The Hidden Cost of the AI Revolution: What It's Really Doing to Your Employees

AI is transforming how we work. But behind the productivity gains and efficiency metrics, a quieter story is unfolding, one that plays out in the day-to-day experience of the people doing the work.

The Assumption We Keep Getting Wrong

When organisations talk about AI and their people, the conversation tends to go one of two ways. Either AI is the villain, stealing jobs, dehumanising work, turning employees into data points or it's the hero, freeing workers from repetitive tasks and unlocking human potential. Both narratives are compelling. Both are too simple.

Recent research offers a finding that reframes the whole conversation, if you want to understand what AI is doing to your people, stop looking at the technology and start looking at the work it's creating. Wellbeing isn't shaped by AI adoption itself, it is shaped by what AI does to daily tasks, roles, and the experience of work.

It means the lever isn't the AI, it is everything the AI changes around it.

What the Research Actually Found

The study identified two key factors sitting between AI adoption and employee wellbeing: task optimisation and safety. AI doesn't affect how your people feel in one direct hit, it changes what their work actually looks and feels like day to day, and that is what shapes their wellbeing. Understanding this changes what you need to pay attention to as a leader.

The Task Optimisation Paradox

Task optimisation sounds like an unambiguous win. Eliminate repetitive work, streamline processes, free up cognitive bandwidth for higher-value thinking. And in many cases, that's exactly what AI delivers.

But here's what that story often leaves out is when AI handles more of the workflow, the remaining human work doesn't simply become lighter. It becomes different and often, more demanding.

Consider what's happened in organisations that have reduced headcount alongside AI adoption. The assumption is that AI fills the gap. In practice, the remaining employees absorb increased oversight responsibilities, greater accountability for AI-generated outputs, and the cognitive load of operating alongside systems they may only partially understand. The work doesn't disappear. It transforms and that transformation lands directly on people.

There's also the monitoring question. AI adoption has brought with it a new generation of performance tracking tools systems that can capture data down to the second, monitoring productivity, engagement, and in some cases, emotional states in real time. Companies including IBM, Unilever, Microsoft, and Softbank have explored AI's emotional analytics capabilities, using them in recruitment and to monitor employee attentiveness and wellbeing.

When every task is trackable and every moment is measurable, the nature of work changes in ways that are difficult to quantify but deeply felt. Higher levels of anxiety, a diminished sense of autonomy, and the psychological weight of constant surveillance are real consequences even when the intention behind the monitoring is ostensibly supportive.

Safety, Trust, and the Black Box Problem

The second pathway research has identified is safety and this is where the conversation gets particularly uncomfortable.

AI systems, particularly the more sophisticated ones, are not transparent by design. The "black box" nature of many AI tools means that the decisions they produce about performance, about hiring, about resource allocation arrive without clear explanation. Even the engineers who build these systems often cannot fully account for how a particular output was reached.

For employees, this creates a specific kind of unease. When a decision affecting your career, your workload, or your standing in an organisation is shaped by an algorithm you cannot interrogate, you lose the ability to question, challenge, or even understand the decisions being made about you. Trust breaks down. And where trust breaks down, psychological safety  the foundation of healthy, productive workplaces follows.

AI systems are trained on historical data, data created by humans, carrying human bias. The problem is that when bias comes from an algorithm, it looks like fact. And it's much harder to challenge a number than it is to challenge a person. The result can be discriminatory outcomes, power imbalances, and a workforce that feels, rightly, that the playing field is neither level nor transparent.

The Scale of What's Already Happening

It's worth stepping back to understand the scale of what's unfolding. AI adoption has now reached. A 2025 McKinsey survey found that 78% of organisations have adopted AI in at least one business function, with generative AI specifically at 72%. Therefore, we are not talking about an emerging trend  we are talking about a wholesale transformation of how work is done.

The frameworks guiding AI development are clear wellbeing isn't supposed to be an after thought, it's meant to be built in from the start. The gap between that intention and what employees are actually experiencing right now tells its own story.

What This Means for Leaders

If the research is right and the logic is compelling then the question for organisations is not simply "are we adopting AI responsibly?" but "are we paying attention to the two levers that actually determine how AI lands on our people?"

A few practical implications:

On task optimisation: 

When you introduce AI into a workflow, map what happens to the remaining human work, not just the work the AI takes over. Where is cognitive load increasing? Where has accountability shifted? If headcount has reduced, are the people who remain genuinely supported  or are they quietly absorbing the shortfall?

On safety: 

Transparency is not optional. Employees should understand, at least in broad terms, how AI tools are influencing decisions that affect them. If your AI systems are opaque, that opacity is a leadership problem, not just a technical one. Invest in explainability. Create space for employees to raise concerns about AI-driven decisions without fear of reprisal.

On monitoring: 

There is a meaningful difference between using data to support employees and using data to survey them. The line is not always obvious, but it is always worth drawing. Before deploying any continuous performance or engagement monitoring, ask honestly, does this serve the employee, or does it serve management control?

Don't Forget the Human in the Room

AI is extraordinary technology. The productivity and efficiency gains are real. The capacity to augment human capability, remove drudgery, and surface insights that no human analyst could reach alone these are genuine benefits.

But the efficiency metrics don't capture what's happening to the person who now manages the AI tool, validates its outputs, absorbs its errors, and carries the cognitive and emotional weight of working alongside a system they don't fully understand or control.

That person's wellbeing is not a footnote to your AI strategy. According to the research, it is the outcome your AI strategy most directly shapes  just not in the way most organisations have been thinking about it.

The path AI takes to your employees' wellbeing runs through the work itself: through how tasks are structured, how safety and trust are maintained, and how the human experience of daily work is designed. Get those factors right, and AI can genuinely improve working life. Get them wrong, and the efficiency gains come at a cost that won't show up in your quarterly metrics, but will, eventually, show up in your people.

Sources:

McKinsey & Company. (2025). The state of AI: How organizations are rewiring to capture value. QuantumBlack, AI by McKinsey. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-how-organizations-are-rewiring-to-capture-value 

Valtonen, A., Saunila, M., Ukko, J., Treves, L., & Ritala, P. (2025). AI and employee wellbeing in the workplace: An empirical study. Journal of Business Research, 199, 115584. https://doi.org/10.1016/j.jbusres.2025.115584