Why doesn’t AI use spread across the organization?

The most common reason is that the skills are left to a couple of enthusiasts. An AI capability that lives in a few people’s heads is not a capability — it is a key-person risk. Real capability is built into roles, processes, and onboarding across the entire organization.

Typically, AI gets adopted fastest wherever someone gets excited. That is a good start and a poor destination. When the capability rests on a few enthusiasts, the benefits inevitably remain modest. At the same time, the organization has built its capability on single individuals rather than as an asset of the whole team.

In our experience, one of the most common reasons AI use fails to spread is that the skills involved are so often thought of as a technical matter — which, these days, they rarely are. The “I’m not a technical person” mindset seems to be one of the biggest barriers to putting technology to use. The truth, however, is that most modern tools no longer require technical skills to get started. They require the courage to experiment. Modern applications are built so that anyone can get going easily — without coding skills or an IT background.

The need for technical skills shrinks by the day. What matter instead are process thinking, requirements definition, a grasp of the technology’s possibilities and risks, and business understanding. None of these is really a technical matter, but all of them take practice — just like anything else.

We have seen the shrinking need for technical skills in real-life examples. Ordinary office workers (non-technical ones) are creating things with technology that have quite significant business impact — things they would not have been capable of at all just a year ago. Things that a couple of years ago would have been bought from outside vendors at great expense.

When it comes to AI, skills must not be outsourced to a handful of individuals. The ability to use AI is essential for absolutely everyone. And when people inside an organization feel empowered and get excited about their own abilities, it rubs off on others too.

You can gauge your organization’s potential with the following question: what could our organization achieve if everyone were as capable with AI as our most advanced users?

Why do AI skills polarize?

In Astu Labs’ client work we see the same distribution that a 2025 company survey by the Finnish Institute of Occupational Health and Statistics Finland confirms: AI spreads unevenly through organizations. More than half (51%) of Finnish companies with at least ten employees already use AI, but the use is polarized. The enthusiasts race far ahead, the rest stay put, and no one is leading the effort to close the gap.

On the other hand, it is worth asking whether organization-wide change and capability were ever even the goal. The MIT Project NANDA report (2025) finds that only about 40 percent of companies have an official AI tool, even though in over 90 percent of companies employees use AI tools regularly on their own.

This leads to a claim that is easy to skip past but expensive to ignore: a capability that lives in a couple of people’s heads is not a capability. It is a key-person risk. A capability becomes real only when it is built into roles, processes, onboarding, and shared ways of working. That building has a name: institutionalization.

And here AI is a threat and an instrument at the same time. A threat, because it accelerates the polarization of skills faster than any earlier tool. An instrument, because few earlier technologies have been able to do what AI can: it gets the knowledge and understanding accumulated over years in memos, reports, and systems out of the silos and searchable for the whole organization. In an expert firm, where value has always rested on the expertise of a few senior people, this is an exceptionally big promise. AI can make accumulated understanding searchable, shareable, and teachable.

What expertise can be transferred to AI — and what stays with people?

In 1966, the philosopher Michael Polanyi formulated an idea that matters more than ever in the age of AI: we know more than we can tell. The most valuable expertise — why an experienced expert makes the right call in a second, relying on intuition — can never be fully written into a memo. From this follow the two levels of institutionalization. Explicit knowledge — documented but siloed — is exactly what AI can make searchable and shareable. At the same time, it reduces the risk that materializes when an expert leaves and takes with them what no one else knew. But the deepest tacit knowledge stays in people. And the more explicit knowledge gets automated, the more valuable the tacit kind becomes. AI does not diminish the importance of people — it raises it.

The same caveat shows up in the experimental evidence. When Dell’Acqua and colleagues gave hundreds of consultants access to AI in a field experiment (2023; peer-reviewed in Organization Science, 2026), performance clearly improved on tasks that fell within AI’s zone of capability, and deteriorated on tasks that looked similar but fell outside it. People trusted convincing but flawed output. The decisive skill was not using the AI but judgment — knowing when the output can be trusted and when it cannot. That is a skill that does not come included in the price of the license. It is a skill that is built into people.

The good news is that polarization is not destiny. Brynjolfsson et al. showed (QJE 2025) that AI raised productivity by about 15 percent on average — and by about a third for the least experienced. The gap can be closed, then, if skills are led to spread rather than allowed to pile up in a few hands.

“But they’re the ones driving the whole thing here”

But isn’t it a good thing that precisely those couple of enthusiasts are driving our adoption forward? Of course — ignition takes enthusiasm. But ignition is not a structure. When the capability rests on them, two things follow inevitably. First, a person is not a system — a person cannot be documented, backed up, or scaled. Second, you have outsourced the productivity gains to individual people’s employment contracts and widened the internal learning gap.

Three questions worth asking right away

We encourage leadership teams to run one exercise at their very next meeting:

1. If your two most advanced AI users left, what capability would walk out the door? And is any of it written down — in a process or in onboarding? If the answer is “a lot, and hardly anything,” you have a risk, not a capability.

2. Whose job is it to spread AI skills from the few to the many? Name a person and a mechanism. If there is no name, the job is not getting done.

3. What understanding currently lives only in your senior experts’ heads and in old memos — and what part of it could AI make available to everyone as early as this year? That is the first concrete step of institutionalization.

Institutionalizing skills is one of four reasons why every strategy question is ultimately an AI question. Alongside it sit workflows, cost structure, and governance. And beneath it lies the same foundation this whole page rests on: sustainable advantage is not in the tool but in people, and in how fast they learn.

Sources

• Immonen, J., Alasoini, T., Siltala, V., Lukander, K., Toivanen, M., Valtonen, T. & Varje, P. (2026). Tekoälyn hyödyntäminen yrityksissä 2025: Tuloksia Digivihreä siirtymä ja työ -yrityskyselystä. Työterveyslaitos (Finnish Institute of Occupational Health). https://www.julkari.fi/items/9818b4ac-8c2b-4031-be73-c0d8e92221d5 (Survey conducted March–July 2025, n = 1,691, response rate 38.0%. Figures: 51% use AI; a written AI strategy at 11% of all companies / 17% of AI-using companies; 49% of GenAI-using companies have trained staff; GenAI use by staff group in GenAI-using companies: top management 93%, experts 85%, operational roles 18%.) 

• Challapally, A., Pease, C., Raskar, R. & Chari, P. (2025). The GenAI Divide: State of AI in Business 2025. MIT NANDA, MIT Media Lab, July 2025. (Shadow use: an official LLM subscription at ~40% of companies, while in over 90% of companies employees use AI tools regularly on their own.) 

• Dell’Acqua, F., McFowland III, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F. & Lakhani, K. R. (2023/2026). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality. HBS Working Paper 24-013; peer-reviewed in Organization Science (2026), DOI 10.1287/orsc.2025.21838. 

• Polanyi, M. (1966). The Tacit Dimension. (“We can know more than we can tell.”) 

• Brynjolfsson, E., Li, D. & Raymond, L. (2025). Generative AI at Work. Quarterly Journal of Economics 140(2), 889–942.