AI Isn’t Replacing You. It’s Rewriting the Rules You Work Under

We’ve been asking the wrong question about AI. It’s not whether it will destroy millions of jobs or save us from work altogether. Those are convenient myths — one sells fear, the other sells software. The real shift is subtler and far more consequential: AI is quietly restructuring power in the workplace. And that restructuring moves in only one direction and means more control, more pressure, more dependency and less room to negotiate on the part of the workforce.

There are important psychological and cultural factors at play here. The corporate narrative promises liberation: “delegating routine tasks” to focus on higher-value work. However, as a growing body of evidence shows, the opposite tends to happen. When somebody incorporates AI tools into their day-to-day work, the organization sees it not as an opportunity to reduce their workload, but to increase their productivity. If you do what used to take four hours in two hours, you don’t get two hours off: you are assigned more work; the bar is raised: your increased productivity simply becomes a new baseline. The result is an intensification of work: more tasks, tighter deadlines, longer days, the same old feeling of always having to run to catch up.

Added to the pressure of human overclocking is another, less visible dimension: anxiety. Not only because of the fear of replacement, but because of the uncertainty about how performance will be evaluated in an environment where AI increases output. If work is measured by quantity (of emails, tickets, deliverables, reports, etc.) AI simply speeds up the metrics. And if evaluation is also automated, the circle closes: tools that produce more, metrics that demand more and systems that monitor employees more.

Another aspect to all this, and one that is hardly debated, is the growth of shadow AI; the use of AI tools not approved by the company, off the IT department’s radar and thus not subject to any security or regulatory compliance. I’m not talking about four geeks trying out a chatbot; this is a practice that spreads because it is useful, fast and, above all, because official channels tend to arrive late, with restrictions and without understanding the real needs of the job. The employee has a task to perform, a deadline to meet… and a free tool that miraculously solves problems. The temptation is obvious.

The hitch is that this convenience comes at a huge cost: sensitive data that travels where it shouldn’t, contract templates, business information, strategy drafts, internal reports, code, customer data, conversations… All of this can end up being copied and pasted into services that do not offer meaningful guarantees of privacy. There are already a number of high-profile cases of accidental leaks precisely for this reason, and it is not difficult to understand why: the stressed out employee is not thinking about intellectual property or business secrets; their focus is meeting the latest tight deadline. The perverse thing is that the company, by not providing safe alternatives, indirectly encourages the use of shadow AI.

And here comes the twist: it’s not just about using AI; it’s about how it’s trained. More and more employees are building their own assistants: a bot that perfectly mimics the department’s style, complete with a knowledge base with internal documentation, a system that summarizes past incidents, a co-pilot fed with emails, presentations or reports. They do it because it works. But what they are doing, de facto, is converting the company’s knowledge based into a model with reusable data. And that raises an explosive question: whose knowledge is it when it becomes a chatbot? Does it belong to the employee who structured it, the company that generated it, the customer whose data appears in the documents, or the provider of the tool that makes the system possible?

The legal answer is clear: much of the work an employee produces in their working day belongs to the company, and internal information can constitute a trade secret. In Europe, the protection of know-how and undisclosed information has a specific legal framework. But the real problem is not what the rule says, but how it is managed in practice when thousands of people are “distilling” that knowledge into systems that can end up out of control, duplicated, shared or simply poorly protected. At best, the company is exposed to leakage and compliance issues. At worst, it is unknowingly financing the construction of tools that make part of human labor dispensable… with data generated by that work itself.

In this context, the promise of the augmented worker is simply the worker as an involuntary trainer of their substitute, because the knowledge that made them valuable becomes an automatable and scalable asset. And, in the meantime, the day-to-day becomes more intense: more tasks, more urgency, more expectations, less respite.

The conclusion is that we need an honest conversation about AI. If the discussion is limited to how many jobs are destroyed, we miss the core question of power. AI, as it is currently being deployed, tends to favor employers, those who control the objectives, metrics, tools and data.

If we want everybody to benefit, then we’re going to need serious internal rules, real governance, safe alternatives to shadow AI and, above all, negotiation: what is automated, with what limits, with what guarantees and how productivity gains are shared. Otherwise, the AI revolution will not bring greater well-being. It will be, simply, another turn of the corporate screw.

—

This post was previously published on Enrique Dans’ blog.

***