AI use grows but risks remain

A phase of discovery and experimentation with AI has led to identified risks such as hallucinations, biases, and errors. These issues highlight the need for caution as AI becomes more integrated into daily tasks.

Employees report benefits from AI

According to a survey by the Greenworking Observatory and Ipsos bva in 2026, 73% of employees report using generative AI at work, with 75% stating they are faster and 70% believing they produce better quality work. These findings suggest a widespread perception of AI's positive impact on productivity and output.

Perceived improvements may not translate to real gains

However, the question remains of what the organization actually gains when individuals perceive improvements in speed and quality. While AI models are becoming more coherent, fluid, and convincing, this may make it harder to challenge responses that are based on fragile reasoning.

AI may reinforce a false sense of control

A better-written response may not necessarily be more accurate, and a more convincing recommendation may not be more solid. This can reinforce a sense of control without actual progress in reasoning. The risk is not only AI "hallucinating" but also providing answers too quickly.

Challenges in verifying AI outputs

As production becomes easier, distinguishing what is truly relevant from what only appears to be becomes more difficult. Individual gains may not automatically translate into collective performance, and what is gained at the production level may not disappear from the work, potentially affecting the assessment of results, decision-making, and responsibility for consequences.

Managers face new challenges

This tension is already evident in the work of managers, with 58% reporting that they must challenge or verify more the productions made with AI. Effort saved in production is being redirected into evaluation, arbitration, explanation, and responsibility for the result.

Managers deal with unclear AI guidelines

This forms a managerial debt, as companies cannot define all good uses of AI in advance due to the diversity of jobs, situations, and the rapid evolution of tools. Managers are increasingly tasked with resolving questions that the organization has not yet clarified, such as what can be delegated, what needs to be checked, how to assess work done with AI, and how to prevent gaps in mastery from widening.

Massification of AI does not equal scaling

The confusion between the massification of AI use and scaling up is frequent. While multiplying licenses, users, or use cases indicates AI's spread, it does not necessarily mean the company is gaining more value from it.

Real value depends on organizational improvement

Scaling up requires that the benefits perceived individually lead to better decision-making, cooperation, learning, or production. A company may have widely adopted AI without successfully scaling it up. The real criterion is not the number of users but what the organization is capable of doing better thanks to AI.