out of that kind of car whistles the mechanic over and says, “Look what’s wrong, engine’s making a strange noise.” The mechanic opens the hood, takes a look, revs the engine a couple of times and sure enough, the engine does make a strange noise. So the mechanic grabs a hammer and gives the engine a whack. Then he revs it again, and the engine runs like a charm. “There you go,” he says. “That’ll be a thousand euros.” The driver almost has a heart attack: “What? A thousand euros for one whack?” And the mechanic, unfazed: “No, the whack's free. The thousand euros is for knowing where to hit.”

There you go. In the world of intellectual work, we’re unfortunately used to requests accompanied by the inevitable, “How long will it take you, ten minutes?”

In a world that retains at least a semblance of common sense, knowing which two lines to add or change in a standard document should be worth just as much as writing the entire document from scratch, while “approving” the document should be nothing more than a useless bureaucratic formality.

I’m not sure I live in that kind of world. While I’m certain that management will draw the wrong conclusions from so-called “Artificial Intelligence.” The tech oligarchs are moving heaven and earth to go public or at leastt secure military contracts.

No wonder they’re in a hurry. Anthropic reported $4.6 billion in revenue, against losses of $42 billion. OpenAI reported revenue of $13.07 billion in 2025, but its final result was a loss of $38.53 billion. On top of that, both companies have non-cancellable agreements worth hundreds of billions of dollars each for future expenses related to the use of data centers that haven’t even been built yet.

If we combine those “geniuses” of Altman and Amodei with those at Amazon, Google, Meta, Microsoft, and Oracle, the so-called “industry” of so-called “Artificial Intelligence” has generated $204 billion in revenue from 2022 to the present, with expenses totaling $1.5 trillion.

So there’s a good chance that the fairy tale of so-called “Artificial Intelligence” will collapse in 2027. There will be a “realignment,” as they say on Wall Street; in other words, it will be a disaster. Afterward, the costs for Claude and company will likely be ten or fifteen times what they are today, if they are still around at all.

Then perhaps we’ll be able to take stock. Because while so-called “Artificial Intelligence” may shrink dramatically, the lesson it has taught us about what work really is, well, that remains.

Things might even get interesting.