There is a difference between exercising a decision and knowing what the decision revealed. Most AI education programs for executives and board members do the first part well: they present cases, provoke debate, hand out a certificate of attendance. What is missing is the second part — someone who says, with criteria declared in advance, what each person did when facing the case and what that means when the support leaves the room.
Three questions
The first is retention. Not what the person knew at the end of the day, but what they can still do weeks later, without the material at hand. The second is transfer: whether what was practiced in one case shows up in another, different one, with another sector and another table. The third is autonomy: whether the person does it without the AI at their side — and knows when to call it back.
None of the three is answered by a satisfaction survey. All three are answered by observation: a case built to force real choices, criteria written before application and a test taken with the AI switched off, after an interval.
Why this matters now
A study with students measured the effect of learning with AI without safeguards: performance 48% higher with the tool on and 17% lower with it off — below those who never used it. The group that learned with an AI designed to teach, not to solve, had no drop. The lesson holds for decision-makers: fluency with the AI on can hide a dependency that only shows up when it leaves.
What we do with this
In Discernia's AI-Ready programs, each participant decides under time pressure and pushback, facing agent-characters with their own role, information and objectives. The debriefing describes what the person did — not who they are — and returns strengths and hypotheses, never prescriptions. And what stayed is measured by the LUMEN standard, published openly, with the AI out of the picture.
- Criteria declared in advance, not inferred afterward.
- Observed behavior, not self-report.
- A test with the AI switched off, after an interval.
We do not predict performance and we do not pass judgment on the person. We measure what is left — because that is what the seat will demand when the meeting is real.
- Bastani, H. et al. “Generative AI without guardrails can harm learning”. PNAS 122 (2025).
- Discernia. LUMEN standard — position paper v4 (Jul 2026), open regime.
Feel free to quote — with the source.