Rumpelstiltskin's Receipt: Why Naming the Demon Won't Save Your Job (But Might Save Your Sanity)

The demon has a name now. Rumpelstiltskin. Or rather, Rumpelstiltskin-Brandolini, hyphenated like some cursed matrimonial alliance between Germanic folklore and Italian software engineering, which is precisely what it is. Alberto Brandolini—programmer, prophet, poor bastard—had been reading Daniel Kahneman before watching Silvio Berlusconi and journalist Marco Travaglio perform their political necromancy on Italian television, and on January 11, 2013, announced the principle that now bears his name: the amount of energy required to refute bullshit is an order of magnitude greater than the energy required to produce it. The tenfold figure was an epigram, not a laboratory measurement, and Brandolini’s law is a law only in the cheerful internet sense by which a recurring catastrophe becomes a feature, but the asymmetry is real enough to anyone who has attempted to clean the Augean stable of a comment section. I can manufacture a conspiracy theory in the time it takes to void my morning bladder—no research, no peer review, just the spasmodic ejaculation of something emotionally satisfying and factually constipated—while you, poor honest drudge, must spend three days excavating primary sources, cross-referencing datasets, constructing syllogisms sturdy enough to survive contact with Facebook, only to be met with “lol, cope” and a frog emoji. This is the bullshit asymmetry principle, and it governs our digital ecology with the iron consistency of a Calcutta municipal regulation: universally acknowledged, universally violated, universally generating more of the very filth it describes.

The Rumpelstiltskin effect is a new clinical name for much older magic. In 2025, the philosopher of religion Alan Levinovitz and the psychiatrist Awais Aftab proposed it for the relief that may occur when suffering receives a clinical diagnosis, even before any treatment has begun. The evidence is suggestive rather than conclusive; diagnosis may organize confusion, validate distress, and return a little agency to the afflicted, but it may also stigmatize, overmedicalize, or harden into an identity with prison walls and a billing code. The name comes, naturally, from the Grimm tale. A miller boasts that his daughter can spin straw into gold, a king locks her up and demands the impossible, and a mysterious little man performs the miracle for progressively extortionate payment, culminating in her promise of her firstborn child. Years later, after she has become queen and produced the contractual infant, he offers her three days to discover his secret name. A messenger overhears it in the forest. She says “Rumpelstiltskin,” and the bargain collapses. The spinning wheel does not fall silent at the naming—the spinning had finished long before—and the imp is not defeated in mortal combat. She has acquired the one piece of information that lets her escape the contract. Naming provides leverage. It is not mastery.

This distinction matters because we have named our current demon “Artificial General Intelligence,” labeled its larger and more theatrical cousin “ASI,” marked the calendar for “the singularity,” and convinced ourselves, through the alchemy of taxonomy, that we have thereby domesticated the future. We have not. We have given the anxiety a pet name, and now it sleeps at the foot of the bed, shedding algorithmic fur into our existential blankets. Sometimes the name brings relief because shapeless dread has become an object one can discuss. Sometimes it brings terror because a fog has acquired teeth. The proposed Rumpelstiltskin effect describes the relief of legibility; it does not promise that the thing made legible will be friendly.

The bullshit, meanwhile, is not merely abundant; it is industrious, prolific, a fecund factory of feculence operating on near-zero overhead. To claim that a doom deadline awaits us—that Tuesday next, or Q3 2027, or whatever calendrical sacrifice the venture capitalists are offering to their spreadsheet deities this fiscal quarter, every white-collar profession will vanish in a puff of neural-network smoke—requires no evidence. It requires only a headline, a pulse of cortisol, and a species with a long habit of booking the apocalypse and forgetting to cancel the caterer. It is easy, this manufacturing of doom. It is cheap. It spreads like dysentery in a monsoon drain, which is to say, efficiently and with complete disregard for social hierarchy.

The less operatic evidence is also less comforting. In 2025, the International Labour Organization examined nearly thirty thousand occupational tasks and estimated that one worker in four worldwide was employed in an occupation with some degree of generative-AI exposure. Yet it concluded that most exposed jobs were more likely to be transformed than erased; the highest-exposure category accounted for about 3.3 percent of global employment. Exposure is not adoption, adoption is not automation, and automation is not necessarily unemployment, although each may place a boot on wages, autonomy, hiring, or the number of people expected to do the work. To explain that white-collar labor is being decomposed task by task, through systems that still require integration, supervision, liability arrangements, regulatory compliance, clean data, and the endless soul-corroding meetings that no large language model has yet been trained to endure, requires the labor of Sisyphus, except the boulder is made of working papers and the hill is made of Twitter.

I spent about fifteen years in American healthcare and research, that great cathedral of credentialist anxiety, and watched the prophecies arrive like seasonal fevers. In 2016, Geoffrey Hinton said people should stop training radiologists because within five years deep learning would outperform them; in the next breath he allowed that it might take ten. The five-year trumpet blast expired, the ten-year hedge is now upon us, and radiology is not a mass grave. Machine-learning systems became extremely capable at particular imaging tasks, and radiology accumulated more AI tools than any other medical specialty, but the profession did not consist of one benchmark waiting to be conquered. It consisted of many tasks, embedded in clinical workflows, legal accountability, consultation, intervention, quality control, communication, and the inconvenient existence of patients whose bodies refuse to resemble curated validation sets.

This is not because the technology failed. It often succeeded, sometimes spectacularly, within the boundary of the task it had been given. The error was smuggling an entire occupation into the task and calling the suitcase science. Benchmark sufficiency is not organizational substitution, and the gap between “can perform this task” and “has replaced this worker” is not a gap at all but a vast, marshy, bureaucratic wetland where projects go to decompose. Most often, at first, the AI does not take your job. It takes a task. Your manager then redraws the workflow, raises the quota, freezes a vacancy, calls the resulting abrasion “augmentation,” and, if any savings survive the consultants, presents them to the shareholders wearing a tiny bow. You are left holding a revised job description and the same lukewarm coffee, wondering where the revolution went. It went to procurement. It is stuck in procurement. It will remain in procurement for eighteen months and then return because someone forgot the data-processing agreement.

The present spiral looks deflationary rather than uniformly apocalyptic: a slow leak in the tire of white-collar bargaining power, prestige, headcount, and price, not one clean guillotine descending upon every neck at noon. The lawyer does not necessarily vanish; the lawyer discovers that document review and first-draft discovery work now require fewer junior hours, while judgment, negotiation, courtroom performance, and the delicate human ceremony of preventing a panicked client from committing fresh felonies remain inconveniently embodied. The accountant does not ascend to the great spreadsheet in the sky; the accountant learns that reconciliation and routine classification have accelerated, that the partners have noticed, and that the partners are not dividing the surplus according to the principles of Christian charity. This is the first spiral, the white-collar deflation, and it feels less like a tsunami than realizing, mid-shower, that the water pressure has been declining for six months and you have gradually adjusted your definition of bathing.

Then, perhaps, comes a second spiral: the blue-collar reckoning, delayed not by mercy but by the stubbornness of matter. Moravec’s paradox was an observation from the robotics and AI of the 1980s, not a commandment engraved in silicon: activities humans experience as intellectually difficult, such as formal reasoning or playing chess, proved easier to automate than perception and sensorimotor abilities acquired by a small child. Modern AI has chewed pieces from both sides of that old boundary, but a machine that writes a sonnet still does not thereby possess wrists, balance, tactile judgment, a plumber’s relationship with corroded iron, or the nurse’s ability to reposition a frightened obese patient without breaking the patient’s bones or her own back. The embodied system needs perception, dexterity, proprioception, fault tolerance, physical safety, cheap hardware, field maintenance, and the kind of tacit knowledge lodged in muscles and tendons that we do not notice until a robot drops the crockery.

If embodied AI becomes capable, cheap, reliable, and insurable across messy human environments, blue-collar work will face its own deflation. That is plausible; it is not calendrically ordained. “Eventually” is doing more heavy lifting here than a Bengali porter at Howrah Station. Nor will the transition respect our neat division between collars. Warehouses, factories, hospitals, offices, roads, and homes contain different combinations of structured space, physical risk, regulation, wages, and tolerance for error. The robot will not awaken one morning and replace Labor as an abstract philosophical category. It will enter through profitable niches, one carefully engineered environment at a time, trailing technicians, lawsuits, maintenance contracts, and men with clipboards.

The AGI discourse, the ASI terror, the eschatological countdown clocks maintained by rationalists who have confused Bayes’ theorem with the Book of Revelation—all of this is an attempt to name the demon precisely, to pin the butterfly of machine cognition to the specimen board of human taxonomy and say: here, here is the line, here is where it becomes real. But AGI and ASI are not standardized units like kilograms or kelvins. They are shifting families of definitions carrying scientific, philosophical, commercial, and theatrical luggage. A system need not satisfy a philosopher’s definition of general intelligence to destroy the price of a task. A company does not require metaphysical proof of machine consciousness; it requires a tool sufficiently useful, cheap, reliable, and defensible to survive deployment. Commercial substitution needs “good enough,” not a soul.

This is why the naming can become marketing, reinforcement learning for the human masses, a scaffold erected to support the weight of capital’s appetite. Each new model is described as nearly AGI, each failure is reassigned to the next model, and each definition retreats just far enough to keep the fundraising round alive. “Good enough” is not one fixed horizon, however. It changes with wages, regulation, error tolerance, insurance, workflow design, and the desperation of the buyer. The miller’s daughter won because she learned one stable secret. Our Rumpelstiltskin changes his name whenever the term sheet requires it.

And so I sit in Calcutta, that great churning compost heap of civilization where the bullshit asymmetry principle operates with particular venom, where a rumor can circumnavigate the city before the truth has found its spectacles, where the man announcing that the bridge will collapse tomorrow receives more attention than the engineer who spent three years calculating why it probably will not. I have bipolar II—major depressions and hypomanias, not the full manic episodes that define bipolar I—which means I know something about the architecture of doom, the manufacturing of conviction from insufficient evidence, and the seductive poetry of the spiral. A dysphoric hypomanic surge can produce an entire cosmology at almost no evidentiary cost: every interruption becomes a sign, every ambition a destiny, every fear a forecast. Testing that narrative against sleep, time, treatment, consequence, and other minds is slower and humiliatingly expensive.

Diagnosis itself can bring a little Rumpelstiltskin relief. The chaos acquires a clinical grammar. It is no longer an unnamed moral failure breeding in the skull. But the name does not identify one simple cause, predict every episode, or cure the illness by being spoken. It can become a map; it can also become a fence. I have learned to distrust the deadlines my own mind imposes. I have learned that “sufficient” is not a destination but a negotiation, and that the demon does not disappear when named but becomes somewhat more legible, perhaps manageable, occasionally billable, and almost certainly surrounded by paperwork.

The truth—which costs more to produce than the lie, always, eternally, by Brandolini’s miserable arithmetic—is that the evidence presently describes the future of work less as one cliff than as an uneven landscape of slopes, potholes, local precipices, temporary plateaus, and men charging admission to maps they drew in crayon. That does not make it safe. A gradual transition in the aggregate can still arrive as catastrophe at the level of one worker, one firm, one profession, one city. Tasks will be automated; some occupations will contract; new work will appear; much existing work will become cheaper, faster, more closely monitored, or less autonomous. White-collar deflation may be followed by deeper physical automation, but neither the order nor the speed is guaranteed. Through it all we will continue naming our anxieties, christening our Rumpelstiltskins, and mistaking the production of vocabulary for the production of control.

The Śatapatha Brāhmaṇa offers no neat aphorism about the gods becoming gods by naming themselves; that line is too convenient, and as far as I can determine, it is not there. What the text actually gives us in 6.1.3 is stranger and therefore more useful. A nameless newborn boy cries and tells Prajāpati that he has not been freed from evil because no name has been given to him. Prajāpati names him Rudra. The boy replies that he is greater than that and demands another name. He receives Sarva, Paśupati, Ugra, Aśani, Bhava, Mahādeva, and Īśāna, each name joining him to another form of the world. Naming does not reduce the being to one manageable thing. It reveals that the thing exceeds every name and keeps returning for another.

That is closer to our predicament. AGI, ASI, singularity, automation, augmentation: each name makes one form visible and leaves another moving in the dark. We name the demon to read the receipt, not to cancel the debt. The deadline will pass and another will come. The work will change, some of it will shrink, some will metastasize into fresh drudgery, and we will remain—stubborn, biological, frightened, ingenious, insufficient—spinning our straw into something that is not gold but is, perhaps, enough.

P.S. Sources and provocations: Alberto Brandolini’s formulation of the bullshit asymmetry principle, January 11, 2013; Alan Levinovitz and Awais Aftab, “The Rumpelstiltskin Effect: Therapeutic Repercussions of Clinical Diagnosis,” BJPsych Bulletin (2025), doi:10.1192/bjb.2025.10137; Geoffrey Hinton, remarks at the Machine Learning and the Market for Intelligence conference, Toronto (2016); Hans Moravec, Mind Children (1988); Paweł Gmyrek et al., Generative AI and Jobs: A Refined Global Index of Occupational Exposure, ILO Working Paper 140 (2025); Jacob and Wilhelm Grimm, Kinder- und Hausmärchen (1812); Śatapatha Brāhmaṇa 6.1.3, trans. Julius Eggeling, Sacred Books of the East 41 (1894). For Brandolini’s law in action, see any comment section where a man with six followers announces that he has disproved epidemiology before breakfast.