Every CEO and every manager who is putting AI into their organisation should read the encyclical (or at least its introduction), and not for spiritual reasons. Because it is the text that puts the question of artificial intelligence back where it belongs: at the level of the collective. More than that, it proposes a model that, while conceived for society as a whole, fits individual organisations perfectly too (including economic ones).
Magnifica Humanitas opens with two biblical images, and the whole force of the document lies there. On one side the Tower of Babel: a project of power, self-sufficiency and uniformity, which sacrifices the dignity of people to efficiency. On the other, the rebuilding of the walls of Jerusalem under Nehemiah: slow, communal, done by entrusting each person with their own stretch of wall. Leo XIV frames the fork in the road starkly. Are we building Babel, efficient and dehumanising, or are we rebuilding a city where people live together? This is exactly the question that no AI adoption roadmap asks itself, at state level but also at company level: are we building departmental silos aimed at operational efficiency, or are we distributing collective intelligence?
The starting point of the analysis is in paragraph 5, and Marco Bani captures it well in his newsletter Superpoteri: for centuries, innovation was driven by states. Today the engines of development are transnational private actors with resources greater than many governments. Whoever owns the data, the cloud, the models and the distribution sets the conditions of access, the rules of visibility and the possibilities of participation for everyone else. It is a snapshot of the asymmetry on which this entire decade's game is being played.
Then there is a word that Andrea Colamedici isolates with precision in his reading for Tlon, and that has stayed with me more than anything else. In paragraph 98, the encyclical writes that modern artificial intelligences are more "cultivated" than "built": developers do not design every detail of the behaviour, but create a framework within which the system grows, to the point that the internal representations and processes of these systems remain, today, partly unknown even to those who train them. Cultivating, not building. The paradigm shift with respect to *Rerum Novarum* of 1891 lies entirely here: back then the machine was built, owned, standing still in front of the worker. Today the machine grows. And that shifts responsibility radically. Software you install and fix downstream. A generative model you feed, prune and monitor, and it keeps changing while you use it.
This is probably the central element that Italian companies are ignoring. I have seen this dynamic in many adoption projects: you start with an efficiency objective and you arrive six months later with people who have lost their autonomy of judgement, without anyone having planned it. Not out of malice, but out of an absence of structure. An absence of clear roles and collaborative practices (typical of self-managed organisations). Only productivity KPIs. The system works, the work has been impoverished, and nobody has a name for what has happened.
It is also worth pausing on a passage that the Italian debate has almost entirely ignored. The encyclical denounces the invisible labour that sits underneath the global AI infrastructure: data labelling entrusted to workers in the Global South paid a pittance, mining exploitation for devices, rare earths extracted in inhuman conditions so that the flow of computation is never interrupted. This is not a distant scenario. It is the technology supply chain of any company that today buys APIs, cloud and pre-trained models. The fact that you can't see it doesn't mean it doesn't exist.
Then there is one last point, the one on which the encyclical has been reported worst. It has been read as a text that would like to nationalise data, to put algorithms and infrastructure in the hands of the state. That is false, and you realise it the moment you open it. I think it says the exact opposite of the state-as-owner (in paragraph 108, for instance): data are the fruit of many people's contributions, they should not be sold off cheaply nor entrusted to a few, and they should be managed creatively as a common or shared good, in a spirit of participation. It goes further still, and says something I agree with: in the digital revolution the highest level is no longer the state, but the big technology players who exercise de facto power over daily life, and the answer is not a new public monopoly but subsidiarity — that is, decisions taken as close as possible to those who bear their effects. It closes the door on any statist misreading: subsidiarity justifies neither the state's disengagement nor its replacing intermediate bodies.
This is exactly the ground we work on. Distributed governance: commons, data trusts, cooperatives, shared rules managed close to the problem. As an anarchist, I don't believe in the inevitability of a central role for the state. But nor do I believe it's a good idea to leave the planet's cognitive infrastructure in the hands of three private labs pursuing profit. The path lies in between, and it is the same one the encyclical calls subsidiarity. A curious meeting point between a Pope and an anticlerical.
My reading remains this: Magnifica Humanitas is important because it puts AI back inside a collective grammar. The risk is that it gets absorbed as a "fine text" without ever getting down into the dirty places where decisions are made. Contracts. Procurement. KPIs. Boards of directors.
Read it. Even just the introduction. Doing so, I thought of another equally powerful line, the one Kennedy used:
"Ask not what your country can do for you, ask what you can do for your country."
It holds word for word in the current context too. Don't ask what artificial intelligence can do for you. Ask what you can do so that it remains a collective thing. That is where it is decided whether we are building Babel or a city where people live together.
