Thinking is changing shape.

Peer reviewedPPIG 2026London

Interruptive Thinking: An Empirical Study of Operator-Side Control in AI Interaction

Sudhanshu Garg · Proceedings of the 37th Annual Workshop of the Psychology of Programming Interest Group

People who work with large language models routinely stop the model and redirect it. This autoethnographic study codes 611 such interruptions from 29 of the author's own working sessions, and describes a six-stage interruption cycle along with where the data does not fit it.

Three findings follow. The three AI models used to code the sessions disagree substantially about what counts as an interruption. User actions are described with 172 different labels, yet six recur far more than the rest. And the moments that most needed an interruption were often those where the model's output was most fluent.

@inproceedings{garg2026interruptive,
  author    = {Garg, Sudhanshu},
  title     = {Interruptive Thinking: An Empirical Study of
               Operator-Side Control in AI Interaction},
  booktitle = {Proceedings of the 37th Annual Workshop of the
               Psychology of Programming Interest Group (PPIG 2026)},
  address   = {London, UK},
  year      = {2026},
  url       = {https://ppig.org/papers/2026-ppig-37th-garg/}
}