The Employees Who Use AI Best May Resist the Company’s AI Rollout Most

On this page

1. A good deal for the company
2. Not necessarily a good deal for the employee
3. What gets handed over is time — and your own irreplaceability
4. The effort is rewarded, but what about the risk afterward?
5. How to respond is an unsolved business question

Have you ever seen someone pay to go to work? There is now a group of employees paying out of their own pocket to top up tokens so they can do their jobs.

As AI agents get easier to use, many people have moved almost their entire daily work into AI: reading files, writing reports, making summaries, setting proposals, writing code, and so on. They are itching to organize everything related to their work into skills and plug them into the system they work with every day.

This is the “distillation” that briefly became popular: taking the part of a person’s experience that can be made explicit and refining it from various documents and historical material into skills an AI agent can call. Distilling yourself, and distilling colleagues too, then handing the work to an agent to do it more efficiently in your place.

Many people do this enthusiastically, even willing to spend their own money on AI to do it.

But when it comes to the company deploying and rolling out an AI system at scale — especially asking employees to connect core business into it, distill the relevant experience and SOPs into skills, and then spread them — that is when they hesitate.

This is another problem many companies hit when adopting AI today: the people who use AI best may be the most resistant to a company-wide rollout.

A good deal for the company

To understand this contradiction, you first have to see clearly why the company makes the request.

What the company wants is an overall gain in effectiveness. Since an employee can distill themselves into skills and produce similar or better output at lower cost, those skills should not stay limited to a few people.

Can other colleagues refer to them? Can a new hire pick them up and fill in directly? Can the same piece of work carry on without a single experienced hand watching it from start to finish?

Turning one person’s ability into something the team can reuse is, for the company, a good deal.

Good for the company — but what about the individual?

Not necessarily a good deal for the employee

When an employee uses it themselves, the skill helps them finish their own work and saves their own time and energy. The company benefits too, with work done faster and more steadily.

Before AI agents, experience and SOPs were passed on too: through mentoring, writing documents, and handovers. But once passed on, the specific task still had to be understood and executed by a person, and in complex situations it was often only the person originally responsible who understood it best.

So in daily work, the happiest is the person handing off their work on the way out; the most headache-prone is the person taking over someone else’s handover.

Now there is a new change. Once an employee has organized part of their judgment and experience into a skill and let an agent into the repeatable parts, others can push the work forward by following the skill, without having to find them first.

For the employee, of course they do not become useless tomorrow. Which work can be handed over depends on how far AI, tools, and business conditions have come.

But as long as some tasks can be steadily taken over by AI or other colleagues, it means the company’s dependence on them may decline.

What gets handed over is time — and your own irreplaceability

Often, a skill you use yourself and one connected at the company level, even with the same purpose, are not a matter of simple copying. What you use yourself only has to fit your own habits and preferences; sometimes it need not be so rigorous, and a few small bugs are acceptable.

To actually bring it out for team and company use, the employee often has to do extra optimization and debugging, tuning a skill that only had to run for themselves up to the point where others can call it.

They may also have to account for more business requirements, add handling for exceptions, prepare typical material, and sit through testing. If problems turn up, they have to go back and fix them, help with acceptance, and guide the person taking over. In many companies, the ongoing maintenance work is theirs as well.

For the company this has real value — this ability to accumulate and be handed over is exactly what it wants.

But for the employee it takes more time and energy, and if the original tasks are not reduced, it is a new task added on.

What makes it worse is that what the employee hands over includes their own know-how. What used to be theirs alone is now something others and AI can call.

Once an organization can call on part of an employee’s experience, it may rearrange tasks and people along with that change. If the same batch of work can be done by AI with fewer people, demand for the original roles may fall.

Stack the two accounts together and, for the employee, the sky is practically falling: a real added burden right now, and as for later — whether they are still needed is even in question.

And this does not wait for anything to actually happen. Once employees start distilling themselves, once they have personally gone through their judgment criteria, steps, and exceptions, the sharpest among them know exactly which parts of themselves have already been handed over.

Others see a skill; they see how much of their own work can already be taken away.

So the people who use AI best are the first to recognize what a full company rollout will mean for them.

The effort is rewarded, but what about the risk afterward?

So what can the company do? As a business leader or manager, you may have thought of many ways:

1. Arrange employees’ participation as formal work, rather than as extra hours or something done in passing.
2. Organizing experience, testing, debugging, and mentoring all take time, so free up that time and reduce some of the original tasks accordingly.
3. Count this contribution in evaluations and, where necessary, add some extra incentive.

These arrangements can answer the near-term effort, but they may not answer the job risk that comes later: once this is done, does the company still need them?

This is the real question behind the resistance: if this AI adoption succeeds, they may lose their role instead — so what does the company offer to make him participate willingly?

At that point, two things have to be discussed further.

First, the return. How much of the efficiency gain after the change actually reaches them? Not the line “you are important,” but what they can actually get.

Second, the way forward. After AI is in place and this work is handed over, where does the employee go? If there is a new role, then it depends on whether there is real, sustained work that needs doing and whether they have a chance to get into it; renaming the original role does not count as a way forward.

If the business goal really includes reducing headcount, then who bears the cost will inevitably be put on the table.

Returns, transfers, transitions — these can all be discussed, but putting them on the table and finishing the discussion does not necessarily mean winning agreement.

The conflict of interest is simply there; pressure, mobilization, persuasion, or indoctrination will not make it disappear.

To get a person to actively cooperate with something that may work against them, the company has to offer real arrangements that answer that risk. Whether they accept them still has to be discussed.

How to respond is an unsolved business question

By this point, what a business leader or manager has to reckon is no longer “how do I get employees to cooperate” but another account: for the expected return of this effort, is it worth losing the current stability and the team’s trust?

This is a business question you must answer yourself, depending on your company’s situation and the market. The answer may be to push on, or to narrow the scope and slow the pace, or even to pause.

There is rarely a solution everyone is happy with. You have to decide what you are willing to give and which costs you can bear, before you can determine whether it is still worth pushing further, and how.

Honestly, writing to this point, my mood has grown a little heavy.

The scale and reach of AI’s impact on society, organizations, and individuals is visible to the eye. My judgment is that this impact will accelerate over the coming years.

For an individual, this may be the best of times, or it may be the worst of times.

Acceleration, resistance, bewilderment — everyone responds in their own way, and everyone faces their own unsolved problems.