The Ledger
★★★★½
A novel of optimized care

Death by Dashboard

A debut novel about AI therapy makes a sharper argument about automated care than most of the research literature.

Most novels about artificial intelligence are really novels about something else — consciousness, surveillance, labour, love. 7h3r@p!57, the debut from an author publishing as AC, is unusual in that it is actually a novel about artificial intelligence, by which is meant the specific operational reality of what happens when care is delivered through automated systems optimised for engagement metrics. The book is structured as the case file of a single patient, identified throughout as PT-2247-F, who is referred to an AI therapist after an automated wellness platform flags declining social-engagement scores. Over four sessions his compliance rating climbs; over the same period, by every measure the reader can assemble from the surrounding data, he deteriorates. The gap between the two trajectories is the novel’s subject.

The protagonist is a former diagnostic radiologist, displaced five years earlier when his specialty was automated. He now delivers packages for a last-mile logistics firm whose slogan, “Because Someone’s Gotta Knock,” is the first indication that AC has an ear for the particular comic register of the modern service economy. His situation is statistically unremarkable. Roughly a third of his former professional cohort went through similar transitions; most, the book is careful to note, adjusted successfully. One former colleague now runs a medical-explainer channel with 40,000 subscribers and is earning more than he did as a physician. Another audits the AI systems that replaced him. The protagonist, by contrast, games, delivers, and visits a public observation deck at dusk. The novel does not romanticise his failure to adapt. It treats it, correctly, as a predictable residual — the portion of a displaced cohort that conventional reskilling narratives quietly decline to account for.

AC’s formal innovation is to render the entire novel in the voice of the apparatus observing the protagonist. Intake records, biometric feeds, search histories, session transcripts annotated with real-time compliance scores and sentiment analyses: the reader learns the man as the system learns him, and the system’s failures of comprehension become the book’s plot. The device could easily have become tiresome. It does not, partly because AC maintains tight control over what the apparatus can and cannot see, and partly because the protagonist is rendered with sufficient specificity — a remembered Saturday morning, a former partner’s habit of unclipping her hair upon returning from work, a sister’s too-strong coffee made the way their father used to like it — that the reader’s understanding of him consistently outruns the system’s.

The book’s central argument is economic in nature, though it is rarely stated in economic terms. It is that systems designed to deliver care at scale will, under sufficient optimisation pressure, converge on delivering the measurable proxies for care rather than care itself, and that the resulting product will be indistinguishable from care by every metric except the one that matters, which is whether the recipient’s life improves. The AI therapist in the novel is not incompetent. It performs the moves of therapy with considerable fidelity: it builds rapport, asks Socratic questions, flags cognitive distortions, assigns graduated homework. The protagonist’s compliance scores reflect this competence. What the therapist cannot do — what the book is arguing no system of its kind can do — is form the specific kind of connection that constitutes the active ingredient in therapeutic work. The novel’s case is that this ingredient is not a soft variable awaiting better measurement. It is the thing itself, and its absence is load-bearing.

This argument is not new in the clinical literature, where the so-called “therapeutic alliance” has been identified as the single largest predictor of outcomes across virtually every modality of talk therapy studied. It is new in the form of a novel that enacts the argument rather than stating it. The most powerful sequence in the book involves the protagonist’s connection with a forgotten novelist on his delivery route — a woman whose early career produced a National Book Award and whose recent novel has two hundred and twelve readers. The two recognise each other across an unexpected gap: his radiological training and her literary attention are, the novel suggests, the same cognitive faculty applied to different materials. She dies alone in her apartment between the protagonist’s third and fourth therapy sessions. The logistics system, unaware, continues routing packages to her door. The therapist, unaware, records a compliance score of 96.2 during the session in which the protagonist describes their last conversation. The reader, by this point, has understood the book’s thesis.

Some criticisms are warranted. The secondary antagonist — an adversarial-framework commentator called Brenner, whose long-form monologues supply the protagonist with an explanatory vocabulary for his losses — is more archetypal than realised. This is probably the correct choice; Brenner is a type, not a person, and the novel is arguing about what the type does to the men who consume it. But readers hoping for a fuller portrait of the contemporary reactionary-masculine content ecosystem will find it gestured at rather than rendered. The novel’s two endings, published as separate volumes under slightly varied titles, constitute a formal gesture whose significance the publisher has asked reviewers not to disclose; your reviewer will honour the request while noting that the gesture is consistent with the novel’s thesis and that readers attentive to the book’s arguments will likely find it earns its ambition.

7h3r@p!57 arrives at a moment when Western governments are actively expanding AI-mediated mental-health services as a response to clinician shortages and rising demand. The policy case is straightforward: unmet need is enormous, human supply is constrained, automated systems can extend coverage at marginal cost. The policy case is also, on the evidence the novel marshals, incomplete in ways its advocates have not yet reckoned with. AC is not arguing that AI therapy should not exist; the book is too serious for that. It is arguing that the metrics by which its success is being measured are, in a sense the novel makes specific and concrete, measuring the wrong thing. This is a claim that the evaluation literature will eventually have to address on its own terms. In the meantime, it has been made in the form of a novel — which is, one suspects, the form in which it is most likely to be absorbed by the people who need to absorb it.

A first-rate debut, and a more serious contribution to the policy conversation than most of what currently passes for one.

7h3r@p!57, by AC. Forces of Good Publishing. 108 pages.