AI Transformation Is a Trust Test Disguised as a Technology Project
Executive Summary
The quality of an organization’s AI strategy may depend less on the sophistication of its technology than on whether people are willing to tell leaders what is actually happening.
AI transformation creates an unusually high demand for candor. Employees need to be able to admit when they do not understand a tool, challenge outputs that appear wrong, question assumptions behind a new process, and say when a supposedly improved workflow is producing worse results.
This requires more than technology access, training, or executive sponsorship. It requires trust.
McKinsey’s 2025 workplace research illustrates the problem. C-suite executives estimated that only 4% of employees were using generative AI for at least 30% of their daily work. Employees put the figure at 13%—more than three times as high.
The specific numbers will change as adoption accelerates. The management implication is more durable: the people closest to the work are often discovering where AI creates value, where it fails, and where formal processes no longer reflect operational reality before senior leadership can see those changes clearly.
The leadership question, therefore, is not simply:
Are our people adopting AI?
It is:
Have we built an environment in which people will tell us what is actually happening as they adopt it?
In AI transformation, earned candor is not a cultural luxury. It is part of the operating infrastructure.
AI transformation is also an information problem
Most organizations understandably approach AI transformation as an adoption challenge. Do employees have the right tools? Have they been trained? Are teams using the technology? Are workflows changing? Are productivity gains beginning to appear?
These are necessary questions. But they are incomplete. As I’ve written elsewhere, AI transformation is fundamentally a leadership challenge, not simply a technology implementation challenge.
Underneath those adoption questions sits a more consequential management question:
What information is actually reaching the people making the decisions?
Consider two organizations deploying broadly similar AI capabilities.
In the first, employees can tell leaders that an output is impressive but unreliable, that a redesigned workflow is faster but producing lower-quality work, or that customer behavior is diverging from what the implementation team expected. They can say they do not understand why a model is behaving a certain way. They can challenge an assumption embedded in the transformation plan.
In the second organization, people gradually learn which messages are rewarded. They demonstrate enthusiasm, report adoption, minimize uncertainty, and quietly work around systems that are not functioning as intended.
Both organizations may be able to show positive adoption metrics.
Only one has built a reliable learning system.
The information gap is already visible
McKinsey’s 2025 workplace research offers a useful example of how easily leaders can lose visibility into what is happening on the ground.
C-suite executives estimated that only 4% of employees were already using generative AI for more than 30% of their daily work. Employees reported 13%. The study covered 3,002 U.S. employees and 118 C-suite executives.
That is more than a modest forecasting error. Employees were more than three times as likely to report extensive AI use as executives believed.
The number itself will become less important over time. The leadership lesson will not:
The people doing the work may know important things that senior leadership does not yet know.
They see where AI is unexpectedly valuable and where it consistently disappoints. They know which formal processes are already being bypassed, where human judgment remains indispensable, and which roles are beginning to change before the organization has formally recognized the change.
This is precisely the kind of intelligence executives need during a transformation. But access to it cannot be assumed.
People will surface uncomfortable information only when they believe doing so is both useful and reasonably safe.
Candor is an operating capability
This is why I think of earned candor as infrastructure.
Leaders do not create candor by announcing an open-door policy or encouraging people to “speak up.” They create it through repeated management behavior that demonstrates that inconvenient information is valuable.
Teams notice what happens when someone challenges an executive assumption. They remember whether a failed experiment is treated as useful learning or becomes an exercise in assigning blame. They learn quickly whether leaders genuinely want disconfirming evidence or merely want objections acknowledged before the original plan proceeds unchanged.
Over time, those signals determine the quality of information that moves upward through the organization. We can already see the consequences when that channel breaks down: recent AI adoption data shows what happens when employees quietly work around a strategy rather than openly challenge it.
AI raises the stakes because the technology introduces new forms of uncertainty into everyday work. Its usefulness varies considerably by task and context. Outputs can range from exceptional to mediocre to confidently wrong. Implementation affects workflows, expertise, decision rights, professional status, and, in some cases, perceptions of job security.
All of this creates pressure to appear more certain than the situation warrants.
Senior leaders face a version of the same pressure. Boards expect an AI strategy. Competitors announce new capabilities. Investors anticipate productivity improvements. Employees hear that transformation is urgent. In that environment, acknowledging that an organization is still learning can sound insufficiently decisive.
Yet in a fast-moving technology transition, the opposite may be true.
Strong leadership is not the performance of certainty. It is the ability to create enough clarity and trust that the organization can discover what is true faster than its assumptions become obsolete.
Better questions produce better information
One of the most consequential things an executive can do during transformation is change the questions being asked.
Rather than asking only, “Is the rollout going well?”, leaders can ask:
“What are we learning that we did not expect?”
“Where are people working around the official process?”
“What appears successful in the dashboard but less successful in practice?”
“Where is AI improving speed while degrading quality?”
“What are people reluctant to tell us?”
These questions change the quality of the conversation. More importantly, repeated consistently, they change what the organization understands leadership actually values.
The signal becomes clear: accuracy matters more than reassurance.
AI has the potential to increase the intelligence available to an organization dramatically. But that advantage matters little if the human system filters out the information leaders most need to see.
The organizations that learn fastest may not be those with the most sophisticated models or the most ambitious transformation programs.
They may be the ones that build the strongest connection between the people doing the work and the people making the decisions.
In practical terms, that means creating an organization where reality can travel upward without being edited along the way.
Sources
McKinsey & Company, Leaders underestimate employees’ AI use, March 4, 2025.
McKinsey & Company, Superagency in the Workplace: Empowering people to unlock AI’s full potential, January 28, 2025.