Large language model agents are increasingly being used to operate desktop applications, but the interfaces they encounter were built for human eyes and hands. A new arXiv preprint argues that this mismatch creates friction for agent-based automation, since existing desktop interfaces were designed for human end users or application developers, leaving agents to interpret them as best they can.

The paper, titled "A Decade in a Day: Synthesizing Agent-Native Interfaces from Human-Oriented Software Evolution," proposes a different approach: rather than waiting for software to gradually adapt to agents, synthesize interfaces that are native to agents by drawing on the evolution of human-oriented software. The title suggests compressing years of interface evolution into a single day, though the abstract does not detail the underlying mechanism.

The work highlights a growing research area at the intersection of human-computer interaction and AI. As LLM agents become more common as software users, interface design may need to shift from a purely human-centric focus to one that also accommodates autonomous agents.