AI in the Workplace: 61% of Employees Demand Human Validation

Generative AI at the Heart of Professional Daily Life

The rise of generative AI in the professional world is undeniable, as demonstrated by a recent survey conducted by TeamViewer. Carried out by Sapio Research, this study surveyed 4,200 employees and managers across several countries, including France. It highlights a clear trend: while AI is widely used to assist professionals, it is rarely allowed to act autonomously without human supervision.

Daily Use but Limited Autonomy

Widespread Adoption, Restricted Trust

According to the study's results, 75% of participants use AI at least once a day, with an average of three daily interactions. The benefits of AI are widely recognized, as 97% of respondents identified at least one tangible advantage, ranging from automating repetitive tasks to reducing workload. Furthermore, 64% of those surveyed described their experience with AI as positive.

However, this adoption does not translate into a willingness to fully delegate decisions to AI. In fact, 61% of respondents prefer that AI does not make any decisions autonomously. More specifically:

• 43% want AI to provide recommendations while leaving the final decision to them.
• 18% do not want any recommendations at all.
• Only 35% are willing to allow AI to act more freely on their behalf.

Conditions for Accepted Autonomy

The refusal to let AI act alone is not a total rejection. 70% of participants feel comfortable with AI making decisions, provided they can intervene at any time. In contrast, only 11% would accept AI operating without any possible human intervention. Thus, it is not autonomy itself that poses a problem, but rather the lack of visibility regarding this autonomy.

Necessary Safeguards According to Employees

For AI's autonomous actions to be more acceptable, respondents expressed several conditions:

• Enhanced security and privacy measures (39%).
• Prior notification of any major changes (36%).
• Clear limits on AI's access (36%).
• A visible history of actions taken by AI (33%).
• An option to cancel or revert AI's actions (31%).

This conditional trust is directly linked to the perceived level of risk. 40% of respondents believe that human intervention is necessary whenever security, data, or business operations are at stake. Conversely, a very small minority would accept a fully independent AI. However, monitoring AI comes at a cost: participants spend an average of two hours per week on this, and 51% admit they do not always know when to trust AI or when to control it.

For Kai Werner, HR Director at TeamViewer, the key lies less in technical performance than in governance: “Trust develops when employees understand how decisions are made, have a say in the areas where autonomy is introduced, and know they will not be held accountable for outcomes they did not control.”

Perception Differences Between Managers and Employees

The study also reveals a gap between the perceptions of different categories of staff. Managers are significantly more enthusiastic than employees. 71% of them describe a positive experience with AI, compared to 58% of employees, and they are more inclined to let AI operate in the background. Employees, on the other hand, feel more exposed to the potential errors of an autonomous system, especially since they rarely have a say in its selection or configuration.

This divide is corroborated by other recent research. A study by Notion highlighted a similar gap, with 60% of executives deeming their organization ready for AI agents, compared to only 36% of employees. Our annual survey also showed a real but highly uneven adoption of AI across different professions. Three distinct methodologies converge on the same conclusion: those who decide on the deployment of AI do not assess its risks in the same way as those who use it daily.