<p>The integration of autonomous agents (AAs) into safety–critical domains requires effective Human–Autonomy Teaming, where operator trust is essential. However, trust must align with the AA’s capabilities to avoid misuse or disuse. This study examines trust calibration mechanisms in a longitudinal setting, addressing a gap in research that largely relied on independent situations experiments. We investigate how the AA’s dynamic transparency and performance feedback influence operator’s compliance. Sixteen maritime experts completed a simulated predictive maintenance task involving 60 interdependent decisions to accept or reject recommendations from an AA with 90% reliability. The AA adapted its transparency (information on its reliability or decision consequences) based on operator’s real-time compliance. After each decision, participants received feedback, including notifications of the AA’s errors. Results indicate that average compliance (90.7%) was well calibrated to the AA’s reliability but concealed inter-individual variability. Analysis of non-compliance showed that 68.5% of refusals were triggered by two factors: reduced transparency emphasizing risk, and feedback on prior AA’s errors. Clustering analysis identified three participant profiles: (1) “under-trust,” characterized by active system probing; (2) “risk-accepting and over-compliance,” marked by low sensitivity to risk information; and (3) “attitudinal over-trust and under-compliance,” defined by strong reactions to negative cues. These findings highlight trust calibration as a dynamic process shaped by the AA communication and experiential learning, but moderated by operator profiles. The main contribution lies in identifying these compliance strategies, suggesting that an adaptive AA should be profile-aware, tailoring transparency and feedback to support appropriately calibrated trust.</p>