AI Won't Just Change How We Work. It Will Change What Work Does to Us.

by Oana Iordachescu

Idea in brief. Most conversations about AI at work are conversations about productivity: how much faster, cheaper, or more scalable work can become. That framing leaves out a second question that matters just as much: what AI-mediated work does to the people inside it.

Why it happens. Work is not a neutral container for technology. It is a social system that teaches people what gets rewarded, who gets seen, and whether they can trust the institution they work for. When organisations bolt AI onto hiring, evaluation, and management without redesigning the systems around it, they don't just automate tasks, they do change what employees learn about their own standing, and that changes how they show up.

The solution. Leaders need to treat AI adoption as an organisational-design problem, not a tooling decision. At least strive to. That means preserving employee agency, keeping decisions explainable, and building in the accountability and transparency mechanisms that research consistently identifies as the difference between AI that earns trust and AI that erodes it.

Ask most executives what keeps them up at night about artificial intelligence, and the answer will be some version of the same question: how much value can we capture, how fast, and who gets displaced along the way.

Those are the questions covered in nearly every AI strategy deck, every earnings call, every McKinsey slide. They are legitimate. We think they are also incomplete.

There is a second question that gets far less airtime, and it may matter more over the long run: what happens to the people who spend their working lives inside AI-mediated systems? Not what AI does for the organisation, but what it does to the humans navigating it.

Work is not a neutral container for technology. It is where people build habits, confidence, professional identity, and a sense of their own competence. It is also a social system with its own quiet curriculum, one that teaches employees, through hundreds of small signals, what gets rewarded, who gets noticed, what good leadership looks like, and whether the institution they work for can be trusted to treat them fairly. AI does not erase that curriculum. It can rewrite it, and it can do so faster and at greater scale than any previous workplace technology.

The evidence is no longer speculative

For several years, claims about AI's psychological and social effects on workers were largely anecdotal. That has changed. A growing body of research from labour economists, organisational scholars, and survey researchers now gives a fairly consistent picture and it is more complicated than either the "AI will empower workers" or "AI will crush workers" narratives suggest.

The OECD's 2023 Employment Outlook, one of the most comprehensive cross-country studies of AI's labour-market effects to date, found that AI can improve certain dimensions of job quality but can also intensify work and increase stress and, notably, that workers subject to algorithmic management were among the least likely to report greater job satisfaction after AI was introduced into their workflow.¹ The same body of OECD research found that AI-driven monitoring can raise perceived fairness in some contexts while simultaneously threatening privacy and worker autonomy, more of a tension, not a contradiction, and one that depends almost entirely on how the system is implemented.²

Gallup's most recent workplace data adds a trust dimension to this picture. As of late 2025, AI use among U.S. employees continues to climb - 45% report using it at least occasionally - but organisational clarity has not kept pace: only 37% of employees say their employer has clearly implemented AI to improve productivity or quality, and nearly a quarter say they simply don't know what their organisation has adopted.³ 

Harvard Business Review's own analytic research, published in December 2025, found that only 6% of companies fully trust AI agents to run core business processes autonomously, with the large majority restricting them to routine or supervised tasks.⁴ 

Adoption is outrunning confidence, and confidence gaps of this kind rarely stay contained to the C-suite - they show up in how employees experience the tools management asks them to use.

Why algorithmic management is a different kind of problem

Much of the anxiety about AI at work is really anxiety about a more specific phenomenon: algorithmic management, the practice of using software to partially or fully automate tasks traditionally performed by human managers. It is worth naming precisely, because it behaves differently than automation of routine tasks.

Management scholars Kellogg, Valentine, and Christin, in an influential Academy of Management Annals review, describe algorithmic control as operating through six mechanisms - what they call the "6 Rs": employers direct workers by restricting and recommending, evaluate them by recording and rating, and discipline them by replacing and rewarding.⁵  The framework is useful precisely because it shows that algorithmic management is not one thing. It is a bundle of design choices, each with its own consequences for how power and visibility are distributed between employer and employee.

Some of those consequences are already visible at scale. MIT Sloan Management Review has documented how UPS outfits delivery trucks with sensors that track drivers' every movement to optimise routes, and how Amazon's algorithms track worker productivity closely enough to generate termination paperwork automatically when targets are missed.⁶  These are productivity systems functioning exactly as designed. They are also, for the people inside them, a fundamentally different experience of being managed - one with less room for context, negotiation, or explanation than a human supervisor would typically provide.

The OECD's dedicated report on algorithmic management is direct about the trade-off: these systems can deliver real productivity and consistency gains, but there is growing evidence of harm to worker well-being when they are deployed without corresponding attention to transparency and accountability.²  The risk is not the automation itself. It is automation without redesign - dropping algorithmic decision-making into a management structure that was never built to explain or contest it.

In a study of Uber drivers, researchers Mareike Möhlmann and Ola Henfridsson found that dissatisfaction with algorithmic management clustered around three recurring complaints: constant surveillance, a sense of dehumanisation, and a lack of transparency about how decisions were made. Their prescription was notably low-tech - share more information, invite feedback, build in human contact, and treat trust as something to be actively built rather than assumed.⁷ 

Six years and several AI generations later, that prescription has not gone out of date. If anything, it has become more urgent, because the systems it was written about have only grown more capable and more embedded.

The psychology of being measured differently

Decades of organisational psychology research, going back to foundational work by Jason Colquitt and earlier procedural-justice scholars, establishes a consistent finding: how fair a decision-making process feels to employees shapes their trust in an organisation even more than the outcome of any single decision.⁸ 

People can accept a decision that doesn't go their way if they believe the process was legitimate. What erodes trust is opacity, not knowing why a decision was made, or sensing that there is no one to ask.

This is precisely the terrain where algorithmic management is weakest. Academic research on algorithmic control consistently finds that workers' negative reactions to these systems are driven less by the decisions themselves than by the power imbalance created when workers cannot see how they are being evaluated, and have no clear channel to challenge outcomes they believe are wrong.⁹  Some studies find that increased transparency helps - it improves workers' sense-making, satisfaction, and even reduces turnover - but only when transparency is paired with genuine avenues for input, not simply a dashboard explaining a decision that has already been made.⁹

Preliminary survey research adds a more personal dimension to this picture. The Human Clarity Institute's 2025 survey on AI, work, and identity - a smaller-scale but useful data point alongside the larger institutional studies - found that a meaningful share of workers report a subtle erosion of ownership over their own output when AI is heavily embedded in their workflow, along with concerns about the sheer volume of data being collected about their performance.¹⁰  That finding points toward something the larger datasets can't fully capture: a slow, cumulative psychological cost of working inside systems that watch more and explain less.

What this means for the people-systems around AI

None of this is an argument against using AI in management, hiring, or performance systems. It is an argument that the core question for leaders is not simply "does the model work?" It is "are the human systems around this model strong enough to use it well?"

Consider four processes that already function as quiet teaching moments inside any organisation, with or without AI:

• A hiring process teaches candidates, often before they've accepted an offer, whether they will be treated with respect.

• A performance review teaches employees what actually counts as success, regardless of what the values statement says.

• A promotion system teaches people whose work gets seen and whose doesn't. A restructuring teaches an entire organisation, in one stroke, what it truly prioritises when it has to choose.

  
  

AI can make each of these processes faster, more consistent, and more scalable. It can just as easily make them colder, harder to challenge, and more difficult to explain - and the difference between those two outcomes is almost entirely a matter of design choices leaders make before the system ever goes live.

The World Economic Forum's 2026 report on AI at work, drawn from more than twenty technology companies and their enterprise clients, arrives at a similar conclusion from the deployment side: organisations that successfully scale AI share five practices - human accountability for AI-assisted decisions, redesigning operating models end-to-end rather than bolting AI onto old workflows, building scalable talent systems, prioritizing transparency as a trust mechanism rather than an afterthought, and disciplined experimentation instead of wholesale rollout.¹¹  Notably, none of these five practices is primarily a technical requirement. They are organisational-design commitments, and they are exactly the commitments that determine whether algorithmic management earns trust or erodes it.

The test that you might want to take

The most important AI question a leadership team can ask is not only whether the technology performs well. It is whether the organisation is deploying it in a way that preserves employee agency, keeps decisions explainable, and keeps trust intact under pressure.

In an AI-enabled workplace, leaders are not merely rolling out tools. They are shaping the conditions under which people learn, adapt, and decide whether they belong. Technology determines how work gets done. The people-systems around that technology - hiring, evaluation, promotion, and the everyday texture of being managed - determine how work feels. Over a career spent inside those systems, that distinction compounds.

Workplaces do not only produce output;

they produce expectations, professional identity,

and a durable sense of whether

one's judgment is trusted.

The real test of AI at work, then, is not only whether it helps an organisation move faster. It is whether it helps the people inside that organisation remain capable, thoughtful, and trusted while they do their jobs - because if it doesn't, no productivity gain will be large enough to offset what gets lost.

Sources

1. OECD, "Artificial Intelligence, Job Quality and Inclusiveness," OECD Employment Outlook 2023. ↩

2. OECD, "Algorithmic Management in the Workplace". ↩ ↩²

3. Gallup, "AI Use at Work Rises", 2025; see also HR Dive, "US Workers Report a 'Major AI Trust Gap'". ↩

4. Harvard Business Review Analytic Services (sponsored by Workato and AWS), The Enterprise AI Trust Gap, December 2025, as reported in Fortune. ↩

5. K. C. Kellogg, M. A. Valentine, and A. Christin, "Algorithms at Work: The New Contested Terrain of Control," Academy of Management Annals 14, no. 1 (2020): 366–410. ↩

6. MIT Sloan Management Review, "Algorithmic Management: The Role of AI in Managing Workforces". ↩

7. M. Möhlmann and O. Henfridsson, "What People Hate About Being Managed by Algorithms, According to a Study of Uber Drivers," Harvard Business Review, August 30, 2019. ↩

8. J. A. Colquitt, "On the Dimensionality of Organizational Justice: A Construct Validation of a Measure," Journal of Applied Psychology (2001); see also Colquitt, "Forever Focused on Fairness: 75 Years of Organizational Justice in Personnel Psychology," Personnel Psychology (2023). ↩

9. See the algorithmic-management literature synthesized in New Technology, Work and Employment (2025), "The Rise of Algorithmic Management and Implications for Work and Organisations", and related transparency-paradox research in Computers in Human Behavior. ↩ ↩²

10. Human Clarity Institute, "AI, Work & Human Identity Survey," 2025 — an independent survey dataset; treat as an early, smaller-scale signal alongside the institutional research above. ↩

11. World Economic Forum, "AI at Work: From Productivity Hacks to Organizational Transformation," 2026. ↩