Gartner: only 40% of CIOs guard against AI dulling staff judgment

A sales strategy built from customer feedback reaches a manager's desk, polished and plausible. The manager asks what the conclusions rest on. The employee who submitted it can't say, because an AI tool did the analysis.

According to BigGo Finance, Gartner offers that scenario as an example in a recently published report titled "Building a Workforce That Doesn't Surrender Its Judgment." The report's main argument is blunt: once AI is part of the work, strong output can no longer be trusted as proof that the workforce is capable. Looking ahead to 2030, Gartner expects 30% of organizations to find at least one core role in which AI use has weakened workers' ability to make decisions on their own.

The reason is scope. Gartner says generative AI now handles analysis, synthesis, problem-solving and decision-making as well as drafting, calculation and search. Two employees can hand in reports that look the same while only one of them can explain how the conclusion was reached. If a team judges competence mainly by the quality of what gets submitted, Gartner's finding suggests that measure now needs a second check, and the CIO who rolled out the tools is likely to be asked who owns it.

The CIO survey behind the 40% figure

Gartner ran the survey from November to December 2025. The sample leans toward one region: 250 respondents came from North America, 150 from Europe and 133 from Asia-Pacific. Turned around, the headline number means 60% of the senior IT leaders surveyed had no strategy for stopping AI use from wearing down employee capability.

With North America making up nearly half the respondents, the 40% figure is best read as a snapshot of large enterprises weighted toward North American practice. Even so, it describes companies with at least $50 million in annual revenue, the ones most likely to have rolled out AI assistants widely already.

Why plausible answers are the real risk

Gartner sees the risk as something bigger than tasks being automated away. When a tool keeps producing plausible answers, people feel less need to check them. Gartner calls this "cognitive surrender": accepting an answer because it looks right and handing over both the reasoning and the accountability along with it.

The mechanism feeds itself. According to Gartner, each satisfactory result makes employees more likely to undervalue direct review or independent thinking. The better a tool performs, the faster people lose the habit of checking it. Gartner treats critical thinking as a core skill that decides whether a workforce is ready for the AI era. That framing suggests the issue belongs in the AI rollout plan, not just the HR training catalog.

There's an awkward side effect for operations leaders. While AI takes over the tasks that used to build judgment, the amount of output that needs human review can actually grow. So the review load can climb just as the skills needed to do the reviewing get weaker.

Four ways people use AI at work

Gartner sorts AI use into four types according to how involved the human is:

1. Lazy delegation: minimal instructions, a quick glance at the result.
2. Accidental hits: the answer meets expectations, but the user can't explain its reasoning or limits.
3. Expert orchestration: the human designs the context, constraints, evaluation criteria and verification steps.
4. Direct execution: the human does the analysis and makes the call, and AI handles side tasks like editing or formatting.

Look at the second item. An accidental hit and real expertise can produce exactly the same deliverable, and nothing on the page tells them apart. The difference only shows when someone asks why. That's what makes Gartner's sales-report example useful: the failure isn't in the report at all.

The list also shows that heavy AI use isn't the problem. Expert orchestration may involve more AI than direct execution, and it still keeps judgment with the person. For managers, the better question is which of the four modes a given role actually runs in. The report defines the modes by behavior and doesn't offer a self-assessment tool, so working that out falls to whoever reviews the work.

Useful friction, built into metrics and tools

Gartner's answer is what it calls "useful friction." Before adopting an AI answer, employees check the evidence behind it, look at the counterarguments and can explain the reasoning. One concrete step it suggests: require people to verify facts personally before a high-importance AI-generated report or analysis goes to a supervisor.

Gartner pairs that with a warning about measurement. Evaluating software developers solely on code volume or commit counts, it says, may push activity metrics up while neglecting verification and understanding of the output. That is the sales-report problem translated into a dashboard.

A metric that counts output could reward lazy delegation and accidental hits as generously as expert orchestration, because all three ship the same volume. For engineering managers, the practical consequence is that a review rubric would need some measure of whether a developer has checked the code and can explain it, alongside how much of it got committed.

The second skill Gartner names is "stewardship." It means adjusting how much human involvement a task needs while making sure people keep ownership of the purpose, the evaluation and the outcome. Gartner also takes the idea to tool design. It recommends that AI assistants stop handing over answers straight away and instead question the request's premises and constraints and raise counterarguments.

This puts verification somewhere new. In a June 2024 Q&A, Duncan Harris, a director in the Gartner HR practice, said organizations can ease employee concerns about job loss by offering training on topics that include how to evaluate AI output for biases or inaccuracies. The new report goes further. It suggests verification belongs inside the workflow and the software, where it no longer depends on someone remembering a training module.

Harris's 2024 advice already sketched some of the plumbing. He listed inaccurate AI, meaning tools that create incorrect or unfair insights, among five employee fears, and recommended that companies co-create solutions with employees and test them for accuracy. He also pointed to companies that had deputized AI ethics representatives at the business unit level to oversee AI policies within their departments.

A departmental role like that could be one answer to the ownership question the new report raises. The framing has shifted, though. In 2024, Gartner treated AI inaccuracy as a fear to calm so employees would trust the tools. In 2026, it treats the employee's ability to catch inaccuracy as a capability that can erode and needs protecting.

Where friction slows the work down

The fair objection is speed. The stewardship framing is the report's answer: friction should match the stakes. That means heavy checking for a report that shapes a sales strategy and little for a reformatted slide. Teams that apply friction everywhere will lose the speed they bought the tools for, while teams that apply it nowhere end up with the 30% problem.

The same judgment gap on the deployment side

A column by Optimove, published in MarTech and Search Engine Land, points to a Gartner prediction from June 2025: more than 40% of agentic AI projects will be scrapped before the close of 2027. Gartner based that forecast on a poll of over 3,400 organizations that are actively investing in the technology. According to Gartner senior director analyst Anushree Verma, most of these projects are early-stage experiments or proofs of concept, driven mostly by hype and often misapplied. That suggests the judgment problem also shows up on the deployment side.

Two more Gartner figures in the column bear directly on how IT teams buy. Gartner has warned about a practice it labels "agent washing," in which vendors repackage the chatbots and automation tools they already sell as agentic AI. It estimates that of the thousands of vendors claiming agentic solutions, only about 130 offer real agentic features. A further Gartner prediction holds that one-third of companies will damage customer experiences in 2026 by rolling out AI too early.

Read next to the judgment report, the agent-washing estimate suggests useful friction applies to procurement too. A plausible product pitch deserves the same follow-up question as a plausible report: what does this actually rest on?

Optimove pushes the argument past Gartner's design recommendation. In the column's view, an agent can optimize the inputs it receives but has no way to ask whether those inputs were the right ones to begin with. The column calls those limits structural rather than something the next model release will solve. Gartner's report, on the other hand, treats questioning premises as a behavior that assistants can be designed for. For buyers, the difference between the two positions matters. If Optimove is right, checking premises remains a human job no matter how the assistant is configured.

The column also cites a Gartner prediction that GenAI use leads to the atrophy of critical thinking skills and that, as a result, 50% of global organizations will require AI-free competency evaluations.

That gives the next signal something to measure against. Watch whether AI-free assessments start appearing in hiring and promotion reviews for analyst and engineering roles. That would mean companies have stopped trusting the deliverable alone.