AI-augmented supervision in shared chat spaces: how transparent, collaborative engagement with generative AI supports domain-specific critical engagement in project-based learning

The rapid integration of generative AI in higher education has often remained opaque, with students and instructors engaging AI privately and without shared accountability. This study examines AI-augmented supervision in shared chat spaces, designed to support domain-specific critical AI evaluation within project-based learning. An exploratory qualitative-dominant two-phase mixed-methods design was implemented in an undergraduate AI and business course at a private university in Lebanon (N = 26; third-year undergraduates aged 20–22; 15 females, 11 males). Phase 1 (weeks 3–7) used shared instructor–student–AI chat threads for individual supervision; Phase 2 (weeks 8–11) added rotating domain-related peer groups. Data included 26 chat transcripts (~600 coded dialogue moves), project revision traces, and a 25-item post-intervention survey. Transcripts were analyzed through interaction analysis and thematic coding (Cohen’s κ = 0.82–0.89). Findings reveal a supervision cycle of constraint setting, AI generation, critical evaluation, and student synthesis. Student-initiated evaluative moves rose from 18% to 38%, and peer participation was associated with substantive revisions in 11 of 26 projects (42%). Survey responses indicated high perceived transparency and educational value alongside bounded, context-specific trust. The study contributes evidence for triadic supervision design functioning as both supervision space and accountable decision log.