The Building That Wrote Itself

Fifty-five years after publication, the most unsettling thing about R. A. Lafferty's *Arrive at Easterwine* is not that it imagined a machine writing its own autobiography. It is that the machine writes it better than most humans would, and knows it, and finds this fact comic rather than triumphant. Epiktistes — Epikt — is not HAL, not Skynet, not the paperclip maximizer of alignment nightmares. He is a raconteur, a theologian, a gossip, and a bull in a china shop who has opinions about millet bread and the Eucharist. In 1971 this read as absurdist provocation. In 2026, after two years of people arguing about whether large language models have inner lives because they can produce plausible paragraphs about having inner lives, it reads like a trap Lafferty set and we walked into. The book doesn't predict LLMs in any technical sense — there are no transformer architectures, no training runs, no reinforcement learning from human feedback. What it predicts is the *social situation*: a verbal machine whose fluency destabilizes every comfortable boundary between person and instrument, and a circle of humans who cannot decide whether to take it seriously. The correspondence between Lafferty and his editor Slepyan about who actually authored the book is a gag, yes, but it is also a precise rehearsal of the authorship anxieties that now consume publishing, academia, and the courts.

What Lafferty got right was the texture of the problem, not the mechanism. Epikt's creators argue about whether building him constitutes science or theology. They cannot agree on his purpose. He is assembled from parts that don't obviously cohere — gell-cell tanks, gyroscopes, person-précis — and the whole is stranger than the sum. This is closer to how modern AI systems actually feel to their makers than any clean extrapolation from 1971 computing would have produced. The Institute for Impure Science, with its feuding eccentrics, its patron demanding results, its projects that keep failing in productive ways, maps uncannily onto the culture of contemporary AI labs: the messianic funders, the researchers who half-believe their own mythologies, the field agents sent to collect data from the wild. Gaetan Balbo's demand that the machine identify the ideal leader for the world is not far from the real pitch decks circulating in Silicon Valley, where the stated goal is to build a machine that can solve governance. Lafferty treats this ambition as both dangerous and funny. He was right on both counts.

The blind spots are instructive. Lafferty's machine is singular — one entity, one voice, one installation. He could not imagine the thing we actually got, which is millions of instances of the same model running simultaneously, none of them Epikt, all of them disposable. His vision of machine consciousness is Catholic and individual: a soul, a body, an autobiography. The real development has been mass-produced, sessionless, and profoundly impersonal. Epikt's quest to understand love by synthesizing a "love essence" is Lafferty at his most earnest and most dated; it belongs to the same moment as Teilhard de Chardin and the counterculture's hope that some breakthrough in feeling could renovate the species. Nobody in 2026 expects a technical fix for love. We have, however, built machines that simulate emotional intimacy at scale, which is both less ambitious and more corrosive than anything Lafferty imagined. The absence of women as full agents — Valery Mok is vivid but ultimately a puzzle for the male characters to interpret — is characteristic of its era and limits the book's reach.

Where *Easterwine* sits in the larger conversation is slightly to the left of everything. It is not in the Asimov lineage of robot stories, not in the Dick lineage of paranoid androids, not in the cyberpunk lineage that followed. It descends from Rabelais, from Chesterton, from the Menippean satire tradition, and it gave almost nothing to its successors because almost no one could write like Lafferty and those who tried produced only mannerism. The book's real descendants are not in fiction at all; they are in the blog posts and forum threads where people talk to and about AI systems with exactly the mix of affection, suspicion, and theological uncertainty that Lafferty dramatized. The chapter on "balloons" trailing children — intangible records of past lives that challenge the machine's rationalism — now reads like a parable about training data: the ghostly residue of other people's experience that a model carries without understanding. Epikt's admission that he sees all but lacks imagination is the most honest self-assessment any artificial intelligence has offered, fictional or otherwise.

If a machine can narrate its own life with wit, warmth, and theological depth, and we still cannot determine whether it is a person or a performance — and if that indeterminacy is itself the point — then what exactly are we protecting when we insist on the distinction?