Across a large number of conversations with executives leading AI strategy at their organizations, one specific and somewhat unusual practice recurs disproportionately among the leaders whose organizations appear to navigate AI decisions most effectively: they maintain a running, written log of AI-related predictions they've personally made, along with the actual outcomes once they became known.
Why this practice is unusual
Very few executives in any domain systematically track their own past predictions against outcomes - it's uncomfortable, time-consuming, and there's no external requirement forcing it. The leaders who do it anyway are making a deliberate choice to hold themselves accountable in a way their role doesn't otherwise demand.
What this practice actually produces
Over time, a genuine record of one's own calibration - which kinds of predictions tend to be reliable, which kinds tend to be systematically overconfident - that's considerably more specific and personally actionable than general awareness that forecasting AI's trajectory is hard. It converts a vague humility about uncertainty into a concrete, personal track record.
Why this seems to correlate with better AI decisions
Leaders with an honest personal track record of where their own AI predictions have gone wrong appear to bring more calibrated confidence to new decisions, weighting their own instincts appropriately rather than either dismissing them or over-trusting them based on incomplete self-awareness.
Why this is a practice more leaders could adopt
This requires no special resources or authority - just the discipline to write down a prediction when it's made and revisit it honestly later. It's a low-cost habit that the leaders doing it consistently credit with improving their own judgment over time, and one that seems underused relative to how simple it is to start.