Giving constructive feedback in a difficult conversation is a clinical teaching skill that benefits from rehearsal, not guidance alone. UBC Faculty Development experts and EdTech are co-designing an AI Feedback Simulator to help practising clinicians develop feedback-delivery skills. Clinicians speak or type feedback to an AI learner who follows an authored case and can respond defensively, resist the feedback, or receive it well. They then receive a coaching debrief tied to the case goal and feedback framework. This is a work in process. At an initial Faculty Development planning session, participants shaped the need, learner behaviours, guardrails, modality, and feedback resources. A technical prototype now runs in our Oracle APEX test environment behind institutional sign-on. It supports case authoring, live voice and reflective text practice, persistent multi-turn sessions, and debrief generation. It is not in production. The next phase is formal case refinement and evaluation with practising clinicians. The next co-design phase is intended to define a small set of difficult learner-feedback conversations, credible learner responses, and what strong feedback delivery looks like. Early evaluation will ask whether the AI holds the authored behaviour, whether the debrief helps clinicians reflect on how they deliver feedback, and what it takes to author and govern a case. This lightning talk will share the prototype, user feedback and impressions, early co-design process, and emerging division of work. Faculty Development experts shape the educational goals, cases, and feedback criteria. EdTech translates those into an authorable rehearsal environment and supports implementation and evaluation. Production use requires faculty validation, privacy review, funding, and a service owner.