AI / GOVERNANCE
ADOPTION → TRUST
TRUST → JUDGMENT
JUDGMENT → ACCOUNTABILITY
The problem is no longer who uses AI. It is who knows when to trust, when to question, and how much judgment they are willing to delegate.
Two out of three employees use what AI answers without checking it. More than half have already made a mistake because of it.
KPMG and University of Melbourne, 2025 · 32,352 employees in 47 countries
And you?
Find yourself. Mark what you do today, not what you plan.
Question 1 of 4 · Understanding
Could you explain to your board why AI can give a false answer with complete confidence?
Understanding: I think so
Confuses fluency with knowledge.
Understanding: I think so
Talks transformation without changing the work.
Understanding: Yes, I have done it
Turns answers into decisions without verification.
Understanding: Yes, I have done it
Leaves opportunities unused.
Understanding: No
Rejects a capability it does not understand.
Understanding: Yes, I have done it
Manages the limits of the technology.
Using AI no longer sets an executive apart. Knowing when not to believe it does.
THE SIX PROFILES
The six profiles are not a psychometric classification and do not come from a survey. They are a decision model for recognizing behaviors that are already observable. The desirable position is not to believe more, nor to distrust always: it is trust that changes with the task.
Believes AI can do almost anything and understands little about how it produces its answers.
Is surprised by the apparent quality of the result and confuses a convincing answer with a correct one. When AI gets it right, confidence rises; when it fails, the failure reads as an exception rather than a property that has to be managed. The problem is not using AI too much. It is attributing to it capabilities the person cannot evaluate.
Talks about AI, shares articles, asks for strategies, promotes transformation. The work itself stays the same.
For this profile AI is a narrative of innovation, not an operating capability. It is possible to believe in the technology and not have redesigned a single decision, workflow or responsibility around it. The pattern also shows up at company level: in the same McKinsey survey, only 21% of organizations using generative AI had fundamentally redesigned at least some workflows, and of 25 attributes measured, that redesign was the one most related to EBIT impact. Adoption is not transformation.
Already uses AI, gets useful results and looks advanced next to peers. Has delegated judgment before building the discipline to review what was delegated.
This is the profile that deserves the most attention. Not because it can be shown to be the most frequent among executives (the available studies do not allow that claim) but because it combines the variables that produce the largest impact: use, trust, authority and little verification. It matters when the person approves budgets, changes processes, selects vendors, evaluates people or sets direction.
A user’s mistake produces a bad presentation. A decision-maker’s mistake changes an organization. And delegation is already advancing: in a Wakefield Research survey for SAP (March 2025; 300 C-level executives at US companies with at least USD 1 billion in revenue), 38% would trust AI to make business decisions on their behalf, and 44% would override a decision they had already planned based on what AI shows them. Meanwhile Gartner (456 CEOs and senior executives surveyed between June and November 2024) found that 77% believe AI is opening a new business era, but only 44% consider their CIO to be AI-savvy. Those with the most authority are adopting fastest a technology their own organization still understands unevenly.
Understands the technology enough to recognize its limits, and uses it where the benefit justifies the risk.
Does not confuse generation with truth. Avoids delegating where consequence is high, the action is irreversible or the uncertainty is large. May look less enthusiastic, but the resistance comes from an evaluation, not from ignorance. The risk is staying too long in defensive mode and leaving unused the activities where AI already adds capacity under reasonable controls.
Does not use AI, does not trust it, does not understand what it could do. The conclusion came before the analysis.
The mirror of the copy-paster: both take a strong position on a weak base of understanding. One overestimates; the other dismisses. Both substitute conviction for judgment.
Does not believe AI by default. Uses it, understands it, verifies it and adjusts trust to the task.
For this profile a brainstorm does not require the same verification as a financial recommendation; summarizing is not deciding; a well-argued answer is not a true one; automating a reversible step is not delegating an irreversible action; and using AI does not transfer responsibility for the result. It is the behavior an organization should build: high use, sufficient understanding, constant verification, explicit human accountability.
BEFORE DECIDING
A spreadsheet can hold a wrong number. A dashboard can show bad data. But a generative model delivers a wrong conclusion with excellent language, logical structure, plausible arguments, a tone of authority, and no visible sign of uncertainty. That is why fluency and accuracy have to be separated in one’s head.
The less prepared user judges the answer by how it sounds. The decision owner asks five questions before turning it into a decision.
Not what the model recommends: which facts the recommendation is built on.
Force the contradiction, the counter-argument and the alternative scenario.
Control discipline rises with consequence.
The reversible admits experimentation. The irreversible demands more evidence and more authority.
AI generates, analyzes, recommends and executes. Accountability does not disappear behind the system.
THE EVIDENCE
Three years ago the question was who used generative AI. Today that question sets nobody apart. In McKinsey’s State of AI survey (March 2025; 1,491 participants in 101 countries), 53% of the executives surveyed used generative AI regularly at work, against 44% of middle managers.
The study by KPMG and the University of Melbourne (2025; 48,340 people in 47 countries) shows the other half: 83% expect benefits from AI, but only 46% are willing to trust it. Heavy use with limited trust would be healthy. The problem is the behavior in the opening: the risk sits with the person who uses it, trusts it, holds decision power and stopped verifying what comes back.
The region makes this discussion more urgent. According to the McKinsey Global Institute (July 2026; job postings in Argentina, Brazil, Chile, Colombia and Mexico between 2023 and 2025), postings requiring AI fluency rose nearly eleven-fold, most of them in non-technical roles.
Value capture is not moving at the same pace. The World Economic Forum and McKinsey study (January 2026) found that only 10% of Latin American organizations link their AI implementation to business strategy.
PwC Interaméricas shows the same tension from another angle in its 17th CEO Survey (February 2026; Central America and the Dominican Republic): 85% of employees say they used AI at work in the last year, while the CEOs who say they apply it to a great or very great extent are 20% in setting direction, 24% in products, services and experiences, and 14% in support services. The regional problem is not access. It is turning use into capability, capability into decision, decision into workflow, and workflow into economic result.
THREE FIGURES
85 · 78 · 51
leaders, managers and front line using generative AI several times a week or daily
BCG, AI at Work 2025 · 10,635 white-collar employees, 11 markets
23%
of Latin American organizations report any economic value from AI
World Economic Forum & McKinsey, January 2026
6%
report significant economic impact
World Economic Forum & McKinsey, January 2026
WHAT THE ORGANIZATION DOES
Recognize uncertainty, fabrication, insufficient evidence, conflict between sources, and the cases that require escalating to a person.
Write down when AI may generate, recommend, prepare, decide or execute. Not every task admits the same delegation.
A material recommendation answers what information it used, what it assumed, which controls it passed, who approved, and what happened afterwards.
A decision can be produced faster and worse. Productivity without error, rework and incidents is an incomplete metric.
Control intensity follows consequence, reversibility and authority, not whether something is called AI. And the control that validates what the model proposes has to be deterministic. That is the next piece.
THE QUESTION
The wrong question is whether you believe in AI. The relevant one is how far you understand what you are delegating, and how far you question an answer before turning it into a decision.
In a few years using AI will be infrastructure. What will keep setting people apart is formulating better problems, interpreting evidence, questioning answers and taking responsibility. AI multiplies productivity, and it multiplies bad decisions. The variable that decides which one happens is judgment.
José Ñáñez
SOURCES
Third-party surveys, cited at the level of their own population. None is a population estimate of all executives.