AI in the Enterprise: Between Freedom to Act and the Need to Stay on Course

Over the past few weeks, I’ve had several discussions about AI in the workplace.

With CIOs looking to protect data and avoid slip-ups. With business leaders who primarily want to solve concrete problems. With teams wondering how to save time without adding yet another complicated tool to their daily routine.

At first glance, these conversations didn't have much in common.

Yet, they always came back to the same point:

How do we move forward with AI without losing control, but also without preventing teams from experimenting?

The question is far from theoretical.

It touches on security, of course. But also on trust, on the way we work, and on everyone's role within the company.

Something is Changing

For a long time, when a team needed a new tool, the path was quite clear.

The business unit expressed a need. The IT department (DSI) studied it. Developers designed the solution. Then came testing, security, and production deployment.

This way of operating still makes perfect sense for certain projects.

But it is no longer the only way.

Today, a human resources professional, a salesperson, a buyer, or a project manager can create an assistant, automate a task, or design an initial prototype without knowing how to code.

In just a few hours, they can sometimes obtain something usable.

This changes a lot of things.

Employees are no longer just users of digital tools. They can also participate in their creation and improvement.

Just as we all learned to use spreadsheets, we will likely have to learn to work with AI.

Not to replace our judgment. But to better prepare work, speed up research, simplify a task, or explore an idea.

The CIO's Concerns are Legitimate

In this context, I perfectly understand why CIOs want to set rules.

We are talking about sensitive data, security, compliance, and decisions that can have real-world consequences.

The risks are not imaginary.

An employee might use an unvalidated tool and upload confidential information. A team might create an application without assessing its vulnerabilities. A decision might be made based on a response produced by an AI, without sufficient verification.

And then there is the risk of seeing a proliferation of tools, subscriptions, and experiments that are impossible to maintain over time.

A company cannot simply say: "Test everything, we'll see what happens."

The IT department needs to protect the organization. That is its role.

But protecting doesn't necessarily mean prohibiting.

The Risk of Wanting to Centralize Everything

Conversely, if every initiative must go through a long validation circuit, teams end up waiting.

Then they bypass.

This phenomenon is not new. We have already experienced it with Shadow IT.

AI can simply accelerate it, because the tools are much more accessible and results can be obtained very quickly.

An employee who finds a solution in two hours to a problem that has been dragging on for months will naturally want to continue using it.

If the only response they get is a series of prohibitions, they won't necessarily give up. They might look for another way to do it, without talking about it.

This is where the subject becomes delicate.

The most interesting needs are not always visible from a central management level. They are often found in the details of daily life: a repetitive task, information that is hard to find, a process that wastes time for an entire team.

The people who encounter these problems every day are often the best placed to imagine a solution.

The Issue is Not Choosing Between IT and Business Units

I believe the debate is sometimes framed in a way that is too binary.

As if we had to choose between two options:

either IT maintains control, or business units are free to do whatever they want.

In reality, neither of these two solutions truly works.

The first slows everything down and ends up discouraging teams.

The second exposes the company to risks it cannot control.

The real question is rather:

How do we create a framework clear enough so that teams can experiment without putting the organization at risk?

It is a matter of trust.

Employees need to know which tools to use, what data they can share, what verifications are necessary, and who to contact when they have a doubt.

A good framework should not be experienced as a series of prohibitions. It should allow for moving forward with greater peace of mind.

The Role of the CIO is Evolving

The IT department will likely no longer be the only one building the company's digital solutions.

Its role will consist more of creating the conditions so that others can build correctly.

This means offering reliable tools, protecting data, defining understandable rules, training teams, and supporting experiments.

The CIO becomes less of a gate that must be passed through and more of a partner to move forward with.

This also requires a change in posture from the business units.

Autonomy does not mean that everyone can do anything. It implies understanding the rules, assessing the risks, and agreeing to share what is created.

And What About Managers?

We talk a lot about the CIO and the employees. We sometimes forget an essential player: the manager.

Yet, it is often the manager who can make the difference between an approach that remains theoretical and one that truly takes root within teams.

A manager does not necessarily need to become an AI specialist.

However, they must take enough interest to understand what it can offer, what it cannot do, and the precautions to be taken.

Above all, they must lead by example.

A manager who asks their team to use AI while never using it themselves risks losing credibility. Conversely, a manager who tests, asks questions, shares their learnings, and is willing to rethink certain habits creates a much more tangible dynamic.

Their role is also to start from the team's real needs.

It is not about using AI just because everyone is talking about it. It is about looking at the work as it is actually performed and asking where AI can help.

Which task is a waste of time? Which information is difficult to find? Which process could be simplified? Where does the team need more support?

The manager must then help bridge the gap with the organization's priorities.

An experiment might be interesting for one team but may not address a major challenge for the company. Conversely, a simple idea can have significant value if it solves a problem shared by several teams.

The manager is therefore a facilitator, a guide, and a point of connection between the field and the organization.

The balance will never be perfect

The companies that succeed will likely not be those that let everyone do whatever they want.

But they won't be those that try to lock everything down either.

They will be capable of leaving room for initiative while setting clear boundaries.

This balance will sometimes be difficult to find.

It will require testing. Making mistakes. Correcting. Stopping certain uses. Encouraging others.

It will also require accepting that rules evolve alongside the tools and the maturity of the teams.

The important thing is not to wait until you have answers to every question before starting. But it is equally important not to start without having considered the consequences.

A transformation that is human above all

At Pivotal Skills AI, this balance zone is what interests us.

We often see that difficulties are not solely related to technology.

They also concern trust, skills, work habits, and the ability to make business units, managers, and the IT department work together.

How do you make teams want to try? How do you prevent the fear of doing things wrong from blocking initiatives? How do you help managers support their employees? How do you ensure that use cases meet real needs and remain aligned with the organization's objectives?

In my view, a successful AI transformation relies on several elements that must progress together:

a clear framework, adapted skills, involved managers, and shared trust.

AI will not sustainably transform a company simply because a new tool has been deployed.

It will do so when people have understood how to use it, why to use it, and within what limits.

That is undoubtedly where the true challenge of AI in business lies.

And perhaps also its greatest opportunity.