The Double and the Deserved Interlocutor: A Behavioral Model of Professional Agency in Human–AI Collaboration

Organizations measure the adoption of workplace AI by volume: licenses issued, active users, queries per day. Yet the returns on adoption — and its risks — depend on something these dashboards cannot see: the mode in which a professional works with the machine. We propose a behavioral model that makes this difference measurable and developable, resting on two axes — epistemic vigilance and proactive goal-setting — whose crossing yields four collaboration types: Co-Author, Executor, Skeptic, and Enthusiast. Theoretically, the model imports A. A. Ukhtomsky's account of the dominant and of the "Double" versus the "deserved interlocutor" into human–AI interaction: preference-tuned language models, with their documented sycophancy, are a technically perfect Double. We describe a multi-method assessment (situational judgment test, questionnaire, dialogue-derived behavioral signals), an operationalization of Vygotsky's zone of proximal development over dialogue text, and a development engine built on five design principles — including deliberate non-sycophancy and the requirement that individual results carry no career stakes, argued to be a psychometric condition rather than merely an ethical one. A validation program is underway; the paper states explicitly what is claimed, what is pending, and what is not claimed. Preprint; not peer reviewed. The measurement instruments described are undergoing psychometric validation.