Artificial intelligence research in dentistry has focused mainly on diagnosis, while the documentation, a major source of administrative burden, remains largely unexplored. This study aimed to identify the barriers to AI adoption in dental practice and to derive and preliminarily validate a clinician-first, non-decisional documentation architecture. A three-source needs assessment identified four barriers, which informed three design principles: the clinician remains the sole decision-maker, outputs are delayed to avoid interrupting the clinical judgement and no content is produced without a traceable source. The system was tested in a single-site pilot study (n=14 encounters). The mean documentation time was 1.86 minutes; the perceived workload was lower than underestimated standard workflow and clinicians edited the generated notes in 78.6% of the encounters. These preliminary findings support clinician-first ambient AI as a safe-by-design approach in order to reduce documentation burden with potential benefits for patient communication and clinical practice.