JMIR Publications

ABSTRACT

Large language models are increasingly being used within and alongside therapy, shifting the key question from whether therapists remain in the loop to what they are there to do. We argue that the enduring therapist role in AI-assisted care is best understood through a framework of three overlapping domains: relational functions, adaptive functions, and accountability functions. These include therapeutic challenge, use of the therapeutic relationship as a mechanism of change, rupture detection and repair, bearing witness to suffering, calibration of pace and treatment burden, and clinical judgement under uncertainty across the broader care pathway. Drawing on clinical psychology, digital mental health, and therapist skill development frameworks, including Bennett-Levy’s declarative–procedural–reflective model, we argue that the therapist functions most likely to remain human-led are those that depend most heavily on reflective, perceptual, relational, and contextual expertise. We suggest that this has important implications for workforce development, supervision, training, and service design in near-term AI-assisted therapy.