AI skills: integrating artificial intelligence without creating organisational debt

AI use is expanding faster than skill development, creating a gap between what tools can do and what organisations can actually use effectively.

Unstructured AI adoption creates organisational debt: tool stacking, fragmented practices, lack of consistency.

Managing AI skills requires clear governance: who uses AI, for what, and with what level of autonomy.

The challenge is not just producing AI outputs, but interpreting them, understanding their limits, and taking responsibility.

Sustainable AI integration depends on alignment with existing skill frameworks, decision-making processes, and ways of working.

Artificial intelligence is gradually establishing itself in organisations, but its adoption reveals a growing imbalance between uses and the AI skills that are actually mastered. Faced with this reality, the question is no longer whether AI should be integrated, but how to do so without creating organisational debt.

Multiplying tools, experimenting without governance, or separating AI from existing practices exposes projects to losing momentum. Conversely, an approach based on critical skills and a progressive implementation of know-how allows AI to be anchored over time.

According to a McKinsey survey, 88% of companies worldwide report using AI technology. This acceleration in use is not matched by structured skill development. 

This growth in usage is not always accompanied by an equivalent effort in terms of training and skills management. The 2024 Work Trend Index Annual Report by Microsoft shows that 75% of global workforce use AI in their daily work while 39% report having received no formal training

In other words, AI is being integrated into daily practices faster than upskilling systems are being developed. This gap creates a paradoxical situation:

• On one hand, organisations experiment with and deploy AI solutions, often under pressure for performance or innovation.
• On the other hand, the skills required to understand and leverage these uses remain only partially structured.

When artificial intelligence is implemented without a clear framework, it generates organisational debt comparable to technical debt: quick choices that produce delayed effects that are difficult to correct.

The first source of organisational debt lies in the multiplication of AI solutions introduced over the course of projects. Each team tests its own tools and explores new functionalities without cross-functional coordination. This logic of continuous experimentation fragments practices and makes overall visibility more complex. It becomes difficult to identify who is doing what, with which reference points, and according to which rules of use.

Over time, this accumulation blurs the understanding of the skills that are actually being mobilised. Organisations have powerful tools but struggle to capitalise on the learning generated.

Organisational debt also appears through dependence on a limited number of experts, often called upon to interpret results or secure usage. This concentration of knowledge creates an imbalance. Under these conditions, responsibility becomes unclear. Who validates a recommendation produced by an automated system? Who arbitrates in the event of an error? AI becomes embedded in processes without real organisational ownership.

The most common mistake in AI integration projects is to treat AI skills as an autonomous block. This approach often leads to juxtaposing tools or roles without clear articulation with the existing organisation.

AI-related skills do not replace business expertise or existing capabilities. They build upon them. Understanding a model or interpreting an automated recommendation first relies on a detailed knowledge of processes and operational constraints. Without this foundation, AI remains an isolated system with limited long-term usability.

A controlled integration of AI skills therefore requires starting from what already exists:

• Skills frameworks;
• Ways of collaboration;
• Decision-making processes.

It is within this perspective that structuring tools such as the skills matrix take on their full meaning. It provides a cross-functional view of already mastered skills and facilitates the identification of leverage points to integrate AI without disruption and adapt training pathways.

Behind the term AI lie different approaches, which do not mobilise the same skills. Clarifying these distinctions helps avoid a double pitfall: overestimating the technical dimension or, conversely, underestimating the human skills required for actual AI use.

• Machine learning technologies rely on analysing large volumes of data to produce predictions or automate certain classifications. They are already widely used in predictive analytics and decision support. These systems most often rely on neural networks, capable of identifying complex patterns by progressively learning from input data.
• Generative AI, for its part, produces content or recommendations based on existing data. The user does not simply read a result; they test new functionalities, formulate hypotheses, and adjust their requests.

Within an HR function, a machine learning model can identify trends in skills evolution based on historical data. AI assistants, for their part, can go further by acting as AI-powered skill generators. They are able to propose personalised upskilling pathways or suggest training adapted to a given profile.

In both cases, the critical skill remains the ability to interpret results, not to develop the model.

Regardless of the technology used, expected AI skills go far beyond tool mastery. They rely on capabilities already present in the organisation but evolving.

These include:

• Understanding what an automated system does, and what it does not do;
• Identifying biases or limitations in a generated result;
• Placing a recommendation within its business and regulatory context;
• Deciding when to follow, adjust, or discard a proposal generated by AI.

A manager using a recommendation from a generative AI system to anticipate skills needs is exercising a judgment capability that already exists, enriched by a larger volume of information. AI changes scale and speed, not responsibility.

At organisational scale, in an AI skills project, the challenge is not only to identify use cases, but to define who mobilises which skills, at what moment, and with what level of responsibility. Without this structure, AI is adopted opportunistically, depending on local initiatives.

• The first condition for effective management lies in governance. Integrating AI into practices requires clarifying existing roles rather than systematically creating new ones. HR, training, business, and IT functions retain their respective responsibilities but must share a common foundation of AI uses.
• Managing AI skills in a project also implies rethinking decision-making processes. AI enables the production of real-time or large-scale analysis, but these capabilities only create value if decision-makers have the skills required to use them.

Finally, managing an AI skills project requires aligning project initiatives with a clear and structured upskilling trajectory.