AI-Mediated Teaching Turns Educators into Message Monitors, Erasing the Relational Core of Teaching

Summary

Early AI tutoring deployments (e.g., Eedi/Google DeepMind trial across 10 UK schools) place teachers in the role of approving AI-generated feedback before it reaches students. Tutors approved 94.7% of drafts unchanged, and while time was saved, teachers reported the work was far less rewarding than writing feedback or teaching students directly. This is an emerging, unsolved pain point: edtech is optimizing for throughput (time saved, message volume) without measuring or protecting the parts of teaching that make the job meaningful — direct student interaction and pedagogical judgment. The result is a quiet degradation of teacher satisfaction layered on top of existing burnout: educators do more surveillance/quality-checking of machine output, absorb accountability for AI errors, and lose the relational feedback loop that motivates them. No product or policy framework exists to define which instructional work should never be delegated to AI, nor to measure and protect teacher satisfaction alongside efficiency gains — so schools risk 'solving' workload while deepening disengagement and attrition.