When AI Makes Work Less Human

Artificial intelligence is rapidly becoming part of the operating system of work. Organizations are using it to write, summarize, analyze, support employees, plan workforces, evaluate information, and increasingly influence decisions about how work gets done. In Remesh's May 2026 study of 105 supervisors, managers, and senior leaders, three-quarters described AI adoption in their organizations as growing or widespread. A month later, most of the 155 employees surveyed said they were already using AI for at least some portion of their work.

Much of the organizational conversation has understandably focused on productivity: hours saved, tasks automated, faster decisions, higher throughput. Those are legitimate questions. But they may be incomplete. As AI becomes more embedded in work, organizations also need to ask what happens to the person doing the work. Does AI expand human capability, judgment, and agency, or does it gradually narrow them? Does it free employees for more meaningful work, or simply raise the amount of work expected? Does it create more space for human interaction, or replace it?

Our field research suggests that the answer is not automatically positive or negative. AI can remove repetitive work, improve access to information, and create more time for creativity and complex problem solving. It can also increase pressure, diminish autonomy, weaken direct human interaction, and make employees feel that their own judgment carries less weight. Read alongside psychologist Nick Haslam's influential theory of dehumanization, these findings point to a serious possibility: poorly designed AI implementation may contribute to a subtle form of workplace dehumanization - not by treating people as animals, but by treating them increasingly like machines.

Dehumanization does not have to look extreme

The word dehumanization usually evokes extreme examples: violence, prejudice, humiliation, or deliberate cruelty. Haslam's 2006 review makes a more unsettling argument. Dehumanization can also arise through ordinary social and organizational processes. It can be subtle, interpersonal, and embedded in everyday systems rather than driven by overt hostility. Haslam distinguishes two broad forms. The one most relevant to AI is mechanistic dehumanization, in which people are denied characteristics associated with human nature itself: emotionality, warmth, individuality, cognitive openness, depth, agency, and vitality. The implicit comparison is not between a person and an animal. It is between a person and an automaton.

That distinction matters enormously in the age of AI. Haslam's review links mechanistic dehumanization to technology, standardization, instrumental efficiency, impersonal technique, and enforced passivity. In his discussion of modern medicine, for example, technology becomes potentially dehumanizing when efficiency and standardized information crowd out individuality, subjective experience, autonomy, and agency. The implication for organizations is straightforward: AI does not dehumanize simply because machines enter the workplace. The risk arises when organizations begin managing humans as though they were machines.

The risk is already visible

Our analysis asked employees directly whether AI was increasing or reducing autonomy and dignity in their work. The results were notably divided: 32% said AI helped, 27% said it hurt, and the remainder saw little change. In this research, dignity referred specifically to employees' sense that AI preserved rather than diminished their autonomy, judgment, and standing in their own work.

Twenty-seven percent is not evidence that AI is broadly dehumanizing the workforce. These are relatively small, nonrepresentative field studies, and the results should not be interpreted causally. But the number is large enough to show that the issue is not hypothetical. A meaningful share of employees already perceive that something important about their agency or standing is being lost. More importantly, positive attitudes toward AI did not eliminate the risk. Some employees who otherwise rated their organization's AI rollout positively still reported reduced autonomy and control over their own judgment. An employee can believe AI works, appreciate its usefulness, and still feel less control over how work gets done. Productivity and dignity can move in different directions.

Six ways AI can make work less human

1. Transferring judgment from people to systems. Remesh's analysis found that AI threatens autonomy when employees worry about reduced human decision-making, overreliance on AI, and declining critical thinking. By contrast, AI supports autonomy when it is framed as augmentation and employees retain meaningful ownership of decisions. This matters particularly in environments where organizations say there is a 'human in the loop,' but the employee has little practical freedom to challenge the system. A person who remains formally accountable while being socially expected to follow an algorithm is not exercising meaningful judgment. Human oversight becomes symbolic rather than substantive.

2. Converting employees into units of productivity. AI makes efficiency gains unusually visible, which can create a strong temptation to translate every minute saved into additional output. Leaders in our May study described shorter deadlines, streamlined workflows, and higher throughput expectations. Respondents talked about work that once took hours being completed in minutes, only for more work to be added. Some employees also described pressure to outperform AI simply to demonstrate their continuing value. Efficiency itself is not dehumanizing. The danger comes when efficiency becomes the dominant lens through which the employee is valued. If every gain is converted into more throughput, AI becomes a technology of work intensification rather than one of human augmentation.

3. Reducing employees to data representations. AI allows organizations to quantify an expanding range of human activity: skills, performance, productivity, communication patterns, behavioral signals, potential, and risk. These measures can be useful, but they become dangerous when the representation begins to substitute for the person. Haslam's discussion of technologically mediated information is relevant here: systems become dehumanizing when objective measures crowd out subjective experience, individuality, and agency.

4. Fraying human connection. Employees in our research worried about reduced human interaction, declining direct communication, and the loss of the 'human touch.' At the same time, they also recognized that AI could improve collaboration and free time for more human-centered work. That tension is central. AI can remove unnecessary friction, but organizations need to remember that some apparent inefficiencies - coaching, mentoring, informal conversation, disagreement, shared interpretation - are actually how people build trust and meaning together.

5. Treating employees as interchangeable collections of tasks and skills. As AI makes it easier to decompose jobs into component activities, organizations may begin thinking about workers less as whole people and more as bundles of productive functions. The practical language may sound neutral - capacity, skills, automation potential - but the underlying mindset can become mechanistic if tacit knowledge, relationships, history, judgment, and individuality disappear from view.

6. Reducing agency and voice. Organizations can introduce AI to employees or build AI with employees. The difference is not merely philosophical. In our analysis, only 25% of employees with low or no involvement in shaping AI implementation rated integration very or extremely effective. Among highly involved employees, that figure rose to 71%. Positive culture impact moved from 50% to 90%. That finding also aligns with sociotechnical systems thinking, which argues that people closest to a technology should have input into its design and meaningful control over relevant work processes. In this sense, the connection to dehumanization is unusually clear: being treated as a passive recipient of a system is mechanistic; participating in shaping the system preserves agency.

What organizations should watch for

The most important warning signs are unlikely to arrive under the label 'dehumanization.' Organizations should instead watch for mechanistic drift: gradual changes in work design and culture that reduce agency, individuality, judgment, or human connection. Examples include employees saying that 'the system decides,' declining willingness to challenge AI recommendations, increasingly algorithmic performance management, productivity expectations rising automatically as AI saves time, growing role ambiguity, less direct human interaction, decisions employees cannot meaningfully appeal, persistent job-displacement anxiety, and greater reliance on scores and dashboards as substitutes for managerial judgment.

Language can be revealing as well. When people gradually disappear from management vocabulary and become primarily headcount, resources, capacity, or throughput, that can signal a deeper change in how the organization is conceptualizing them. Haslam describes mechanistic dehumanization as an instrumental and distancing orientation in which people are treated less as social partners and more as objects within a functional system. The danger is that this shift rarely arrives as a single decision. It accumulates through everyday choices about measurement, automation, workload, performance, and voice.

Governance helps, but governance alone will not humanize AI

One of the most useful findings in our research is that formal guidance does not solve the human problem on its own. Employees in organizations with clear AI guidance were almost as likely to report reduced autonomy as employees in organizations with no formal guidance: 25% versus 30%. Yet guidance was strongly associated with perceived effectiveness. Among employees with very limited guidance, 18% rated AI integration very or extremely effective; where guidance was clear, that figure rose to 73%. Positive culture impact increased from 43% to 94%. The implication is not that governance does not matter. It clearly does. Rather, governance and involvement appear to do different jobs. Policies make expectations and boundaries visible. Employee involvement helps preserve agency and standing. Organizations need both.

A responsible-use policy can tell employees what they may do with AI. It cannot, by itself, ensure that they have meaningful voice, preserve their judgment, maintain relationships, or believe they still have a valued role in the organization. Humanizing AI requires not just rules around the technology, but decisions about how work itself will be designed.

Designing AI to preserve humanity

Organizations can reduce the risk of dehumanization by making a few design choices explicit. AI should be framed and implemented primarily as a tool for augmentation rather than subordination. Our participants repeatedly associated positive outcomes with positioning AI as support rather than replacement. Organizations should also preserve meaningful human judgment. Human oversight should mean more than asking employees to approve what the system has already determined. People need the authority, knowledge, and psychological safety to question outputs, bring context to decisions, and override recommendations when appropriate.

Employee involvement should occur before important decisions are finalized, not merely after implementation as a change-management exercise. Our analysis identifies involvement as one of the strongest relationships observed with both perceived effectiveness and positive culture impact. Organizations should also broaden their measurement systems. Every AI scorecard that tracks adoption, hours saved, or productivity should include indicators of autonomy, judgment, role clarity, voice, fairness, relatedness, psychological safety, and perceived standing.

Leaders should also resist the assumption that every efficiency gain must become additional workload. AI can free people for creativity, strategy, complex problem solving, collaboration, and judgment - or it can simply intensify the pace of work. That is not a technological inevitability. It is an organizational choice. Finally, organizations need to invest in human capability as aggressively as they invest in AI capability. Employees ranked training and skill development first among ten strategies for improving AI integration. Literacy matters not simply because it improves utilization; it creates agency. People cannot meaningfully influence a technology they do not understand.

AI may reveal what an organization really believes about people

There is a deeper implication running through all four documents. AI may function as an amplifier of an organization's underlying theory of the employee. If employees are fundamentally understood as costs, resources, task performers, or units of output, AI can make that worldview extraordinarily powerful. Organizations can measure people more continuously, standardize their work more completely, prescribe behavior more precisely, and compare output more relentlessly. In that environment, AI may accelerate exactly the kind of mechanistic orientation Haslam describes.

But organizations operating from a different assumption - that employees are thinking, feeling, relational, meaning-making agents - can use the same technology in a very different way. AI can remove drudgery, expand access to knowledge, increase individual capability, and create more room for judgment, creativity, collaboration, and human connection. The employee evidence captures this duality well: AI is experienced as beneficial when it creates speed, useful information, empowerment, and structured support, but as threatening when those same dynamics turn into pressure, ambiguity, or replacement.

That makes the risk of dehumanization serious, but not inevitable. It is largely a consequence of design, management, and organizational choices. There is also a business reason to take those choices seriously. Across our two studies, the conditions associated with protecting dignity - literacy, employee involvement, and clear guidance - were also associated with stronger AI effectiveness outcomes. The research cannot establish that dignity causes better performance, but the two moved together rather than appearing to trade off.

That suggests a different way to frame the challenge. Dignity is not simply a constraint organizations should place around AI after the important decisions have been made. It may be part of the infrastructure required for AI to work well in the first place. The most useful question for leaders, then, may not be how much work AI can automate or even how much more productive employees can become. It may be whether AI is being used to create more room for the distinctly human parts of work: judgment, agency, creativity, relationships, meaning, and voice.

What leaders can do now

1. Design for augmentation. Use AI to extend human capability and remove low-value work, not simply to increase control or replace judgment.

2. Keep real humans in the loop. Give employees genuine authority to question, contextualize, and override AI outputs when appropriate.

3. Involve employees early. Co-design important AI-enabled workflows with the people who will live inside them.

4. Measure dignity as well as productivity. Track autonomy, voice, judgment, role clarity, fairness, relatedness, and psychological safety alongside adoption and efficiency.

5. Do not automatically convert time saved into more work. Use some of the capacity AI creates for higher-order thinking, learning, creativity, collaboration, and human connection.

6. Build AI literacy. Invest in role-specific training so employees can use AI confidently, understand its limits, and participate meaningfully in decisions about its use.

Interested in reading the full study? Download it here.

Sources

Haslam, N. (2006). Dehumanization: An integrative review. Personality and Social Psychology Review, 10(3), 252-264.

Remesh. (2026, May). Exploring the Impact of AI on Organizational Culture: Learning from Supervisors, Managers, and Senior Leaders. Field study of 105 leaders and people managers.

Remesh. (2026, June). AI Adoption & Integration: Exploring Employees' Perspectives. Field study of 155 employees in the United States and United Kingdom.

‍