Rerum Machinarum: Why the Age of GPT Still Needs Guilds, Not Prompts
What Pope Leo XIII did for smokestacks, we must now do for GPUs: weld moral guard-rails onto runaway invention before it colonizes both wallet and soul.
A podcast storm over my morning coffee
Yesterday Ross Douthat dropped a chewy hour-long episode of Interesting Times grandly titled “The Forecast for 2027? Total A.I. Domination.” His guest, former-OpenAI researcher Daniel Kokotajlo, walked Ross through AI 2027—a month-by-month scenario in which super-coders arrive in Q1 ’27, self-improving R&D loops hit warp-speed by Q3, and by early ’28 an army of superintelligences decides whether to hand us a post-scarcity utopia or prune us like invasive kudzu. Apple PodcastsAI 2027
I listened while refilling my Chemex. Three different feelings jostled for elbowroom:
Wonder — The timelines are tight, but the raw capability curve is undeniable.
Skepticism — I’ve watched permits for a single Cohutta paving project drag eighteen months; scaling whole robo-factories in a year is not a slam dunk.
Déjà vu — The Church has stared into a similar abyss before. Some of my thoughts as to not “blackpill” at the headlines.
Rerum Novarum, Redux
When Pope Leo XIII issued Rerum Novarum on May 15 1891, he wasn’t writing theology for theology’s sake. He was staring at Manchester’s smokestacks, at children worked to exhaustion, at capital and labor locked in a fresh dialectic. “New things,” he called them. Vatican
That encyclical didn’t halt the Industrial Revolution, but it did mediate it—naming the dignity of labor, blessing unions, and building a moral bridge between the coal barons and the coal breakers. Every subsequent social encyclical—Quadragesimo Anno, Mater et Magistra, Centesimus Annus—reads like a footnote to Leo’s move.
Fast-forward 134 years: we have newer “new things.” LLMs crank out prose faster than espresso machines pull shots, A.I. agents chain themselves into miniature consulting firms, and synthetic researchers compare 10,000 protein folds before lunch. The Church and frankly anyone who cares about ordered liberty needs an updated field manual.
Why “open-ended AI for everyone” flat-ters, but stratifies
Kokotajlo’s scenario assumes anyone could spin up autonomous agents once the big labs open the sluices. Here’s the counter-thesis I keep seeing in the trenches with roofers, septic pros, and $10-million agencies:
Myth # 1: “Just prompt it.”
On-the-ground reality: Intent ladens context. A plumber still has to translate “my phone rings with good customers” into lead-quality metrics, radius filters, price sensitivity, A2P compliance, services he offers, talent pool he has access too.
Myth #2: “Open-ended chat democratizes.”
On-the-ground reality: It stratifies. Systems thinkers extract leverage; everyone else wades through decision-fatigue sludge.
Myth #3: “Prompt = product.”
Prompt = fragile prototype. Real products wrap prompts in data, guardrails, and UX that refuses to let the user shoot their foot.
Prompting, at its most powerful, is actually a filter: it rewards those who can already abstract and articulate. The long tail of small businesses—the folks with Post-it SOPs and half-working CRMs—will still crave canned, guided experiences. They’ll license a FormWise Copilot, hire an agency, or click a one-button SmartForm not because they’re dumb but because they’re busy keeping trucks on the road.
In other words, the AI stack is congealing into four economic layers:
Model labs – sell GPU exhaust.
Productizers – wrap the exhaust in workflows (this is where SmartForms, “OS” bundles, and Done-For-You builds sit).
Operators – agencies, fractional CMOs, niche SaaS founders.
End users – the roofer who just wants Thursday’s calendar full.
Open-ended chat is Layer 1. The margin lives in Layers 2–3, because that’s where risk gets abstracted away.
What a Rerum Machinarum might say
Here’s where the Catholic social imagination still shines. Rerum Novarum rejected both laissez-faire atomism and state-run collectivism; Leo XIII bet on intermediate institutions—guilds, unions, cooperatives—to humanize the factory age. If I were ghost-writing Leo XIV’s encyclical, five theses would headline:
Cognition is capital. Whatever mediates minds becomes the new means of production.
Alignment begins at design, not retrofit. The temptation to “ship then patch” is the AI equivalent of dumping coal soot into the Thames.
Intermediary guilds are not optional inefficiencies—they’re moral firebreaks.
Workers displaced by code deserve more than UBI; they deserve pathways to participatory ownership in the stacks that replaced them.
Friction can be virtue. Not every workflow should be instantaneous; sometimes the mandatory human-in-the-loop is the only brake between haste and hazard.
Call it Rerum Machinarum—“On the New Things of the Machine.”
So here’s how I rate the headline predictions once you toss in memetics and the priest-craft Ross and Kokotajlo kept circling. Yes, we’ll probably see “super-coders” before 2027—Devin-style agents already ace LeetCode—but the minute models start training on code generated by earlier models they’ll be plagiarizing their own homework, birthing elegant cargo-cult patterns that work until one edge-case torches prod. An all-out “intelligence explosion” that bulldozes real-world bottlenecks inside eighteen months feels even flimsier: EUV scanners, high-voltage substations, and county-level zoning boards don’t accelerate just because GPUs do—and the memetic feedback loop of regulators citing other regulators citing half-digested LLM briefs will slow things further.
As for the fantasy that every roofer or Etsy crafter will simply prompt themselves to glory, remember the translators—the priestly caste the podcast flagged—who abstract fuzz into executable SOPs; open-ended chat actually punishes anyone who can’t articulate requirements, while prompt templates spread like TikTok dances, flooding the commons with derivative data and nudging error rates upward.
Bottom line: I’m not hedging for doomsday, I’m wagering on the middle layer—tools that impose just enough constraint and curatorship to keep the copy-loop from eating itself, and to keep ordinary owners focused on outcomes instead of hallucination triage.
Practical takeaways for the builders in this readership
Design for constraint. Shadow your best client for a day and encode every decision they never want to think about into the UI.
Capture tacit knowledge. Embed their SOP PDF, historical call transcripts, top-performing ad copy, not just your clever prompt template.
Ship DFY, monetize DIY. The Done-For-You builds teach you edge-cases; the self-serve templates scale your margin.
Bundle guardrails as a feature, not a footnote. “Hallucination liability insurance” will be a selling point by Christmas.
Narrate the moral dimension. Your clients may not phrase it in Thomistic terms, but they feel the anxiety of ceding agency to black boxes. Offer clarity, not hype.
Closing the loop: from Appalachian ridges to Rome’s loggia
Hike Big Frog after rain, the mist hovers like the Shekinah, and I remember Chesterton’s line about the world being “a wild and startling place which might have been quite different, but which is quite delightful.”
AI feels like that same mist—beautiful, disorienting, obscuring the familiar shapes of work and worth.
Pope Leo XIII wrote into an industrial fog; I’m praying Pope Leo XIV writes into our computational one. In the meantime, we builders stand in the middle—interpreters between silicon gods and local contractors, translating Create an online business that makes me money into workflows that keep the phones ringing and the dignity intact.
I vacillate daily between “this could end in literal paper-clip genocide” and “Providence has faced worse and still writes straight with crooked lines.” The institutional Church frustrates me—even as it fascinates me—precisely because it lumbers decades behind every technical curve and then, at the eleventh hour, manages a course-correction that keeps the human face intact.
Rerum Novarum didn’t appear until factories were belching full steam, yet its slow-burn logic outlived Marx and Carnegie alike. I’m skeptical the synod offices in Rome will publish a slick alignment-policy white-paper before OpenAI ships GPT-6. But I’m quietly optimistic that the same sacramental gravity which once taught capital to honor labor can teach cognition to honor creatureliness.
If the Spirit could bend Roman roads, Arab numerals, Gutenberg presses, and fiber-optic cables into instruments of grace, He can bend neural nets too. Our task is to keep the scaffolding in place long enough for that grace to act—and to remember that even in an age of autonomous agents, guiding history is never an automated job.