Hands-on use (level 2) shows you have tried AI tools but they are not integrated into your regular practice. Calibrated trust (level 2) is at the stated-habit level without task-by-task differentiation or named failure cases from your own experience. Fit judgement (level 3) is generic rather than workflow-specific. Agentic supervision (level 3) shows some configuration but the autonomy-boundary reasoning is light.
The risk is that organisational decisions about AI scope, risk thresholds, and vendor selection are being made without the calibration that comes from regular contact with the technology. Closing this gap is straightforward in mechanism (daily use across two or three workflows over a quarter) and high-leverage in effect.