Jagged AGI is a term for the state in which an AI model performs at the level of human experts on many cognitive tasks yet fails in unpredictable ways on others. It was proposed in 2025 by Ethan Mollick, a professor at the Wharton School, and reflects the view that artificial general intelligence arrives unevenly across domains rather than as a single event.
The term extends the concept of the jagged frontier, which describes the uneven boundary of AI capability, to the discussion of AGI. Where the jagged frontier refers to the shape of that boundary, with some tasks handled easily and others not at all, jagged AGI refers to a model that has such a boundary while approaching general intelligence overall. It is used to stress that where an AI is strong and where it is weak can only be learned by using it.