AKI · ENTERPRISE KNOWLEDGE
The missing talent is the company’s domain knowledge
To bring AI into production, technical skills are needed, but to make it useful, I believe company’s domain knowledge must become part of the organization’s heritage.
Part One · The talent required to deploy AI into the company
Anthropic is investing one hundred million dollars to train ten thousand engineers to deploy models into production. Meanwhile, McKinsey reports that 89% of companies adopt AI, but only 6% derive significant value from it.
Though these are two different pieces of news, I tried reading them side by side, and I believe they converge toward the same skill to develop: the value of enterprise AI grows when the knowledge that makes it useful within a specific organization becomes explicit, transmissible, and preserved. Simply put, when know-how is collected in a shared, understandable form and kept up to date over time.
Per realizzare questa trasformazione non basta adottare un modello di AI; abbiamo bisogno di professionalità distinte: chi conosce il modello e il suo impiego; chi possiede il significato di ciò che l’impresa decide ogni giorno.
What evidence is needed to make a decision? Who can approve an exception? In which cases is a specialist’s judgement required?
It is in questions like these that the company’s knowledge takes shape, and we need to ensure that it becomes a shared asset of the organization, rather than remaining confined to the memory of a few individuals.
Part Two · The signal to recognise
“When we train people in AI, are we training them to use a model, or also to make what our company knows reusable?”
On October 2nd, Anthropic announced the Claude Frontier Academy: an initial investment of one hundred million dollars to train ten thousand Frontier Deployed Engineer by the end of 2027. According to the announcement, the programme includes an intensive in-person course based on a simulated enterprise deployment, from choosing the use case to security review and handover, followed by twelve weeks in which each engineer leads a real project in their own organization.
Reading this announcement, I think the stated reason deserves particular attention. Anthropic said that talent is one of the most urgent issues in AI implementation, adding that it is difficult to find people able to make AI work inside a real company and that a small group of highly skilled people can produce a disproportionate share of the value. I find even more interesting the way it summarised to CNBC what customers are asking for: people who combine business context with a high level of AI proficiency.
Part Three · The second news, read alongside the first one
The day before Anthropic’s announcement, David Pralong, senior partner at McKinsey, published an analysis in Fortune of the firm’s latest research. It revealed that organizations using AI in at least one function increased from 88% to 89% in a year, but the number of companies attributing at least 5% of their operating results to AI remained stable at 6%.
Individual productivity is obviously increasing: 15% more cases resolved per 15% more cases resolved per hour in a sample of approximately 5,200 customer service representative, and around 26% more tasks completed by developers equipped with a coding assistant. Despite this progress, Pralong notes that enterprise value does not follow the same trend.
I also find instructive the example Pralong chose: a bank uses AI to read small business loan documents, saving time in the review process. However, the benefit to the customer remains limited if the process does not clarify what evidence is needed to approve a solid loan, who has the authority to decide, and which exceptions require a specialist’s review.
We can conclude that speeding up reading is one thing, but redesigning the process is another: it means clarifying decisions, responsibilities and next steps.
Pralong encourages us to look at how work is organized, although this does not mean that redesign alone guarantees value in every context.
These two news items, read together, tell us about the transition we are living through: on one hand, there is a shortage of people who know how to bring the model into production, technicians and AI engineers able to implement the best infrastructure to support business processes. On the other hand, they show us that, when the model is adopted, its value also depends on how the company organizes work and makes evidence, authority, and exceptions explicit.
Two different skills are therefore needed.
If the first can be trained (the aforementioned Frontier Deployed Engineers), the second (company-specific knowledge) is not a ready-made package to transfer. It must be brought to light with the contribution of those who know the context and with the support of the necessary technologies.

Part Four · From digital to cognitive
As we have read, an engineer can follow training programmes to learn how to select a use case, design an agent, handle security reviews, and deliver a working system. This is a valuable and, in part, scalable skill.
But knowing “the reason why behind things” offers a completely different reading of the events that happen in a company.
What evidence makes a loan solid in that bank? Who can approve an exception in that plant? What exception distinguishes a strategic client from an important one?
This knowledge is not standard: it lives in the minds of the most experienced people, partly in systems, and partly in unwritten rules. It cannot be transferred simply through a course. It must be collected, made explicit, tested, validated, and maintained.
Those who deploy the model can be trained technically.
Those who keep the company knowledge heritage must be listened to, and what they know must become an asset of the organization: an explicit, verifiable, and adoptable resource for any AI model in use.
It is in this Human + AI convergence that the epochal shift occurs, because it is no longer enough to make data available through digitized processes. To use it in daily decisions, it must be traced back to the specific domain and clarify what it means for the company.
In other words, knowledge has to be put into action.

My reading is that Pralong’s example describes precisely this shift, from digital to cognitive. Evidence, levels of authority and exceptions are part of the meaning of the credit domain. For an AI agent to apply them correctly, they must be represented and retained.
Part Five · Understanding where to start, what remains in the company and how to measure results
“How much of the knowledge that makes AI useful in our company is explicit, verifiable and reusable today, even when people or tools change?”
Answering this question precisely is likely complex, and it was for us at Altea Federation as well. But as Newton said, “If I have seen further, it is by standing on the shoulders of giants.” This reminds us that innovation happens in stages, building on solid foundations. Consequently, we should not think of cognitivizing the entire company in a single project. However, we do need a holistic vision of business processes to identify the most strategic area to start from.
The smartest approach is to start with a well-defined domain, where we observe decisions that recur daily and where responsibilities are already clear and applied.
Some conditions can guide us along the path from digitization to cognitivization.
Part Six · Training people, preserving knowledge
I believe that ten thousand engineers trained by 2027 is a good news for the market: they will have followed a programme to bring models into organizations’ systems and processes. However, for these skills to translate into value, I also believe it is necessary to make the meaning of what the company knows explicit and create a governed context in which people and AI can work together.
This transformation within a company develops different skills and professions, following parallel tracks.
The first skill is the race toward upskilling people, with education and AI usage paths. The second skill, however, must be built more deeply, starting from the core of company knowledge: it is a capability that grows with those who possess distinctive knowledge, proceeding one domain at a time, even by introducing new professional figures.
New roles are indeed emerging: in a sense, we might call them “the guardians of knowledge”.
The result is knowledge that does not depend on a single person or a single model in use. It is knowledge made explicit, governable and reusable, even when tools and key people change.
This is the missing talent for companies: transforming their own knowledge into the most enduring strategic asset and making it a heritage of the entire organization, thanks to a cognitive infrastructure like AKI, which makes it usable, protected, and governed over time.
At the next AI training meeting, let’s try adding a second question alongside “How many people know how to use the model?”: “How much of our knowledge is currently in a form that a model—and an auditor—can read?”
Because making that knowledge explicit is an important step towards reducing the gap between the 89% of organizations that use AI in at least one function and the 6% of companies that attribute at least 5% of their operating result to it.
Andrea Ruscica
Chairman & CEO, Altea Federation
FAQ – Frequently Asked Questions
Bibliographic Note
SOURCES FOR THIS ARTICLE
- Anthropic, 2 October 2026. Claude Frontier Academy: $100M to train 10,000 engineers.
- CNBC, 2 October 2026. Anthropic to invest $100 million to train AI engineer talent.
- David Pralong · Fortune, 1 October 2026. McKinsey senior partner: Why AI’s easiest wins are misleading CEOs
The details of Anthropic’s programme are those announced by the company. The figures on adoption and productivity are reported by Pralong in Fortune, based on McKinsey research and the studies cited in his commentary; they are not Italian data or AKI results. The banking example is illustrative. The link between the two news stories and enterprise knowledge is the author’s interpretation.
