"AI will figure it out": System Compensation and Intent Deferral in Open-Ended Generative AI Interfaces

Abstract

Open-ended generative AI systems may be reshaping how cognitive effort is distributed before and during task execution. This work-in-progress paper introduces two interrelated concepts: system compensation, the system’s ability to generate coherent responses from underspecified inputs, and intent deferral, the user behavior of relying on the system to resolve missing task parameters. We argue that system compensation may be removing a critical calibration signal to the user, potentially shifting automation-like overreliance into the input construction phase of generative AI interaction. Three exploratory observational sessions with 23 participants reveal four recurring interaction patterns. An intervention introducing constructive friction, requiring explicit intent specification before a system response, provoked resistance, suggesting that system compensation shapes expectations of immediacy. We discuss implications for cognitive ergonomics and open-ended interface design.