The rapid integration of generative AI into higher education has created a distinctive institutional challenge: how to develop critically grounded AI capability across an entire academic workforce, including academics, professional services staff and Associate Lecturers, without bypassing legitimate concerns about academic integrity, expertise erosion, equity and pedagogical values. Existing approaches typically default to either enthusiastic adoption or cautionary awareness sessions. Neither adequately addresses the complexity of the institutional context or the diversity of the staff it seeks to serve.

This paper presents a framework for human-AI collaboration in learning and teaching, synthesised from two complementary approaches: the first is grounded in Critical AI Literacy (CAIL) and the Responsible by Design (RBD) principles developed at The Open University, which insist on structured scepticism as a precondition for adoption and direct connection between development and live curriculum work. The second draws on human-centred service design, foregrounding co-creation, psychological safety and the recognition that AI capability is an essential dimension of academic professionalism.

The framework operates across five design principles: psychological safety as the precondition for meaningful human-AI collaboration; disciplinary-specific development co-designed with subject leads; capability distributed through peer and community leadership including AI Academic Leads, Champions networks and promptathons; development time as productive time generating tangible curriculum outputs through the RBD Framework; and visible progress through an AI Pedagogical Use Tracker and performance dashboard reported to Academic Board.

A three-layer architecture covering Governance, Capability, and Curriculum and Outcomes aligns policy to UK Office for Students B3 conditions and monitors student continuation, completion and progression as primary impact indicators.

Drawing on the FAIESTA model, this paper offers a theoretically grounded, practically tested framework that is critical, inclusive and scalable, demonstrating that scepticism and adoption are sequential conditions and that the endpoint of human-AI collaboration is improved outcomes for a diverse student body.