AI in the Workplace Part 3: Human-Centered AI – Designing Healthier Workplaces

AI in the Workplace Part 3 of 3

Part 2 of this series examined the hidden demands of AI-enabled work. While AI can reduce workload and improve efficiency, we also explored important questions about critical thinking, workplace stress, collaboration, movement, and employee well-being. Understanding these emerging challenges is only the first step.
 

Human-Centered AI: How to Design Healthier, More Sustainable Workplaces

Generative artificial intelligence (AI) is quickly becoming part of everyday work. As discussed in Parts 1 and 2, these tools can make work easier and faster, but they can also change workload, learning, collaboration, decision-making, and the physical and cognitive demands of work.

This means that introducing AI into a workplace should involve more than simply giving employees access to a new tool. Research on human-centred AI suggests that AI implementation should be approached as a work-design process, with consideration for how the technology affects the people using it.¹

Start With the Work, Not the Technology

One of the easiest mistakes organizations can make is purchasing an AI tool and then trying to find as many ways as possible to use it.

A human-centred approach starts in the opposite direction.

Employers should first understand the work itself. What tasks are employees completing? Which tasks create unnecessary workload? Which require professional judgment? Which provide important learning opportunities? Where are employees already experiencing bottlenecks? Only then should organizations decide where AI fits.

Employers should also consider the entire workflow rather than measuring success solely by how quickly one task can be completed. AI may make one part of a task faster while creating new responsibilities for reviewing, correcting, contextualizing, and approving its output.²

Do Not Automatically Fill Every Minute AI Saves

If AI turns a two-hour task into a one-hour task, it can be tempting to immediately fill that hour with additional work. If this happens repeatedly, however, increased efficiency may become increased work intensity.

As discussed in Part 1, AI can reduce some demands while creating others.³˒⁴ Employers should therefore consider not only how much more work employees can produce, but whether expected work volume has increased, how much checking is required, and whether AI is actually reducing workload.

Productivity and sustainable performance are not necessarily the same thing.

Teach Employees How to Evaluate AI, Not Just How to Prompt It

AI training often focuses heavily on prompts: how to ask better questions, provide appropriate context, and obtain better outputs.

Those skills matter, but they are only one component of AI literacy.

Employees also need to understand what AI is good at, where it struggles, how to verify claims and sources, when information should be independently verified, and when AI should not be used. Training should also address privacy, confidentiality, bias, ethical concerns, and organizational expectations surrounding AI use.

This includes developing metacognitive skills, or the ability to think about and monitor one’s own thinking.⁵ In practice, this can include recognizing gaps in your own knowledge, questioning how confident you are in an answer, noticing when you are relying on AI rather than reasoning through a problem yourself, and recognizing when you need another source or someone with greater expertise.

Employees also need to know when to stop an AI activity. Because generative AI can continue producing more information, alternatives, and revisions, employees can easily spend unnecessary time prompting without meaningfully improving the result. Setting a clear purpose before using AI, recognizing when enough information has been gathered to complete the task, and knowing when additional prompting is no longer adding value can help prevent employees from becoming lost in information.

The goal should therefore not simply be to create employees who are good at using AI.

It should be to create employees who are good at deciding how to use it, how much to trust it, and when to stop using it.

Protect Learning Opportunities for Junior Employees

Efficiency cannot be the only consideration when deciding what to automate. Some inefficient tasks may also be developmental tasks.

Organizations should deliberately identify tasks where independent practice remains important and avoid automatically delegating every junior-level task to AI. Emerging research suggests that junior professionals may face learning challenges when they are expected to work with emerging AI technologies without having the deeper knowledge and experience needed to identify novel risks.⁶

AI could instead be incorporated into the learning process. For example, a junior employee might complete an initial analysis independently, compare it with an AI-assisted version, discuss the differences with a senior colleague, and then revise the final product.

This preserves something AI cannot easily replace: feedback from someone with experience across a variety of contexts and environments.

Preserve Human Collaboration in AI-Enabled Workplaces

AI should not quietly become the default coworker.

Workplace interactions serve purposes beyond information exchange. Employees learn how colleagues may think differently, creating diversity in thought and creativity. They develop relationships, observe experienced workers, debate alternatives, receive informal feedback, and develop communication and collaboration skills.

Research has demonstrated that individuals working with generative AI can perform some collaborative tasks at levels comparable to human teams.⁷ That capability makes it even more important to deliberately preserve valuable human interactions rather than assuming that any interaction AI can replace should be replaced.

Mentorship, peer review, team problem-solving, case discussions, brainstorming, and opportunities for junior employees to observe experienced colleagues should remain part of AI-enabled work.

Build Human Oversight Into the Workflow

Telling employees to “double-check AI” is not really a system.

Organizations need clear expectations about what needs to be checked, by whom, and when. The level of oversight should also reflect the consequences of an error.

An AI-generated brainstorming list and an AI-generated recommendation that could influence a clinical, financial, legal, or safety-related decision should not receive the same level of scrutiny.

The higher the consequence of being wrong, the stronger the human oversight should be.

Low-risk tasks might require basic employee review, while higher-risk work may require independent source verification, subject-matter review, documentation, or formal human approval.

Most importantly, responsibility for the final work should remain clear.

AI can contribute to a task. It cannot take accountability for the consequences.

Give Employees Permission to Question AI

AI literacy alone is not enough if workplace culture rewards speed over judgment.

Employees need to feel that questioning an AI output, independently verifying information, asking another person for help, or choosing not to use AI for a particular task is acceptable.

Meaningful human oversight therefore requires more than telling employees they are responsible for the final result. Organizations must create enough time, autonomy, and psychological safety for employees to actually exercise that responsibility.

Involve Employees in AI Implementation

The people actually performing the work often know things about it that decision-makers do not. They know where bottlenecks exist, which tasks require more judgment than they appear to, and when a new technology is making a workflow worse rather than better.

Employee participation should therefore occur before, during, and after AI implementation. Research on human-centred AI has highlighted employee participation as an important part of socially sustainable AI adoption.¹

Rather than simply asking employees whether they like the technology, employers can ask: What became easier? What became harder? What new work did AI create? What are employees doing less often?

These questions help organizations understand how AI is changing the actual work system.

Design Movement Back Into Increasingly Digital Work

If more parts of a job can be completed from the same computer, employers should not assume movement will happen naturally.

Movement may need to become more deliberate through task variation, brief movement and posture changes, opportunities for in-person collaboration, appropriately designed workstations, and productivity expectations that allow employees to step away from their screens.

Current evidence does not establish that generative AI itself causes prolonged sitting. However, prolonged occupational sitting is already an established concern in computer-based work, and workplace interventions incorporating digital components can reduce sedentary behaviour.⁸

Physical ergonomics should therefore remain part of AI implementation rather than becoming an afterthought.

Measure Whether AI Is Improving the Job, Not Just the Output

Finally, organizations should evaluate AI adoption using more than productivity.

An employee might produce more work while experiencing greater mental workload. A team might work faster while junior employees receive fewer developmental opportunities. An AI system might reduce administrative time while increasing the amount of checking employees perform.

Organizations should therefore consider workload, trust, job satisfaction, learning opportunities, quality, collaboration, physical demands, and well-being alongside productivity.

This is where ergonomics offers an important perspective.

The goal is not simply to optimize the technology. The goal is to optimize the work system.

Working Better, Not Simply Faster

Generative AI presents an enormous opportunity to improve work. But productivity should not be the only measure of progress.

Human-centred AI does not mean resisting automation. It means being intentional about what we automate, what we preserve, and what we redesign.

From an ergonomics perspective, perhaps the most useful question employers can ask is not:

“How much work can AI do for our employees?”

but:

“How can AI help us design better work for the people doing it?”

That distinction may ultimately determine whether AI creates workplaces that are simply faster or workplaces that are healthier, more sustainable, and genuinely better to work in.

Frequently Asked Questions

What is human-centered AI implementation?

Human-centered AI implementation is an approach to introducing artificial intelligence that prioritizes the people performing the work, not just the capabilities of the technology. It begins with understanding existing tasks, workflows, cognitive demands and opportunities for learning and collaboration. Organizations then evaluate how AI changes those conditions, including its effects on workload, employee well-being and human judgment. The goal is to use AI to improve both organizational performance and the quality of work, rather than focusing solely on efficiency or productivity.

How can employers prevent AI from increasing employee workload?

Employers can reduce the risk of AI-related work intensification by evaluating how the technology changes the entire workflow, rather than measuring productivity gains alone. Although AI may reduce the time required for certain tasks, it can also introduce new demands, including reviewing outputs, correcting errors and managing higher performance expectations. Organizations should monitor actual workload, involve employees in evaluating changes and avoid automatically replacing every minute saved with additional work. The goal is to ensure that efficiency improvements support sustainable performance rather than simply increasing work demands.

How can organizations maintain human oversight when using AI at work?

Organizations can maintain human oversight by establishing clear responsibilities for reviewing AI-generated outputs, verifying information and making final decisions. The level of oversight should reflect the potential consequences of errors, with higher-risk tasks requiring more rigorous verification. Employees also need sufficient time, training and authority to question AI recommendations rather than accepting them automatically. Effective human oversight is built into workflows and accountability structures, rather than relying solely on employees being told to “double-check” AI.

How can employers evaluate whether AI is improving the quality of work?

Employers can evaluate the impact of AI by measuring more than productivity, speed and output. A human-centered approach also considers changes in cognitive workload, employee well-being, opportunities for learning, collaboration and physical work demands. Organizations should gather employee feedback, assess how tasks and workflows have changed, and monitor whether AI introduces new demands alongside its benefits. The goal is to determine whether AI is helping people work more effectively and sustainably, not simply enabling them to produce more.

References

  1. Haipeter T, Wannöffel M, Daus JT, Schaffarczik S. Human-centered AI through employee participation. Front Artif Intell. 2024;7:1272102. doi:10.3389/frai.2024.1272102.
  2. Lee HP, Sarkar A, Tankelevitch L, Drosos I, Rintel S, Banks R, et al. The impact of generative AI on critical thinking: self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In: Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery; 2025. Article 1121. p. 1-22. doi:10.1145/3706598.3713778.
  3. Demerouti E, Nachreiner F, Bakker AB, Schaufeli WB. The job demands-resources model of burnout. J Appl Psychol. 2001;86(3):499-512. doi:10.1037/0021-9010.86.3.499.
  4. Su F, Beh WF, Cheah KSL. Efficiency or intensification? A systematic review of generative AI and academic workload in higher education. Acta Psychol. 2026;268:107353. doi:10.1016/j.actpsy.2026.107353.
  5. Tankelevitch L, Kewenig V, Simkute A, Scott AE, Sarkar A, Sellen A, et al. The metacognitive demands and opportunities of generative AI. In: Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery; 2024. doi:10.1145/3613904.3642902.
  6. Pachidi S, Berends H, Faraj S, Huysman M. Novice risk work: how juniors coaching seniors on emerging technologies such as generative AI can lead to learning failures. Inf Organ. 2025;35(1):100559. doi:10.1016/j.infoandorg.2025.100559.
  7. Dell’Acqua F, Ayoubi C, Lifshitz-Assaf H, Sadun R, Mollick E, Mollick L, et al. The cybernetic teammate: a field experiment on generative AI and teamwork. Organization Science. 2026. doi:10.1287/orsc.2025.20702.
  8. Parés-Salomón I, Señé-Mir AM, Martín-Bozas F, Loef B, Coffey A, Dowd KP, et al. Effectiveness of workplace interventions with digital elements to reduce sedentary behaviours in office employees: a systematic review and meta-analysis. Int J Behav Nutr Phys Act. 2024;21(1):41. doi:10.1186/s12966-024-01595-6.

EWI Works offers many services that can improve your quality of life. We have developed several cost-effective remote services to help you transition to remote work. Find out more about our Online Training, Services, and Resources. 

Follow us on Twitter, LinkedIn, and Facebook for more updates and offers, or sign up for our newsletter below.