Two postures toward machine intelligence are licensed in the trade press, and each is held with a conviction wildly out of proportion to what either can know: the system will take the drudgery and return the worker to the dignity of judgement; the system will take the work and return nothing at all. What deserves attention is not that both are argued but that they are argued about the same weights, the same vendor, the same procurement cycle, and frequently by the same executive within a single quarter. Melanie Klein, working with very young children in London through the 1930s and 1940s, described the mental arrangement in which this is not a contradiction but a method. In Notes on Some Schizoid Mechanisms (1946) she set out the earliest organisation of the psyche as one that cannot hold its object whole: unable to bear that the source of nourishment is also the source of frustration, the infant divides it into an idealised object and a persecuting one, and rids itself of its own destructiveness by projecting that destructiveness into the second, which then quite reasonably appears to be coming for it. The persecutor is one's own aggression, returned with a face.

Klein chose the word position rather than stage, and the choice carries the argument. A stage is passed through; a position is a configuration of anxiety and defence to which the mind returns whenever the load exceeds what it can integrate. In the 1935 paper on the psychogenesis of manic-depressive states and the 1940 paper on mourning, she described what follows when the split closes: the recognition that the object attacked and the object needed are one and the same, the guilt this brings, and with the guilt the wish to repair. Reparation is not apology and not sentiment; it is the sustained labour of restoring an object one has damaged, in phantasy and sometimes in fact. It is also reversible. Under sufficient pressure, the whole object comes apart again into its two ghosts, and the person who yesterday could hold ambivalence finds himself surrounded by enemies and saviours.

The move from a nursery to a sector is not mine. Elliott Jaques, in Social Systems as a Defence against Persecutory and Depressive Anxiety (1955), proposed that institutions are used by their members to externalise and share defences they cannot sustain alone, so that an org chart is among other things a settlement about what nobody has to feel. Isabel Menzies Lyth tested this against a London teaching hospital and published the result in Human Relations in 1960: the nursing service had organised itself around the task list, so that a nurse performed one procedure on forty patients rather than forty procedures on one, and she read this fragmentation not as an efficiency but as a defence, a way of preventing the nurse from meeting any patient whole and therefore from being exposed to the anxiety that a whole patient carries. The service that split its object was not irrational. It was managing something unbearable at a cost it declined to name.

Consider radiology, where the split has been most publicly performed. Speaking in Toronto in 2016, Geoffrey Hinton advised that we should stop training radiologists at once, since it was completely obvious that within five years deep learning would read images better; the radiologist, he said, was the coyote already over the edge of the cliff. Two things are now true and are almost never held together. The vacancy did not arrive on that schedule, and in many health systems the reading backlog grew. Meanwhile, at the reading desk, the arriving system is not experienced as the emancipator described in procurement but as something closer to a persecutor: it triages, it flags, it produces alerts that must be dispositioned, and it establishes an asymmetry in which the radiologist who overrides it and is wrong has committed two errors instead of one. Lisanne Bainbridge named the mechanism in Ironies of Automation (1983): automation takes the tractable portion of a task and leaves the human the residue that could not be automated, then requires him to monitor a process he no longer performs, with the predictable decay of the very competence on which the emergency handover depends. The idealised object of the board deck and the persecuting object of the workstation are one system, deployed once, seen twice.

Law shows the same division running along a different seam. In Da Silva Moore v. Publicis Groupe (S.D.N.Y. 2012), Magistrate Judge Andrew Peck held that computer-assisted review was an acceptable method of searching electronically stored information, and the profession's relief was audible, because document review was the part of litigation that partners despised and juniors endured. This is the pattern I want to name: *the drudgery split, the readiness of a trade to surrender first the operation it holds in contempt, and the discovery afterwards that the despised operation was the ligament by which the loved one was learned. Michael Polanyi's formulation in The Tacit Dimension (1966), that we can know more than we can tell, is not a decorative epigraph here but a description of what happened in those review rooms: the junior acquired the smell of a case, the sense of which document was load-bearing, by handling thousands that were not. Harry Collins, in Tacit and Explicit Knowledge* (2010), distinguishes the collective tacit knowledge that is acquired only by immersion in a practice from the somatic and relational kinds, and it is precisely the collective sort that document review was quietly manufacturing. Automate the hated task and you have not merely saved a cost; you have removed the apprenticeship and left the profession to discover in a decade that it has partners who have never been young.

In commercial translation the persecution is more intimate still. The industry has rebuilt its rate structure around post-editing, on the premise that correcting a draft is easier than composing one, and the translator now confronts a text that is fluent, plausible, and wrong in ways that require her full competence to locate while paying her as though she were proofreading. Wilfred Bion supplies the exact instrument. In his account of thinking, the absence of a satisfying object is not experienced by the immature mind as an absence but as the presence of a bad object; the no-breast is a thing, not a lack. The blank page is an absence that a translator can fill with authority. The machine's confident paragraph is a bad object present, an entity to be attacked, and the working day of the post-editor becomes a sequence of attacks on links she did not make, in a text that will bear her name and none of her syntax. Session musicians and library composers are meeting the same object from the other side: their catalogues have been treated as training material by an industry that licenses the craft while professing that the craft was never difficult.

At the level of the firm this behaviour has a Kleinian name that fits it exactly. Klein described the manic defence against depressive anxiety as a triad of control, triumph, and contempt: the object is not mourned but mastered, gloated over, and belittled, precisely because mourning it would require admitting that it was loved. I propose *manic procurement for the institutional version: the deployment programme whose real function is not throughput but the avoidance of grief for the trade being dismantled. Its control appears as the dashboard that renders a practice as a set of monitored rates; its triumph appears as the headcount slide, in which the reduction is the achievement rather than a consequence of one; its contempt appears in the vocabulary of routine cognitive labour applied to work that took eleven years to learn. Max Weber saw the wish beneath it in Economy and Society, where bureaucracy is said to develop the more perfectly the more it is dehumanised, eliminating love and hatred and every purely personal element, sine ira et studio: an administrative order designed so that nobody in it need feel ambivalence about anything. That is the paranoid-schizoid wish drawn as an organisational chart, and Jacques Ellul was right in The Technological Society* (1954) that such a system augments itself, since each defence generates the anxiety that requires the next.

There is a further transaction, and it is the one that most repays attention. The firm's difficulty is not the machine; the firm's difficulty is its own aggression toward its workforce, which it cannot own and therefore decants into the vendor's system, so that the redundancy arrives as an actuarial fact rather than a decision by named persons in a room. The machine becomes the *designated persecutor: the object into which an employer projects the hostility inherent in the wage relation, whereupon that hostility returns wearing a logo and everyone in the building can agree that nobody wanted this. Marx had already described the material half of the operation in the fifteenth chapter of Capital*, where machinery figures not merely as a productivity gain but as the most powerful weapon for suppressing strikes, a power confronting the worker as inimical to him. Klein supplies the half he did not need: why the weapon is so useful to the mind that wields it, and why the sector that has projected its aggression into a model will thereafter experience the model as fate.

I have argued elsewhere that the useful unit of analysis in automation is the task rather than the occupation, since jobs are bundles and only bundles get displaced. That formulation now looks to me like the disease presenting as the diagnosis. Decomposing a trade into tasks, sorting them into the automatable and the residual, is not a neutral analytic act; it is the splitting operation itself, performed at consultancy rates, and it is the exact procedure Menzies Lyth found in the task list that kept the nurse from meeting the patient. The task frame permits an industry to congratulate itself on removing what it hated from work it never had to grieve, because a trade decomposed into tasks is no longer an object anyone could be sad about. The whole object of a practice includes its unpaid hours, its wasted motions, its apprenticeship overheads, and the tacit stock which no ledger has ever carried as an asset.

What, then, would the depressive position look like in a sector rather than a person? Not reconciliation, and certainly not the reassurance that displaced workers will be retrained into something. It would look like the capacity to state without flinching that the same system finds the small lesion and empties the room where radiologists learned to find it; that predictive coding is both a genuine advance in the discovery of truth and the demolition of the ladder by which advocates were made; that these are properties of one object, deployed by identifiable people, and that the damage is real and repairable at a price. Thomas Aquinas, holding with Augustine that evil is a privation rather than a substance, offers the structural point that the split cannot tolerate: a defective good is not a demon, and a model's failure is a shortfall in a distribution rather than a malice, which is why the mind that requires a persecutor can never make accurate use of one. Reparation in Klein's sense is driven by guilt and by love toward a damaged object, and here the analysis must stop short of comfort: a firm is not a subject, cannot mourn, and possesses no organ of memory except the balance sheet. The integration, where it occurs, will be the work of bodies that outlive quarters, the colleges and faculties and unions and licensing boards that carry a practice across generations and can therefore afford to notice what a decade of removed apprenticeship will cost in twenty years.

A trade does not die of automation. It dies of splitting: of the refusal to hold in one view the instrument that reads the scan and the instrument that empties the reading room, since whichever half is disowned returns as an unnegotiable force. The finding to be carried forward is this. The integration of the split is not a threshold that an industry crosses once but a labour it must perform continuously and can lose under any sufficient shock, and the institution that proves able to bear the ambivalence, rather than the one that resolves it fastest in either direction, will own the trade that emerges.