We thought AI would save us time. Why do we feel like we have even less of it?

February 2026. Harvard Business Review publishes a study that should cool the prevailing enthusiasm around generative AI in business. Aruna Ranganathan and Xingqi Maggie Ye, researchers at Berkeley, observed approximately 200 employees at an American tech company for eight months after the deployment of AI tools. Their conclusion is unequivocal: AI has not reduced workload. It has transformed and intensified it.

This finding is not anecdotal. It reveals a structural mechanism that many organizations will discover too late: the productivity gains promised by AI translate neither into free time nor reduced workload. They fuel a spiral of acceleration and task expansion that ultimately exhausts teams.

A Technological Promise We’ve Heard Before

This is not the first time a technology has promised to liberate work.

Mechanization was supposed to reduce physical effort. Information technology was supposed to eliminate paperwork. Lean was supposed to eliminate waste. ERP systems were supposed to streamline processes. Each time, the gains were captured by intensification.

Industrial automation did not reduce engineers’ working time: it multiplied the variants to design, the configurations to optimize, the standards to maintain. Computerization did not lighten administrative tasks: it created new reporting requirements, new interfaces, new traceability demands. Lean did not simplify work: it eliminated buffers, accelerated flows, and transferred the load onto those who remained.

AI follows this trajectory. It does not break with the logic of continuous optimization that has structured organizations for fifty years. It accelerates it.

The difference this time lies in the cognitive nature of the work being impacted. AI does not simply replace gestures or calculations: it transforms how we think, decide, and arbitrate. And this transformation occurs without anyone truly managing its effects on collective mental load.

What the Study Actually Shows

The observed employees do indeed work faster on certain tasks. But three dynamics absorb these gains:

Acceleration creates new expectations. As soon as AI allows a case to be processed in two hours instead of five, the norm becomes two hours. The time saved is never recovered: it is immediately reallocated. Managers adjust their expectations, often implicitly. Speed becomes the standard, not the exception.

Work perimeters spontaneously expand. With AI, employees take on tasks they previously delegated or would never have considered. They write more complex syntheses, explore more scenarios, assume responsibilities that exceed their job descriptions. This expansion is not imposed: it happens naturally, because it is now possible. The organizational effect is direct: recruitment is delayed, more workload is absorbed with the same headcount.

Cognitive load increases instead of decreasing. AI does not eliminate intellectual work: it shifts it. Employees now spend their time verifying AI outputs, arbitrating between multiple suggestions, controlling result consistency. This supervision and validation work is demanding, fragmented, and less visible than the tasks it replaces. It generates a diffuse cognitive fatigue that productivity indicators do not capture.

Result: employees work as much, if not more. Working time does not decrease. Exhaustion increases. And AI’s benefits end up neutralized by an overload that no one explicitly decided upon.

What This Could Mean in Industry

The study does not focus on the industrial sector, but its mechanisms allow us to anticipate what could concretely unfold.

Imagine, in a chemical plant, a process engineer using AI to optimize production parameters. Before, he tested two or three configurations per week. With AI, he could generate fifteen. The model runs fast, results arrive immediately. But each simulation would need to be verified, contextualized, arbitrated. Time saved on calculation would be absorbed by analysis. And because it’s possible, he would now be expected to systematically explore all variants.

In an automotive design office, a project manager could use AI to generate technical syntheses. She used to produce one report per project. Now, she could produce three: one for management, one for suppliers, one for compliance. AI would allow her to handle this volume. No one would explicitly ask her to multiply deliverables. But since it’s feasible, it would become the norm.

In a predictive maintenance center, a manager would now receive fifty alerts per day instead of ten. The model would detect more weak signals. All would need to be sorted, prioritized, deciding which merit intervention. The gain in anticipation would be real. But the mental load of constant filtering would exhaust the team.

AI would not eliminate work. It would create invisible volume: more variants, more scenarios, more controls, more arbitrations. This volume would not appear in any productivity indicator. It would manifest in fatigue, in meetings that run long, in days that overflow.

Why This Mechanism Escapes Organizations

Work intensification through AI is insidious because it develops without clear directive. No one formally decides to increase workload. Tools are deployed with the intention of relieving teams. But the absence of a structuring framework—usage norms, workload limits, redefined objectives—leaves the field open to an implicit logic: if it’s technically possible, then it’s expected.

Productivity gains do not automatically translate into well-being or collective efficiency. They require conscious organizational arbitrations: deciding what to do with freed time, redefining responsibility perimeters, adjusting managerial expectations. Without these arbitrations, AI becomes a pressure accelerator, not a liberation tool.

The Ranganathan and Ye study shows that this dynamic is not a temporary side effect. It is a structural consequence of unregulated AI adoption. Organizations that do not actively govern AI usage will see their teams burn out while believing they are gaining efficiency.

Open Questions for a Necessary Debate

This study opens several fronts that European companies, particularly industrial ones, must confront.

Can we truly capture AI’s productivity gains without profoundly transforming work organization? If processes, metrics, and managerial expectations remain unchanged, AI will merely accelerate an existing system. It will not reinvent it. Technical gains will be absorbed by organizational inertia.

Does AI reveal a deeper tension between efficiency and workload? For decades, automation has promised to free up time. Yet working time in skilled professions has not decreased. AI could worsen this trend by creating a perpetual race to productivity, where each gain becomes a new floor.

How to measure AI’s real impact beyond performance indicators? Classic dashboards show tasks completed faster, volumes processed up. They show neither cognitive fatigue, nor implicit perimeter extension, nor degradation of decision quality under pressure. What indicators would detect intensification before it becomes unsustainable?

Who is responsible for regulating AI usage within the company? Employees adopt AI to stay performant. Managers adjust their expectations. Executives measure gains. But no one explicitly pilots overall workload. This governance absence creates a void where intensification develops without anyone truly deciding it.

Can AI be a transformation tool if not accompanied by a redefinition of work’s purposes? Accelerating for acceleration’s sake makes no sense. If AI is to serve organizations, we must decide what to do with the time it theoretically frees. Otherwise, that time simply disappears into ambient rhythm.

Technology Doesn’t Reduce Work: It Reveals What the Organization Decides to Do With It

The Ranganathan and Ye study recalls an often-forgotten truth: technology does not decide its own usage. AI does not reduce work by magic. It changes the nature of tasks, but the net effect on workload depends entirely on organizational choices.

The question is not whether AI saves us time. The question is: what work organization do we want to build with it?

For AI to become a transformation tool and not merely a pressure accelerator, companies must govern its usage with the same rigor they apply to budgets or industrial processes. This means defining clear policies, setting limits, measuring effects beyond apparent productivity, redefining what we truly expect from teams.

Failing this, AI will remain what it is today in many organizations: a promise of liberation that transforms, in practice, into silent intensification.