Agentic AI systems increasingly handle tasks that take minutes per turn, leaving users in a strange limbo: the wait is long enough that staring at the screen feels wasteful, yet short enough that walking away is impractical. A new arXiv paper examines this 'micro-wait' and the design challenges it creates.
The researchers conducted a formative study with 12 people who use agents daily. Based on that work, the paper argues that micro-waits are a distinct phase of interaction, not just a technical delay. The usual options—watch closely or leave—do not fit comfortably at this timescale.
The paper then outlines design considerations for these moments, suggesting that interfaces could offer lightweight ways to track progress or resurface results when ready. The work is preliminary, but it points to a growing usability gap as agents take on longer, more autonomous tasks.