ABSTRACT With artificial intelligence (AI) and automation embedded in most aspects of life from healthcare to transportation, this article conceptualizes how increasing automation will benefit or inhibit the therapeutic encounter. Five levels of a model are outlined, from the human analogue to the fully autonomous virtual (AI) therapist. The levels are referenced against key elements of the therapeutic encounter known to contribute to an effective therapeutic outcome, from the working alliance, to collaborative elements, to a confidential setting. Losses and gains are discussed through these and other aspects, such as the neurobiological, and the capability to notice what is not there in the process as well as what is, around the client’s aim of change or awareness. The objective is not to pit human therapists against non-human systems, but to propose a human-centred framework for the integration of AI into clinical reality, by introducing a taxonomy model of human–machine interaction (HMI), to see how each contributes, and to evaluate areas of risk and harm. Questions are raised around ethics, therapeutic process and crucially the very nature of what it is to be in a therapeutic encounter for a human client together with a human or AI therapist.