Proactive Thought Partner

A proactive thought partner is an AI agent that offers higher-level cognitive support during an open-ended task, such as writing, by choosing its own moment to intervene rather than waiting to be asked or filling in the next few words like autocomplete. Chao Zhang, Abe Davis, Chih-Wei Chen, and Chin-Chia Hsu introduced the term in "Designing Proactive Thought Partners for Writing" (arXiv:2609.01588, September 2026), work Zhang carried out during an internship at Google DeepMind with Cornell University collaborators. The paper distinguishes this from existing proactive writing tools, which mostly provide generic textual assistance such as predicting the next word or sentence, by focusing instead on support aimed at a writer's thinking process: generating ideas, noticing gaps, or prompting reflection on what has already been written.

The authors built a technology probe, a working prototype meant to surface design insights rather than ship as a finished product, that let users configure one or more "partners" by assigning each a role and a proactivity level. They deployed it with 16 participants for one week of real writing. As a person wrote, the configured partners judged when to take the initiative and surface a suggestion, rather than responding only to explicit requests.

The study found that participants set up their partners prospectively, anticipating situations they expected to encounter rather than reacting after being interrupted. They used the suggestions for both generating new ideas and for self-monitoring their own writing, a use case proactive tools are rarely designed to support. How intrusive a suggestion felt depended more on its presentation, such as a lightweight visual cue and non-directive phrasing, than on its timing alone. The authors derive design implications for proactive writing assistants organized around customization, timing, engagement, and representation.