AI has turned access to information into a commodity. A model summarises documentation, proposes code, drafts a plan and explains any concept to you before your coffee goes cold. It is enormously useful, and it is less and less of a differentiator, because tomorrow your team, your competitors and half the industry will have access to the same models, the same courses and the same playbooks. If everyone has exactly the same thing, the advantage has to be found somewhere else.
It is not in piling up open tabs or signing up for yet another e-learning platform and celebrating the percentage of courses completed. It lies in something considerably harder to buy: the ability to connect all that information with context, experience, your own mistakes and real decisions. It has an old-fashioned name, judgement, and one inconvenient property: it cannot be downloaded. The explanation of why that is so is a hundred and thirty years old, and its implications for any company spending its entire training budget on licences are considerable.
The mysteries are in the air
In 1890, the economist Alfred Marshall asked why trades prospered when clustered in industrial districts, and left one of the most quoted observations in economic history: in those places, the mysteries of the trade become no mysteries; they are in the air. An apprentice absorbed more metallurgy walking between the workshops of Sheffield than by studying every manual available. The economics of agglomeration that Marshall founded still explains today why talent crowds into wildly expensive cities when, in theory, it could work from anywhere.
With generative AI, that observation has stopped being an academic curiosity and become an investment criterion. Everything in the manuals, the model already knows: it has read them cover to cover, better than any of us. What sits outside that price is exactly what Marshall placed in the air. Technology can give you answers; what it cannot deliver, packaged in a PDF, is the instinct with which a designer spots that a flow is going to fail before the data confirms it, the commercial nose that tells a real opportunity from a meeting of corporate entertainment, or the intuition of the engineer who knows that this elegant solution is going to blow up in six months. That is learnt by watching, asking and working close to others.
What the model knows and what the room knows
Knowledge management has spent decades naming this frontier. Ikujiro Nonaka, the Japanese theorist who turned that discipline into serious subject matter, formalised it in the nineties: explicit knowledge is stored in documents and databases; tacit knowledge only passes from one head to another through socialisation, that is, by sharing experience. It is the same distinction drawn by the Stanford study on AI and young workers’ employment that we analysed in another article on this site: AI replicates what is codified well and what is tacit badly.
And when physical contact disappears, the tacit is the first thing to die. Microsoft measured it almost by accident during the pandemic, analysing the activity of 61,000 employees. The result, published in Nature Human Behaviour, deserves framing: with the whole workforce remote, collaboration time with people from other teams fell by around 25%, and internal networks became more static and closed. Conversations within each team survived without trouble. The bridges between disciplines did not.
It matters because teams that only talk to themselves become very efficient at repeating their own certainties: design talks to design, technology to technology, and everyone optimises their own patch without anyone seeing the whole system. Interesting ideas rarely arise inside a single department; they arise when the person building the product hears how the person selling it reasons, or when a creative director frames a problem a developer would never have thought of. A recommendation algorithm brings you closer to people like you. A well-mixed room forces you to listen to the people you are missing.
A case worth studying: IAHub Barcelona
Since late 2024, Barcelona has had a textbook example of this thesis. And it is worth saying so without romanticism, because the city already has plenty of gatherings where business cards are swapped, collaboration is promised and nobody writes back. IAHub Barcelona is something else: fortnightly meetups where creatives, technologists, artists, educators and business profiles share real work, with more than twenty speakers over two seasons, technical masterclasses and an annual selection of creators. The author of this article knows it from the inside: he has attended since the first sessions, keeps a close relationship with its founders and, cards on the table, holds an admiration for the event and its concept that borders on devotion.
What holds the case together
What is interesting about the case is not the enthusiasm it generates but the mechanics that sustain it, because they are replicable:
Every two weeks, without exception. A one-off meeting can inspire; a recurring appointment creates habit, trust and collective memory. Consistency is what turns an activity into an institution.
Profiles that would not normally share a room, brought together by a common question rather than a sector. The collision between disciplines is not a side effect: it is the product.
Speakers talk about doubts, failed experiments and half-baked discoveries. That material does not exist in any course, precisely because nobody packages it: success stories teach little once they strip out the hard decisions and the shortcuts that were discarded.
The talks are recorded and published in a video library open to anyone.
That last point invites the obvious objection: if the talks are recorded, the knowledge has already been codified. Why attend? Because the video captures the presentation, not the transfer. The tacit does not live in the talk: it lives in the question someone dares to ask out loud and would never write in a form, in the objection that dismantles an idea in time, in the ten-minute conversation with someone who looks at your problem from an angle that did not exist in your team. The video library is an act of generosity and a magnificent tool for spreading the word. But it is the record of the event, not the event.
Generosity without hot air
Here it is worth separating two things that get mixed up too often. Charging for training, support or consulting is in no way suspicious: good training does not sell information, it accelerates the application of judgement to a specific context; good consulting does not hand over a template, it takes on uncomfortable decisions with data, constraints and consequences. What AI is exposing is a different business: selling as secret knowledge information that is already published, indexed or available in five minutes to anyone who can put together a decent search. Hot air charges for the wrapping. Serious work charges for solving.
That is why sharing the process in public does not weaken those who know: it singles them out. Anyone with experience does not fear having a presentation copied, because they know their value lies not in a list of prompts but in knowing when to apply them, when to ignore them and what to do when the result goes wrong. And there is a return no spreadsheet will ever attribute precisely: reputation, relationships, opportunities that arrive years later by paths impossible to trace. As an indicator for deciding who to listen to in today’s AI noise, intellectual generosity works better than almost any other.
That knowledge does not travel over fibre optics. You have to go and get it.
What to do with this
If your company’s learning budget is concentrated in licences and platforms, there is no need to throw it out of the window. What is needed is to stop believing it is enough.
Outwards: attend these spaces, and do not just delegate attendance; send your team, including the profiles who in theory do not need it; support these initiatives with visibility, space or sponsorship; and if nothing like this exists in your city, seriously consider setting one up, because the investment is negligible compared with any software line item.
Inwards: create your own cross-disciplinary spaces around real problems, not generic presentations about innovation; reward experiments, mistakes and improvements circulating between teams, because if every lesson stays where it was born you are paying for it several times over; and measure changes in practice, not content consumed, since someone completing a course proves little and someone making better decisions proves everything.
AI will make access to knowledge even cheaper, and that is good news. But let us not confuse access with learning, or learning with judgement. Having a model available to everyone does not make a company smarter; what does is its ability to turn answers into better decisions. And that, for now, still happens above all when someone stands up in front of a room, tells what went wrong for them and someone else asks what nobody dared to ask. That knowledge does not travel over fibre optics. You have to go and get it.
If this kind of conversation interests you, we carry on over on LinkedIn.
Sources
- Alfred Marshall, Principles of Economics, 1890 — industrial districts and agglomeration economies.
- Yang, Holtz et al., The effects of remote work on collaboration among information workers, Nature Human Behaviour, 2021 — study of 61,000 Microsoft employees.
- Nonaka and Takeuchi, The Knowledge-Creating Company, 1995 — explicit vs. tacit knowledge, the SECI model.
- IAHub Barcelona — talks, masterclasses, La Lista and the video library.