Half Your Workforce Claims AI Expertise. Your Upskilling Strategy Is Built on That Lie.

There is a number sitting inside your last employee survey that deserves far more alarm than it's getting. According to a recent INTOO survey covered by HR Executive, 52% of employees call themselves AI experts. That same survey found those workers are simultaneously asking for basic guidance on how AI should be used for work-related tasks. Both things cannot be true — and the contradiction is the problem.

Self-reported expertise is not expertise. It is confidence, which is a different variable entirely, and one that is far more dangerous to build a transformation program around. When organizations design AI upskilling initiatives using employee self-assessments as their baseline, they are effectively letting the least-calibrated voices set the floor. The result is training that is too advanced for most of the room, too elementary for the few who actually know what they're doing, and operationally useless for everyone.

The harder question is why organizations keep making this mistake. Part of the answer is that measuring real AI capability is uncomfortable — it implies judgment, and judgment implies accountability. It's easier to run a survey, report the high confidence numbers to the board, and call the culture "AI-ready." The gap between reported readiness and operational readiness is where initiatives quietly die.

This is not purely a training design problem. It is a taxonomy problem first. A workplace behavior expert, speaking to HR Dive, warned that AI upskilling initiatives are making a consistent error: they focus on software integration while discounting human skills and the contributions of older workers. The framing of "AI expertise" has become synonymous with tool proficiency — knowing how to prompt a model, navigate an interface, generate an output. That definition excludes the judgment layer almost entirely: knowing when to trust the output, when to push back on it, and when to escalate the decision to a human who has context the model cannot access.

This matters structurally. MIT Sloan Management Review has argued directly that companies do not have to slash jobs because of AI — the choice is a design decision, not an inevitability. But that argument only holds if organizations can demonstrate genuine human value-add at the judgment layer. If your workforce's AI "expertise" is entirely surface-level tool use, the economic case for retaining headcount over automating it collapses faster than the HR team can write the talking points.

The workers most likely to be undervalued in this framing are exactly those the HR Dive expert flagged: older employees with deep domain knowledge who can catch what a model gets wrong, and who have spent careers developing the contextual judgment that no prompt-engineering course replicates. HR Dive (2025) reported that training initiatives are systematically failing to account for this cohort. Organizations that flatten AI readiness into a single confidence score will miss this, not because the data is unavailable, but because the questions being asked are too shallow to surface it.

The board conversation version of this is simple: if your AI transformation plan is calibrated to employee self-assessments, you do not have a transformation plan. You have a morale exercise with a technology budget attached. Boards are increasingly asking whether workforce AI readiness is a real operational capability or a number that makes the CHRO's slide look good. The honest answer, in most organizations, is that nobody has run the test to find out.

What would running the test actually look like? It means defining AI competency not as tool familiarity but as a set of decision-quality outcomes — can this person identify when an AI-generated output is wrong, incomplete, or missing context? Can they use AI to improve the quality of a judgment call rather than replace it? That kind of assessment is harder to administer than a survey, which is precisely why it keeps getting skipped.

The 52% figure is not a success metric. It is a diagnostic. The organizations that treat it as the former will spend significant resources training a workforce that already believes it doesn't need training. The ones that treat it as the latter will have a real conversation about what capability actually means — and build something that holds when the model is wrong, which it will be, and often.

The question your operating committee needs to answer before the next planning cycle: what specific decision quality improved because of your AI program, and how do you know?