The uneven boundary between tasks an AI system does well and tasks it does badly, which often does not line up with what people find easy or hard. A model may draft a strong report yet fail a simple-looking puzzle. The danger is that the boundary is invisible, so people trust the system on exactly the tasks where it lets them down.
Technical view
The term comes from a September 2023 Harvard Business School working paper by Dell’Acqua, Mollick and colleagues, a field experiment with 758 consultants at Boston Consulting Group. On tasks inside the frontier, those using AI completed more tasks, faster and to higher quality; on a task designed to sit outside it, they were markedly less likely to reach the correct answer than those working without AI. The boundary shifts with each model generation and is not obvious from a task’s surface, which argues for task-level evaluation and calibrated trust rather than blanket adoption or rejection.
Related
Part of AI Terminology Index, section: Product and Adoption.