How Much of the Job Is Real?

In 1930, John Maynard Keynes looked ahead at a century of automation and made a prediction. His grandchildren, he wrote, would work about fifteen hours a week. The machines would do the rest, and the hard problem of the future would be what to do with all the free time. He was half right. The productivity arrived, output climbed for decades, and the fifteen-hour week never came. The hours that automation was supposed to give back went somewhere else. The anthropologist David Graeber spent the last years of his life arguing that he knew where they went: into jobs that did not need to exist.

Graeber called them bullshit jobs, and he meant the term with some precision. A bullshit job, in his telling, is paid work so pointless, unnecessary, or harmful that even the person doing it cannot justify why it exists, while feeling obliged to pretend otherwise. He was careful to separate these from hard, badly paid, useful work, the cleaners and care workers and drivers without whom everything stops. That work is often miserable, but it is real. The bullshit job is stranger. It tends to pay well, carry a respectable title, and produce work whose existence is easier to defend than its value.

He sorted them into five kinds. Flunkies exist to make someone above them look important. Goons do work that is aggressive or deceptive and exists only because the other side employs them too, the lobbyists and the cold-callers. Duct tapers spend their days patching problems that ought to have been fixed at the source. Box tickers produce the appearance that something useful is happening, the report filed so an organization can say the thing was done. Taskmasters invent work for people who did not need it, or supervise people who needed no supervising. The puzzle Graeber kept circling is why a market system, supposedly allergic to waste, breeds so much of it. His answer was that the waste is political. He called it managerial feudalism, a world where importance is measured in underlings and a manager with a bigger team outranks one with a smaller team, whatever either team produces.

It is a seductive idea, and it deserves a little care, because the evidence runs both ways. The viral 2013 essay that started it leaned partly on a poll in which 37 percent of British workers said their job made no meaningful contribution. The same poll found 63 percent said their work was personally fulfilling, a reminder that private satisfaction and social usefulness are different questions. A 2022 study of European workers rejected the thesis, finding that the share who feel their work is rarely or never useful is small and falling, and that the high-status occupations Graeber named are not where that feeling concentrates; it turns up more often among manual and essential workers. A 2023 study using American data reached a different conclusion, finding support for Graeber after all, since workers in the occupations he named are likelier to call their own jobs socially useless, though it also found that the sense of uselessness has several sources and that his theory cannot carry the whole explanation. It suggested the pattern bites hardest in heavily financialized economies like Britain and the United States. So take bullshit jobs as a sharp lens rather than a settled law. It names something real about how far a job's pay and prestige can drift from its usefulness, even where it overstates how common the worst cases are.

That brings us back to the machines, and to a question worth sitting with. The pitch for automating white-collar work with AI rests on a premise it rarely states out loud, that a great deal of what people do at desks is routine enough to hand to a machine. Listen to how the technology gets sold. It will take the busywork and the drudgery, the rote tasks nobody wanted anyway. That premise is a thin, commercial cousin of Graeber's, the part about busywork without the politics about why it survives. The labs themselves put it more carefully than their marketing does. When Anthropic's Dario Amodei says that automating ninety percent of a job frees people to do more of the valuable tenth, or when OpenAI's Sam Altman admits he expected white-collar jobs to disappear faster than they have, neither is calling the work pointless. Both are assuming it is routine enough to compress or hand off. That assumption is close enough to Graeber's observation to be useful to anyone selling the machine, even as it travels without his conclusions.

Here is the part the technologists leave out. Graeber's argument ran the other way. He did not think the market would clear away pointless work. He thought the system had evolved to create and protect it, for reasons only loosely tied to efficiency, and that it would not automate the bullshit away on its own. If he was right, AI will not end it. It will grow a new layer of it. When the automated tenth expands, the new work may not be better work. It may be model supervision, output certification, and cleanup, fresh box-ticking with a technical sheen. Automate a task and the managerial reflex hires three people to watch the automation. Graeber supplied the mechanism in his other book, The Utopia of Rules, where he argued that technologies sold as liberation tend to become new occasions for rules, paperwork, and the people paid to administer them. AI may be the most powerful paperwork engine ever built. And there is a last irony in it. Graeber was an anarchist who wanted people to work less, and proposed a basic income so they could. He has become a posthumous mascot for an automation project he would have despised.

Now bring it home, because cybersecurity is one of the clearest exhibits Graeber never got to use. Plenty of security work is real and hard and necessary. Someone has to take the call at two in the morning, build the detection that catches the intrusion, find the flaw before an attacker does. None of that is bullshit. Much of the field, though, is organized around producing the look of security rather than the thing itself. Compliance is the clearest case. Done well, it is a forcing function, the lever that finally gets patching, access control, asset inventory, and logging funded. At its worst, it is box-ticking at industrial scale: attestations that reassure an insurer without changing anything on the ground, risk registers no one reads, framework maps whose product is a record of diligence rather than a safer system. The analyst closing the same false positive for the hundredth time is duct-taping a tool nobody tuned. The program kept alive mainly to move liability around is box-ticking in legal armor, and the one copied because a competitor has it edges toward Graeber's goons only when its real job is to look fierce rather than to defend. Above all of it can sit a layer of managerial reassurance, the dashboards and status meetings and the warm feeling, upstairs, of being covered. The result, in too many organizations, is a machine whose most visible product is the appearance of security, the kind that holds up in an audit. The people inside it tend to know. Burnout in this field has more than one source. The long hours are real, and so are the understaffing and the steady pressure of the alert queue. But there is also the demoralization Graeber named, the slow corrosion of spending your day on work you suspect does not matter.

There is a second mechanism at work, the managerial habit of turning judgment into a counterfeit science, and Matthew Stewart named it. Stewart had the right scar tissue for the job, a philosopher who spent years inside management consulting. His argument in The Management Myth is that management theory, the kind sold by consultants and business schools, claims a scientific authority it never earned. The founding legends rested on cooked or flimsy evidence: Taylor's famous efficiency numbers leaned on guesses and unpublished data, while Mayo built a theory of human relations on experiments confounded by changed workers and altered incentives. Security management often wears the same costume. The maturity model that scores a program from one to five, the risk register that can discipline judgment or launder a hunch into a heat map, the posture dashboard that renders safety as a number trending up: treated as aids to judgment these are useful, and treated as measurements of safety they become security's Taylorism, the look of rigor laid over a judgment call. Jerry Muller explains why the costume survives. What is easy to count becomes what gets managed, and in security the count is often not the thing, while the most countable thing is rarely the thing that matters. Bruce Schneier gave the result its native name back in 2003, security theater, the visible performance of safety without the fact of it. He noted the cruel asymmetry too: real security is mostly invisible, while the theater is built to be seen. The visible thing, in most organizations, is also the easier thing to fund.

This is why the subject belongs with the exits. It explains them better than fatigue alone does, because people walk away from work that feels hollow, not only from work that is hard. It also tells you which tasks the machines come for first: the rote triage, the report generation, the evidence-gathering and control-mapping that never change a decision. A lot of entry and mid-level cyber jobs are full of exactly that work. Those jobs can still matter, and the people in them are not doomed. The sharper question to ask of any role is whether it teaches judgment: the sense of what matters, what is noise, what to fix first, and how you know. That judgment is the part most resistant to both the bullshit and the bots, the work a model cannot fake and an auditor cannot reduce to a checkbox. So run a test on the work in front of you. Ask what it changes. Does it shrink an attacker's opening, shorten the time to detect or contain, sharpen a real decision, make a system better able to take a hit, or teach you how incidents unfold? If the answer is yes, you have found the part worth keeping, and it is worth defending. If it keeps coming back no, the problem is more likely the place than the person. And if you manage the work, run the same test before you ask anyone else to spend a year of their life on it.

An organization that rewards the binder and starves the threat hunt, that dresses a judgment call as a maturity score and calls it rigor, will not reward you for seeing it clearly. Keynes thought a century of productivity would buy back our time. It mostly bought a great deal of work that did not need doing, and the freedom he promised was never going to be handed down from inside the apparatus that swallowed it. Whether AI takes your job is the smaller question. A fair number of jobs in this field exist only because something upstream was left broken, unaudited, untuned, or too inconvenient to fix. The larger question is whether the job deserves your working life. Keep asking it, and do not flinch from the answer. For some readers it points back in, to the real work, buried and worth the digging. For others it points at the door. Both are honest. The only unforgivable move is to keep mistaking the theater for the work.

Sources

David Graeber, Bullshit Jobs: A Theory (Simon & Schuster, 2018), and the original 2013 essay, "On the Phenomenon of Bullshit Jobs" (Strike!), for the definition, the five types, and managerial feudalism.

David Graeber, The Utopia of Rules (Melville House, 2015), for technology and bureaucracy.

J. M. Keynes, "Economic Possibilities for Our Grandchildren" (1930), for the fifteen-hour-week prediction.

Empirical debate: Soffia, Wood and Burchell, "Alienation Is Not 'Bullshit'" (Work, Employment and Society, 2022); Walo, "'Bullshit' After All?" (Work, Employment and Society, 2023); YouGov (2015).

Matthew Stewart, The Management Myth: Why the Experts Keep Getting It Wrong (W. W. Norton, 2009; in paperback, The Management Myth: Debunking Modern Business Philosophy), for management theory as pseudo-science and the Taylor and Mayo critiques. The argument first appeared in The Atlantic (June 2006).

Jerry Z. Muller, The Tyranny of Metrics (Princeton University Press, 2018), for metric fixation and the habit of measuring inputs rather than outcomes.

Bruce Schneier, Beyond Fear (Copernicus, 2003), for "security theater," and his essay "Beyond Security Theater" (2009) for the point that real security is largely invisible while the theater consumes resources.

AI-leaders' statements via Fortune (May 26, 2026), reporting Altman's "pretty wrong" reversal and Amodei's productivity framing.