AI is everywhere, at least in discourse and intentions. Sometimes it is invisibly integrated into our everyday tools, sometimes it takes the form of new tools that we try to shoehorn into our daily routines, but it is there. And yet, when we talk to practitioners, we have to admit that nothing is really changing.
As is often the case, promises of transformation are slow to materialize, productivity gains are not so obvious, and, on top of that, employees are worried about the future that awaits them.
And yet this is hardly surprising, since once again we are confusing the adoption of a tool with the transformation of an organization (The limits of technology-driven transformation).
All this is confirmed by the BCG report AI at Work 2025. As is often the case when faced with a technological wave that is supposed to sweep everything away in its path, we discover a much more contrasting reality where AI generates uses but finds it more difficult to generate value.
And since this is a scenario that plays out with every technological revolution, we realize that the obstacle is no longer AI itself, but the way businesses are structured to (not) take advantage of it.
In short:
- The widespread adoption of AI in business has not been accompanied by real organizational transformation, as it is often integrated as a simple tool without questioning existing processes.
- The productivity gains announced are limited by the lack of mechanisms for organizational reappropriation of the time freed up, which dilutes the value created instead of concentrating it within the business.
- Current AI training is insufficient and inadequate: it must evolve towards an approach focused on reasoning, exploration, and interaction with non-deterministic systems.
- Businesses that rethink their workflows (and not just their tools) see significantly greater benefits in terms of efficiency, strategic reallocation, and decision quality.
- The lack of human and managerial support around AI generates mistrust and insecurity among employees, highlighting the importance of clear leadership, contextualizing management, and transparent communication.
Adoption does not mean transformation
72% of employees surveyed say they use generative AI regularly (several times a week), compared to 56% in 2024. Its use is therefore widespread and growing rapidly. Better still, at the risk of upsetting the naysayers, France (64%) now surpasses the United States (51%), a far cry from the usual refrain about our digital lag. However, India (92%) and the Middle East (78%) are still way ahead.
But this widespread adoption masks a flaw: AI is mainly used as a one-off aid, integrated into existing tools (Copilot, ChatGPT, etc.), without questioning work processes and organization.
While 72% of businesses say they are deploying AI (Deploy), only 50% of businesses say they have started redesigning their workflows (Reshape), and 22% are inventing new models (Invent).
Note that I expected much more worrying figures than this.
In any case, adopting AI is not enough because what matters is what the organization does with it.
AI is therefore increasingly present in businesses, but without being transformative. It is not a question of usage, and the figures are there to prove it, but a question of design, of “work design”. In other words, it is not the usage that is at issue, but what we are trying to do with AI and what we want to achieve through it.
We deployed Copilot, activated the CRM’s intelligent agent, added a “generate” button to the intranet, but at no point did we ask ourselves what this should change in terms of work and how we create value. AI is often treated as a technical building block, not as a lever for organizational alignment and simplification.
Where have the productivity gains gone?
47% of regular GenAI users say they save more than an hour a day. But only a third receive guidance on how to reallocate this time, with the rest split between secondary tasks, personal exploration, training, discussions with colleagues, or even shorter working days.
In fact:
- 52% finish their work earlier (sometimes with improved quality),
- 44% work on strategic topics,
- 36% experiment with generative AI,
- 34% train themselves,
- 29% connect with others or take time for themselves.
The problem is not the gain but the way in which the organization reappropriates it.
We seek productivity gains without being organized to absorb them. There is often no system in place to capture, redistribute, or leverage the time freed up. There is no management, no redesign, and so everyone muddles through, hoping that something tangible will come of it.
In any case, when technology actually frees up time, the business does not automatically capture the value created. Some of these gains are absorbed by the employees themselves (comfort, exploration, breathing space), diluted in low-value tasks or even used for new tasks made necessary by the use of technology. They can also be captured by customers (sometimes), service and technology providers (often) in such a way that in the end the business captures only a small portion of its ROI.
This is nothing new and is not specific to AI, but it is worth pointing out.
Several economic theories explain this phenomenon of “gain dissipation“, from the organizational rebound effect to the spillover logic (Are you familiar with Sauvy’s spillover theory on productivity gains?) and redistribution to other actors, but I will return to this in a dedicated article.
Training in AI is not the same as training in Word
Only 36% of employees feel well trained in AI. However, as soon as training exceeds five hours, in person with coaching, adoption rates skyrocket to 89%. BCG emphasizes that generative AI is non-deterministic, contextual, iterative, and therefore requires a new approach to teaching (Businesses are deterministic, generative AI is not, and that’s a real problem.).
Training is not about learning a sequence of clicks to achieve a given result, but about reasoning in ambiguous situations and dialoguing with a probabilistic tool that will not react identically to the same prompt twice in a row.
Training in AI is above all training in an attitude: the ability to formulate intentions, explain a context, explore several formulations, cross-reference results, verify them, structure reasoning, and interact with an imperfect system.
The reflex to train people in Copilot as we trained them in Excel is therefore logically a dead end. You can’t learn to use a generative tool without learning to think with it.
It’s not tools that create value, but flows
While Moderna is currently one of the most ambitious (and intriguing) examples of transformation by and for AI, the concept of flow work is central (HR and IT merger: Moderna redesigns its organization for and with AI).
And the figures put forward by BCG speak for themselves: in businesses that are reshaping their workflows (Reshape), the indicators are skyrocketing:
- Time savings of more than 1 hour/day: +26 points
- Reallocation of time to strategic tasks: +9 points
- Improved decision-making: +13 points
It is not the tool that creates value, but the organization of work and workflows.
AI will not do the explaining for you
46% of employees in businesses undergoing transformation believe that their job could disappear, compared to 34% in businesses that have only begun deployment. Spain (61%), India (48%), and Japan (40%) are among the most anxious countries.
This comes as no surprise: AI without human support and clear communication generates more confusion than confidence. I would even add that the greater the trust in leaders, the more employees will trust the way AI is deployed, the objectives pursued, and the potential impact on them (Employees Won’t Trust AI If They Don’t Trust Their Leaders).
It’s a simple organizational rule: any change that is not explained is perceived as a threat, and AI, by its very nature, reshuffles the deck in terms of value, roles, and expertise. If the business does not support this, employees will be unsettled, and sometimes rightly so.
This brings us back to the role of the manager. A good manager is not a task distributor, but a context creator (The fictional interview with Ted Lasso, the manager who manages without expertise), and with AI, this becomes even more true. It is not AI that will drive meaning, priorities, and trade-offs, but the augmented manager, not the manager who has been relieved of responsibility (The minimalist manager: a promising model, but one that needs clarification).
Bottom line
The BCG report shows how widespread AI is, how it is used and adopted, but it also reveals the scale of the organizational work still to be done.
The technology is ready, but the organization is much less so.
The promise of generative AI will not be realized in businesses that think in terms of deployment, but in those that dare to rethink the way they work, organize work, train their employees, and manage.
AI is not a miracle solution, but rather a test of maturity that forces us to clarify what was no longer being questioned: what is the purpose of work, what meaning do we give it, and how do we conceive it?
Illustration: generated by AI via ChatGPT.
Related articles:
AI in the workplace: going beyond augmentation to actually transform
Why enterprise AI can’t keep up with consumer AI: beyond ChatGPT, a more complex reality
Generative AI in the workplace: revolution or illusion?
AI heading for an economic dead end?
Automation is done with people, not against them.
Why use cases don’t work with freeform technologies
