What a teacher cannot delegate: a pedagogy-first framework for integrating conversational AI agents across the simulation cycle

The benefits of simulation-based learning are well established, but they depend on the quality of each phase of the cycle: preparation, enactment, and debriefing make demands of different kinds, and the better each is met, the more a student draws from the simulation. This Curriculum, Instruction, and Pedagogy article presents a framework for meeting those demands with conversational AI agents under pedagogical direction. Its principle is that the teacher's reading of what each phase requires, not the capability of the technology, decides what is deployed, where, and whether. Agents built for each phase, holding the epistemic stance the phase demands and the knowledge of the course they serve, extend what a teacher can do rather than substitute for it. The framework specifies the conditions under which this is done responsibly, with agents operating within the institutional environment, under supervision structurally visible to students, and with interaction data remaining under institutional governance. It also specifies where integration must stop. The fiction contract, the commitment to psychological safety, and the terms of assessment are constitutively non-delegable, because their force depends on someone who can be held to them. The framework is illustrated through a complete deployment: five teacher-authored agents across one undergraduate simulation cycle whose simulation carries summative weight, a condition that sharpens rather than dissolves the boundary. The deployment is a pilot; no data were collected or analyzed for research, and the observations reported are the teacher's. The framework, not the case, is the contribution.