Artificial intelligence is presented as a performance lever, but as announcements follow one after another, one question is rarely addressed: what really works today, without any particular effort, in organizations that are under pressure and already saturated with transformation projects?
Here, we are taking the opposite approach to “AI First” discourse and structural transformations, and are trying to look at the bottom of the ladder, at what works naturally and can bring something to employees on an individual level, but without having any real impact on the collective or the business other than a marginal one (Collective appropriation of AI: the only condition for tangible impact).
What is the point? To know how to build even a minimalist foundation for adoption, so that we can then build on it, but also, because we talk a lot about trains that arrive late, to shine a little light on those that arrive on time.
In short:
- The AI use cases that work today are those that integrate naturally into existing practices, without requiring organizational transformation or significant adoption efforts.
- These uses are mostly individual, linked to local and repetitive tasks, and do not redefine collective roles or processes.
- Their effectiveness is based on their gradual “invisibility”: they cease to be perceived as AI, become integrated into everyday tools, and are sometimes used informally or hidden.
- Adoption is based on empirical practice, without ideological discourse, which allows users to develop pragmatic confidence in these tools.
- These uses do not constitute a model for transformation but rather an indicator of what organizations can absorb without resistance, thus laying a minimal foundation for wider adoption.
When AI is not an issue
When we look at how artificial intelligence is used in business after several years of discussion and experimentation, we see that while much of what is launched disappears without leaving a lasting mark for various reasons, some uses persist. They are not necessarily identified as strategic, but they endure.
What works today rarely corresponds to what was most heavily promoted. Uses that have found a place that is sufficiently compatible with existing operations and have required minimal change, to the point that they are no longer even noticed or mentioned.
Priority to hyperlocal optimization
Usages that endure without any particular effort almost always fit into processes that are already in place. AI does not appear as an organizing principle or an end in itself, but as a marginal addition, at a moment or in a tedious, repetitive task, most often at the user level, not the collective level. It does not lead to an immediate redefinition of roles or a redistribution of responsibilities.
It is precisely this absence of disruption that allows the use to take hold. There is no need to believe in it or defend a vision. The work does not change, there is no redesign, but it simply becomes a little less restrictive. This limited but concrete gain is enough to explain why the use persists where more ambitious systems quickly run out of steam.
When usage becomes unconscious
A recurring feature of these uses that are working today is their gradual disappearance from discourse. They are no longer presented as AI and have become just another mechanism, integrated into the working habits of some. There are two reasons for this invisibility.
The first is that an initiative, even a modest one, has become standard practice.
The second, let’s not forget, is “shadow AI” and those employees who don’t want to show that they are using AI!
There is also the evolution of tools with AI that are no longer used intentionally but are integrated into everyday tools. We are moving from the “art of prompting” to the use of a feature that breaks down a barrier to adoption and makes us forget that we are using a new technology (Nobody wants to prompt).
It is often said that technology is a word that describes something that doesn’t work yet but we could just as easily say that it describes something that requires effort to use.
We no longer talk about what works, only what doesn’t work or doesn’t work yet.
Practice governs the relationship between humans and AI
The uses that persist are not accompanied by grand speeches about the place of humans, and this may be why they are not frightening. Things are handled pragmatically, and teams know when to follow a recommendation, when to deviate from it, when to use AI, and when to take control. This understanding is empirical, often imperfect, and not always formalized, but it comes from experience and therefore inspires confidence rather than fear.
When this confidence based on experience exists, AI does not become a sensitive issue. It is used and discussed, but rarely rejected. Conversely, where this confidence does not exist, its use remains fragile, even when the technology is considered relevant.
Lessons learned from local use cases
What works today with AI does not provide a model to replicate or a path to follow. It does not provide any methods or trigger any transformation, but simply shows what organizations accept without any extra effort.
It tells us what the business naturally tolerates in its day-to-day operations. It is imperfect, the gains are limited, but it is a basis for writing the next chapter of the story.
Bottom Line
What works today are practices that have become established without any additional effort or transformation or change to the existing system. They do not carry any ambition, but tell us what we can do at a minimum.
In this sense, they are more of a revelation, as they show the forms of automation, recommendation, or assistance that the organization is capable of absorbing without putting itself under strain.
To answer your questions…
What works is based on marginal uses, integrated into the existing system, without any particular transformation or effort. These are one-off aids at the individual level, for repetitive or tedious tasks. These uses are not strategic, but they bring immediate and concrete benefits. Their strength lies in the absence of disruption: they do not change the organization or roles, which allows them to become established in the long term where more ambitious initiatives fail.
The most visible projects are generally highly ambitious and require profound changes. In organizations already under pressure, these additional disruptions create fatigue and resistance. Conversely, practices that endure are compatible with existing practices and require neither ideological engagement nor reorganization. This discrepancy explains why what is most prominent is not what has the most lasting impact.
Ultra-local optimization allows AI to act as a simple tool for lightening the workload. It acts at a specific point in the process without calling the whole thing into question. This approach reduces friction, avoids debate, and makes its use acceptable without conscious effort. The benefit is limited but tangible, which is enough to anchor the practice in everyday life, without the need for a global vision or motivational discourse.
When a use case works, it ceases to be perceived as AI and becomes a habit. This invisibility also stems from “shadow AI”, when employees prefer not to reveal their practices. Integrating AI directly into everyday tools also plays a role: the effort disappears, particularly that of prompting. The less the use is prescribed, the more natural and sustainable it becomes.
These uses do not outline a method or model, but reveal what the organization accepts without creating tension. The gains are modest but real, and the experience creates pragmatic trust between humans and tools. This minimal foundation shows the forms of assistance or automation that can be absorbed naturally. For a decision-maker, it is a starting point for gradual development, without forcing a transformation that is disconnected from actual practices.
Image credit: Image generated by artificial intelligence via ChatGPT (OpenAI)
Related articles:
AI First: lots of debate and few answers
Adoption and impact of AI: lessons (and limitations) from the latest McKinsey and BCG studies
Who is handling your artificial intelligence projects? Probably not the right people.
How can you prepare your organization for AI?
AI adoption does not replace productive appropriation
The AI-first company: the origins of an ambiguous concept that grew too quickly
