The ongoing AI boom seems to permeate, and sometimes call into doubt, every aspect of almost everybody’s profession.

Wonder what on earth comes next? You’d do well to ask Gleb Tsipursky, Ph.D. He’s widely acknowledged for his expert reads on the future of work. In fact, the man dubbed “The Office Whisperer” by The New York Times has already written seven books and is about to release his eighth, The Psychology of AI Adoption at Work: From Resistance to Results.

Circumspect and painstakingly researched, Tsipursky’s writing is backed by more than 100 of his consulting projects, more than 50 interviews with leaders at top firms, and his 25-plus years of experience as a consultant.

Prior to publication, Tsipursky spoke with Quality Digest about the upcoming book and shared his thoughts on some of its salient points.

Quality Digest: Please tell us a little more about yourself and explain how you wound up living in Columbus, Ohio.

Gleb Tsipursky: I come from a Ukrainian family, and that heritage strongly shaped my interest in how people respond to coercion, uncertainty, and institutional change. I came to the United States for my education, earning degrees at New York University and Harvard before going to the University of North Carolina at Chapel Hill for my doctorate. I stayed at UNC as a lecturer after completing my graduate work.

My academic research focused on behavioral science and historical cases of persuasion and resistance, and I later brought those insights into consulting. Today, I lead Disaster Avoidance Experts from Columbus, helping organizations manage the human side of major workplace changes, including AI adoption.

QD: Your doctoral research examined how Soviet suppression of Western music in the 1940s and 1950s backfired. How did that affect your consulting philosophy?

GT: Soviet authorities tried to suppress Western-influenced music through censorship, punishment, and public denunciation. The repression often increased the music’s appeal because young people treated it as authentic, exciting, and forbidden. That research taught me that coercion may produce surface compliance while strengthening private resistance. In consulting, I apply the same lesson: Leaders should explain the purpose of a change, involve the people affected, demonstrate concrete benefits, and give them meaningful agency.

Persuasion takes more effort at the beginning. But it produces stronger commitment, better information from the front line, and more durable behavioral change than an order backed primarily by authority.

Persuasion takes more effort at the beginning. But it produces stronger commitment...

QD: Why is persuasion more effective than coercion in return-to-office decisions?

GT: Return-to-office mandates often fail because employees interpret them as a signal that leaders distrust them or disregard the practical benefits they gained from flexibility. A mandate can produce attendance without producing engagement, collaboration, or retention.

Persuasion works better when leaders identify which activities benefit from in-person work, use evidence to establish an appropriate schedule, consult employees about implementation, and explain the trade-offs honestly. Teams accept inconvenience more readily when they understand the business purpose and have some influence over how the policy works.

The same principle applies to AI adoption: Leaders gain stronger results when people understand the goal and participate in shaping the change.

QD: Is AI adoption fundamentally a leadership and behavioral challenge rather than a technological one? Has that always been your approach?

GT: Yes. The technology matters, but the decisive bottleneck usually appears after an organization acquires a capable tool. Employees must trust it, managers must redesign workflows around it, and leaders must create incentives and guardrails that support responsible use.

My psychology-first approach emerged from the work because I repeatedly saw technically sound initiatives stall through anxiety, unclear expectations, weak training, or management resistance. The pattern resembled earlier changes I studied and advised on, especially remote and hybrid work. The distinctive feature of AI lies in how directly it touches professional identity, judgment, and job security, which makes leadership behavior and organizational culture central to adoption.

Technical capability determines what an AI system can potentially do, while leadership and behavior determine whether people use it safely, consistently, and for valuable work. Leaders must decide which problems deserve attention, clarify accountability, create protected space for experimentation, invest in skills, and connect use to measurable outcomes.

Managers then translate those decisions into daily routines and respond when the tool fails. A company can buoy an excellent platform and still obtain little value if employees distrust it or managers discourage experimentation. Conversely, an organization with modest tools can achieve meaningful gains when its people understand the purpose, own the workflows, and improve them continuously.

QD: What is genuinely different about AI adoption, and which older change-management practices still apply?

GT: AI differs from Six Sigma, lean, ERP, and earlier automation because it can generate content, recommendations, and analysis in nearly every knowledge role while producing outputs that remain probabilistic and sometimes wrong. It also evolves so quickly that organizations can’t treat adoption as a one-time implementation.

The mistake lies in discarding proven change-management discipline because the technology feels unprecedented.

Yet the strongest older practices still apply: Involve users early, begin with specific problems, run controlled pilots, build cross-functional teams, train for real workflows, identify peer champions, standardize what works, and measure outcomes.

The mistake lies in discarding proven change-management discipline because the technology feels unprecedented. AI requires faster iteration, but it still succeeds through trust, participation, feedback, and operational rigor.

QD: What is the biggest blind spot leaders have when implementing AI, and what do they misunderstand about employee resistance?

GT: The biggest blind spot is assuming that resistance reflects ignorance, laziness, or hostility to innovation. Employees often raise legitimate concerns about job security, accuracy, workload, privacy, accountability, and whether leadership will use productivity gains primarily to reduce head count. Resistance therefore contains useful information about mistrust, unclear policies, and poorly designed workflows.

Leaders make matters worse when they launch generic training before answering those concerns. They should first diagnose the sources of resistance, distinguish among different employee mindsets, address credible risks openly, and demonstrate one or two visible improvements in daily work. People become more receptive when they see both honesty about the risks and evidence of personal value.

QD: What misconceptions do organizations have about employees’ fear that AI may replace them?

GT: Organizations often assume that job-loss fears come from misunderstanding AI and will disappear after a demonstration or training session. Employees usually understand the basic issue quite clearly: If AI reduces the labor required for important tasks, staffing and career paths may change. Leaders lose credibility when they offer blanket reassurance that nobody’s job will be affected.

A better response explains which tasks may change, which human capabilities will grow in value, how the organization will invest in reskilling, and how staffing decisions will be made. Employees can tolerate uncertainty more readily than evasiveness. Their resistance often reflects a rational demand for clarity about workload, status, accountability, and future opportunity.

QD: What are the early signs of psychological safety, and how can leaders encourage experimentation without tolerating poor performance?

GT: Early signs include employees openly reporting AI errors, asking basic questions without embarrassment, sharing unfinished experiments, and challenging an AI recommendation or a senior leader’s assumption.

Psychological safety doesn’t mean lowering standards. Leaders should create bounded experiments with approved data, defined use cases, required human review, and clear escalation rules. They should distinguish a good-faith learning error from negligence or repeated disregard of safeguards. Teams can then examine failures without blame while still holding people accountable for verification and judgment.

The strongest cultures combine permission to experiment with demanding quality controls.

The strongest cultures combine permission to experiment with demanding quality controls. Employees know they can speak candidly about a mistake, but they also know that hiding one or skipping required review carries consequences.

QD: What do your data show about AI’s biggest advantages and downsides for team knowledge-sharing and connection?

GT: AI can make expertise more accessible by helping employees retrieve internal knowledge, summarize complex material, translate specialized language, and learn from patterns previously concentrated among experienced staff. It can also support cross-functional collaboration by giving people a common starting point for discussion.

The main downside is that AI can spread weak or false information at great speed, especially when teams treat fluent output as verified knowledge. Overuse can also reduce direct conversation, weaken mentoring, and produce homogenized thinking.

Organizations gain the most when AI supports human exchange, rather than replacing it, with source-grounded systems, visible verification, expert review, and deliberate opportunities for people to interpret and challenge what the tool produces.

QD: You discuss ethical guardrails. What is ethical, and who defines it?

GT: No single executive, vendor, or ethicist should define organizational AI ethics alone. An ethical framework should begin with applicable law and professional standards, then incorporate the organization’s stated values and the interests of employees, customers, affected communities, and other stakeholders.

A cross-functional governance group should translate those principles into practical rules concerning privacy, fairness, transparency, human oversight, accountability, security, and prohibited uses. The process matters because people trust guardrails more when those affected have a voice in creating and revising them.

Ethics also requires ongoing judgment. New capabilities create new dilemmas, so organizations need review mechanisms that can update policies rather than treating a static checklist as the final answer.

QD: What role does middle management play in AI adoption?

GT: Middle managers often determine whether AI strategy becomes daily practice or remains an executive presentation. They translate broad goals into workflows, answer employee questions, approve experiments, allocate time for training, and decide how teams respond to mistakes.

At the same time, managers can become a major source of resistance because AI may threaten their expertise, authority, staffing levels, or even parts of their own role. My research identified five distinct middle-manager profiles, ranging from strong skeptics to highly enthusiastic catalysts. Leaders should therefore involve managers early, give them practical experience, clarify how their roles will evolve, provide relevant metrics, and channel both caution and enthusiasm through appropriate governance.

QD: How can companies amplify AI’s positive effects and counteract the negative ones?

GT: Companies should begin with narrow, valuable use cases that remove friction from real work, involve frontline experts in designing the workflow, and measure quality as well as speed. They should give employees approved tools, practical training, peer support, and a safe channel for reporting failures. At the same time, they need tiered controls based on risk, including data restrictions, human review, audit trails, and clear accountability.

Companies should begin with narrow, valuable use cases that remove friction from real work, involve frontline experts in designing the workflow, and measure quality as well as speed.

Leaders should reward useful outcomes rather than raw AI use, since pressure to use AI everywhere produces low-quality work and distorted metrics. The guiding model is augmentation: Let AI handle pattern processing and routine drafting while humans retain responsibility for context, judgment, relationships, and final decisions.

QD: Does AI resistance look different in highly regulated quality environments, and does your advice change?

GT: Resistance in medical devices, aerospace, food safety, and similar fields tends to center more strongly on traceability, validation, liability, data integrity, and personal accountability for a harmful outcome. That caution often reflects professional responsibility rather than backward thinking. The human principles remain the same, but the operating model must become more rigorous.

Employees experiment more confidently when they know the boundaries, review process, and escalation path.

Organizations should start with lower-risk uses, validate performance against established baselines, document data sources and revisions, preserve human sign-off, conduct failure-mode analysis, and involve quality, legal, compliance, security, and frontline specialists from the outset. Strong guardrails can increase adoption because employees experiment more confidently when they know the boundaries, review process, and escalation path.

QD: How does frontline distrust of AI tools for inspection, process control, and documentation manifest itself?

GT: Distrust often appears as quiet behavior rather than open refusal. Employees might avoid the tool, perform the entire task manually, and then enter the expected AI result, double-check every output so extensively that no time is saved, withhold useful feedback, or treat each error as proof that the system has no value.

Others may use unapproved tools because the official system feels cumbersome, creating shadow AI risk. Frontline teams could also fear that reporting a problem will make them appear resistant or incompetent. Leaders need visible error-reporting channels, rapid correction, transparent performance data, and evidence that human expertise retains authority when the tool conflicts with observed conditions.

QD: How can quality’s continuous-improvement culture become an AI adoption advantage?

GT: Quality teams already understand that a process improves through measurement, root-cause analysis, corrective action, standard work, and repeated learning cycles. That mindset fits AI unusually well because AI outputs require ongoing testing and refinement rather than one-time approval.

Quality professionals can help define acceptable performance, identify failure modes, compare results against baselines, document human review, and turn incidents into improvements. They also bring a useful skepticism to vague claims of transformation. When quality teams apply familiar plan-do-check-act discipline to AI, they make experimentation safer and more credible. That discipline helps the organization scale proven uses while stopping weak or risky ones before they become embedded.

Dashboard theater occurs when leaders count licenses, prompts, or active users and mistake activity for business value.

QD: How does measurement separate real AI value from dashboard theater, and how should organizations do it?

GT: Dashboard theater occurs when leaders count licenses, prompts, or active users and mistake activity for business value. Making AI real requires connecting a specific use case to an operational baseline and an outcome that matters. For inspection, that might include defect escape rates, false positives, rework, and cycle time. For documentation, it may include completion time, correction rates, audit findings, and user burden. Organizations should run bounded pilots, compare performance before and after deployment, track quality and risk alongside efficiency, and ask how saved time gets redeployed. They should also review results with frontline users, since a favorable average can conceal additional work, workarounds, or risks in a particular part of the process.

QD: Can you describe the field evidence from more than 100 organizations? Was it drawn from previous work or developed for the book?

GT: The book draws on more than 100 consulting projects conducted throughout my work with organizations adopting AI and other major workplace changes. I combined those field cases with research gathered specifically for the book, including more than 50 interviews with leaders and systematic analysis of employee and manager attitudes. For the psychographic profiles, I used more than 35 focus groups involving more than 250 middle managers, plus more than 90 focus groups involving nearly 600 nonmanagerial employees. I coded recurring themes, refined the profiles, and checked them against later client settings. So the evidence combines a long consulting record with a more structured research effort developed specifically to explain AI adoption.

QD: Have the viewpoints of leaders and other sources evolved as AI has developed?

GT: Yes. Early conversations often veered between excitement about productivity and fear of broad disruption. As organizations gained experience, the discussion became more operational. Leaders began asking less about whether AI mattered and more about workflow redesign, governance, data quality, manager readiness, and measurable returns.

Many enthusiasts became more cautious after encountering hallucinations, security concerns, and weak pilots. Some skeptics became more positive after seeing how narrow uses save time without compromising accountability.

The most important evolution has been a move away from treating AI as a tool purchase and toward treating it as an organizational capability that requires continuous learning, employee involvement, and regular adjustment as both the technology and the work change.

QD: As AI becomes better at processing information, what becomes more valuable about human expertise?

GT: Human expertise becomes more valuable in framing the problem, judging relevance, recognizing unusual conditions, understanding consequences, and taking responsibility for the final decision. AI can process vast amounts of information, but it does not reliably know which goal deserves priority, when a pattern fails to fit the situation, or how a technically efficient choice will affect trust, safety, and relationships.

Experienced people also possess tacit knowledge that rarely appears fully in a database, including contextual cues and lessons from rare failures. The strongest professionals will use AI to expand their reach while becoming more disciplined about questioning assumptions, verifying outputs, exercising ethical judgment, and communicating decisions to other people.

The Psychology of AI Adoption at Work: From Resistance to Results, from Georgetown University Press, comes out Sept. 1, 2026, featuring a foreword by Nick Bloom, the William D. Eberlie Professor of Economics at Stanford University. It’s available now for preorder; sample two free chapters here.