There is a comfortable story leaders tell themselves about employees who hesitate to use AI: they do not understand the tool, or they are simply resistant to change. However, the 2026 data on workplace AI suggests a more uncomfortable and useful explanation.
A Writer and Workplace Intelligence survey of 2,400 knowledge workers found that 29% of employees said they had sabotaged their company’s AI strategy. Among Gen Z employees, the figure was 44%. The study covered 1,200 employees and 1,200 C-suite executives across the United States, United Kingdom, France, Germany, and Australia.
The word sabotage is provocative, and this was a vendor-sponsored, self-reported survey rather than a direct measure of behavior. Still, the finding deserves attention. AI adoption resistance can reflect fear, misaligned incentives, unclear expectations, or legitimate doubts about quality and accountability. In other words, treating every form of reluctance as a training deficit hides those signals.
At the same time, the Writer study found pressure on both sides of the adoption divide. Sixty percent of surveyed executives said they planned layoffs for workers who could not keep up with AI. In addition, 92% said they were cultivating an “AI elite” class of employees.
When employees hear that adoption may determine whether they keep their job or qualify for leadership, hesitation stops looking irrational. It becomes a predictable response to a transition framed as a test of employability rather than an investment in capability.
A separate Prudential workplace-benefits study found another perception gap. Seventy-eight percent of employers viewed AI positively, compared with 51% of employees. Because that survey focused on AI-enabled benefits experiences in the United States, the figures should not be generalized to every workplace use case. However, the 27-point gap still illustrates how differently leaders and workers can experience the same technology.
Leaders often see an investment, a productivity target, and a transformation roadmap. By contrast, employees may see surveillance, deskilling, unclear liability, or a future reduction in headcount. Therefore, a rollout can be technically sound while remaining socially unstable.
Research from Harvard Business School professors Das Narayandas and Shunyuan Zhang offers a useful explanation for this pattern. They describe AI and similar forms of automation as self-disruptive technologies: innovations that can improve organizational performance while eroding the authority, discretion, or legitimacy of the people expected to approve them.
The researchers identify three identity threats. Role compression turns judgment-heavy work into monitoring. Control shift moves decisions to a model or a centralized team. Span erosion reduces influence over people, budgets, or processes.
As a result, a tool can work exactly as promised and still feel threatening to the person adopting it. For example, the employee may become faster while feeling less expert. A manager may gain better forecasts while losing the discretion that previously defined the role.
This is different from a usability problem. More training may improve skill, but it does not answer whether the organization still values the employee’s expertise. Nor does it explain who remains accountable when AI-generated work fails.
Similarly, project abandonment data needs careful interpretation. Gartner reported in 2026 that at least half of generative AI projects had been abandoned after proof of concept by the end of 2025. The cited causes included poor data quality, inadequate risk controls, escalating costs, and unclear business value. User trust is part of the operating environment, but it should not be presented as the sole cause of that abandonment rate.
The good news is that the research points toward practical interventions. In practice, access to AI tools alone does not produce durable adoption. Employees need a clear strategy, role-specific guidance, useful workflows, and managers who can discuss both benefits and risks.
Even so, Gallup’s midyear 2026 findings show how strongly those conditions matter. Frequent AI users with a clear implementation plan, active manager support, and strong engagement were nearly three times as likely to give AI the highest productivity rating as frequent users overall: 50% versus 17%.
Moreover, employees who frequently used AI and had active manager support were engaged at a rate 14 points above the national average. Gallup concluded that leaders should set explicit expectations, equip managers to coach AI use, incorporate AI into performance conversations, and track adoption alongside engagement.
Ultimately, trust grows when employees can see what the technology is for, where it should not be used, and how success will be judged. It also grows when people can raise concerns without being labeled resistant.
Inside engineering teams, reluctance is often misread as a productivity problem when it may be a quality signal. An engineer who slows down to verify an agent’s output is not necessarily behind on adoption. They may be protecting a system for which the team will remain accountable long after the demo.
Likewise, an engineer who opposes merging generated code without review may be applying sound professional judgment. For instance, AI can produce insecure dependencies, misunderstood abstractions, brittle tests, and code that appears plausible without fitting the surrounding architecture.
Leaders should distinguish among at least three kinds of resistance. One comes from unfamiliarity and can improve with training. Another comes from fear about status or job security and requires honest role design. A third comes from evidence about reliability, security, or maintainability and deserves technical investigation.
If leaders treat all three as noncompliance, they teach careful engineers to stop speaking up. Consequently, reported adoption may rise while actual system risk becomes harder to see.
State the workforce implications plainly. If AI is intended to augment roles, explain how responsibilities, evaluation, and career paths will change. If headcount reduction is part of the plan, vague reassurance will destroy credibility.
Involve engineers in selecting use cases. The people closest to the work know where automation removes friction and where it creates hidden review costs. Participation turns adoption into joint design rather than imposed compliance.
Preserve review and override authority. Teams need documented situations in which human judgment controls the outcome. In addition, leaders should protect the time required for verification rather than counting every review step as resistance.
Measure quality with speed. Adoption dashboards should include escaped defects, rework, security findings, maintainability, and incident impact. Usage volume alone rewards activity rather than value.
Make dissent actionable. Create a path for employees to report weak outputs, policy conflicts, and unsafe workflows. Then show what changed because of that feedback. Visible response is one of the clearest ways to build trust.
Reluctant AI adoption is not automatically a workforce failure to be managed away. It is a signal about whether a new tool arrived with the communication, safeguards, role clarity, and genuine involvement needed for people to use it responsibly.
Some resistance will be based on misunderstanding. Some will reflect anxiety. Yet some will be accurate professional judgment from people who understand the work better than the rollout team does.
The engineering organizations that gain durable value from AI in 2027 will not simply be those that mandated adoption fastest. They will be the ones that listened closely enough to separate obstruction from expertise, and then redesigned the work around what they learned.
For a related look at why human review remains essential as AI accelerates delivery, read The Quality Engineer’s Renaissance: Why AI ROI Depends on Human Oversight.
Healimpilo is a technology advisory firm working with companies across healthcare, fintech, life sciences, and SaaS on engineering strategy, talent, and delivery.
