Magnifica Humanitas

135 — the years between Pope Leo XIII signing Rerum Novarum on May 15, 1891 and Pope Leo XIV signing Magnifica Humanitas on May 15, 2026 — to the day. Yesterday he released it at the Vatican with Anthropic's interpretability lead Christopher Olah standing next to him. The same week, Microsoft canceled Claude Code for ~100,000 of its own engineers because finance couldn't afford how much they loved it. The Pope picked a fellow seeker of the truth. Your CFO picked the leaderboard. One of those is a distribution deal.

Stop measuring AI usage by the column. Measure it by the slope. Microsoft just canceled Claude Code for ~100,000 engineers because the CFO measured total spend instead of productivity-per-token. Uber ran an internal leaderboard ranking engineers by usage volume — adoption hit 84%, the 2026 AI budget died in April. The right leaderboard ranks employees by the slope of their improvement curve across the quarter — who turned their token budget into a 20% better output next week, then a 20% better one the week after. Build that report. Kill the usage report.

Distribute small token budgets widely. Find your Verstappens. The employees whose learn-fix-rerun cycle is 20% tighter than everyone else's are not on your performance review system's radar. They might be a senior staff engineer. They might be the operations analyst quietly using Cursor on her own time for eighteen months. The Maxinomics anecdote from last week is the canonical case: a Fortune 50 employee, the only person on her team shipping at scale, burned through her token allocation in one day and is now petitioning three layers of management for more. She is the only one in the building who knows how to drive the car. Protect her from the next round of layoffs — don't run her over with it.

Read Magnifica Humanitas before your competitors do. Pick a side. Pope Leo XIV released the 42,300-word encyclical at the Vatican yesterday with Anthropic's interpretability lead next to him. It will be the reference document for every AI governance conversation in every legislature, newsroom, and union hall for the next ten years. Companies that align with its framing — AI is a tool that requires vigilance — get a defensible posture. Companies that align against it become Bethlehem Steel by 2030. Pick a side. The Pope already picked his.

Today's actions touched on the measurement system around AI deployment and the moral framework forming around it — both conversations we have weekly with clients trying to figure out which of their employees are the actual drivers and which AI investments are about to become liabilities. If your CFO is about to cancel a Claude Code rollout, that's the call we're built for.

Outsider Labs Advisory →

Tuesday, May 26

1. The Pope just made a distribution deal. Pope Leo XIV released Magnifica Humanitas — his first major encyclical, 42,300 words, signed exactly 135 years after Rerum Novarum reshaped Catholic doctrine for the Industrial Revolution — at the Vatican yesterday. Standing next to him: Christopher Olah, Anthropic's interpretability lead. Not Altman. Not Amodei. A fellow seeker of the truth — the researcher whose entire career has been about making the AI black box legible to human inspection. That is not a photo op. That is a 1,500-year-old institution signing a strategic alliance with the people building the dictionary of the next dominant medium. (Full analysis below.)

2. The round-trip is breaking before the round-trippers finish counting. Microsoft put $5 billion of equity into Anthropic last November on the same day Anthropic committed $30 billion right back to Azure compute — six dollars of cloud-services revenue for every one dollar of investment. This week, Microsoft told ~100,000 of its own engineers to stop using Anthropic's Claude Code by end of June because token-based pricing made the internal bill untenable. Uber went 32% → 84% adoption in four months, then the 2026 AI budget was "blown away already" by April; CTO Praveen Neppalli Naga himself burned $1,200 in a two-hour demo. The AI subsidy era is ending in real time — and the people who measured use more are discovering they should have measured use better. (Full analysis below.)

3. DeepSeek just permanently set the floor at 25¢ on the dollar. V4-Pro pricing dropped 75% on Sunday — permanently, not promotionally. The model now runs at roughly a quarter of the single-token compute and a tenth of the memory footprint of its predecessor at long context. Analyst quote from InfoWorld: "It is not a discount. It is an efficiency gain being passed through." CIOs who built workflows assuming OpenAI and Anthropic premium pricing would hold are about to renegotiate. The labs' bull case for IPO multiples requires sustained pricing power. That case just got harder.

4. Anthropic bought its way into Palantir's foxhole. Acquisition of Fractional AI yesterday — generative AI enterprise applications — plus a new consulting venture launched in the same press release. Translation: Anthropic is no longer content being a model vendor. They want the forward-deployed-engineer business Palantir has run for twenty years. Combined with the Stainless acquisition eight days ago — the SDK generator for every major AI lab — this is a firm operating posture, not a lab posture. The IPO is coming.

5. IREN's CEO put a date on the gigawatt timeline. It is 2030. Daniel Roberts on Bloomberg Tech in eleven words: "If you wanted to start today and build a gigawatt AI factory, you are looking 2030 before you get the first compute online." Utilities need 18-24 months just to determine whether a site has available power. The hyperscalers' $725 billion 2026 capex does not change that timeline — it just confirms how many people are trying to bid for what is already mostly bought. Power is the new GPU.

The Same Week

Yesterday morning at the Vatican, the most influential moral institution on Earth signed a strategic alliance with the AI lab whose entire research program is making the model legible. The same week, the most valuable software company on Earth told 100,000 of its own engineers to stop using a product its own portfolio company built — because the engineers loved it so much that the bill became unaffordable. One institution moved toward AI. One institution moved away from it. Both moved decisively. Both moved in the same news cycle.

These are not two stories. They are one story told from two angles. The angle is: who has the authority to act on AI right now, and who is going to be left adjusting around it later.

Pope Leo XIV has founder-authority. He inherited it from Peter. He can sign an encyclical unilaterally, and the College of Cardinals cannot vote it down. He signed Magnifica Humanitas on May 15 — exactly 135 years after his predecessor Leo XIII signed Rerum Novarum — and released it yesterday with Olah at the lectern. He didn't pick the salesman. He didn't pick the CEO. He picked a fellow seeker of the truth — the interpretability researcher whose career has been about making the AI black box legible to human inspection. The Church is not buying paintings. They are buying the dictionary. The same move Cosimo de' Medici made in 1462 when he funded Marsilio Ficino to translate Plato into Latin and build the philosophical infrastructure the Renaissance ran on for the next two centuries.

Microsoft, on the other hand, has hired-manager authority. The CFO who killed Claude Code for 100,000 engineers is doing what hired managers do under cost pressure: he is straw-polling finance, optimizing the column on his spreadsheet, and missing the report he is not running — which of those engineers turned their tokens into compounding output this quarter. He measured spend. He did not measure slope. Uber did the same thing on a bigger stage at faster speed — adoption climbed from 32% to 84% in four months, the budget died in April, and the company's internal leaderboard ranked engineers by token consumption when it should have ranked them by improvement-curve slope. The Maxinomics anecdote makes it explicit: a Fortune 50 employee, the only Verstappen in her building, burned through her token allocation in one day shipping work that everyone in the office called her for, and is now petitioning three layers of middle management for more. The measurers cannot see what they are not measuring. They are about to lay her off in the next round.

The Church has lived this pattern four times in five hundred years. They lost the printing press to the Reformation, the university to the Enlightenment, the broadcast tower to the network executives, and the internet to the technologists. Each time they lost the medium, they lost it to a hired manager — a printer who sided with Luther, a Berlin chancellor who made theology a department, a TV programming executive who scheduled politics over liturgy, a platform product manager who decided what got into the feed. Each loss happened because the Church arrived too late and tried to teach the catechism harder on a medium it did not control. This time they are not making that mistake. Magnifica Humanitas is not a complaint about AI. It is a positioning document. The Catholic Church and the effective altruism-adjacent AI lab are at the same table not because they share a metaphysics — they don't — but because they want the same governance outcome: legible compute, distributed power, dignified labor. They came at the agreement from opposite ends of the philosophical map and arrived at the same destination.

Your CFO is moving in the opposite direction. He is cutting the people who know how to drive. He is canceling the tools they love. He is rewarding the leaderboards that measure the wrong axis. He is doing it because hired managers manage by coalition and short-term cost, and founders are in increasingly short supply. The clock on founder-authority is the clock to the IPO — and OpenAI is reportedly filing this week. The window in which the people running the AI industry can still make alliances like yesterday's is narrowing.

Pick your side now. The Pope already picked his.

Drive the car. Tune the car. Don't count the laps.

At COAI today: Full analysis of the Pope-Olah handshake, the Cosimo/Ficino parallel, and what Magnifica Humanitas means for AI governance over the next ten years at getcoai.com/news-letter/magnifica-humanitas.

— Harry and Anthony

Sources: