Jagged Frontier of AI

Definition

A concept articulated by Ethan Mollick describing the irregular, non-linear boundary of artificial intelligence capabilities, where a model may excel at highly complex tasks while simultaneously failing at seemingly trivial or adjacent ones.

Key Characteristics

• Non-Intuitive Competence: Human intuition about task difficulty fails to predict AI performance (e.g., writing sophisticated code vs. basic spatial counting).
• Dynamic Shifting: The frontier continuously expands and contorts with new model updates, prompt strategies, and tool integrations.
• Necessity of Empirical Exploration: Practitioners cannot guess capability from first principles; they must directly test the model against specific domain workflows.

Applications

• Studio pedagogy: teaching students to probe and map model boundary limits through iterative experimentation.
• Task delegation in agentic workflows.

Mentions in Source

• “Tasks of seemingly identical difficulty may fall wildly inside or outside current model competence.” — Co-Intelligence