Designing Proactive Thought Partners for Writing
Chao Zhang and colleagues at Cornell, with work done during an internship at Google DeepMind, study what proactive AI should actually do during writing and deploy a probe with 16 participants for a week.
Ask this paper
The design space is the contribution. Existing proactive writing tools mostly do generic textual assistance such as autocomplete, and this work asks what higher-level cognitive support looks like when the agent chooses its own moment to intervene.
Users configure partners rather than receive one. The probe lets a writer create partners by setting their role and their proactivity level, and relevant partners then take initiative at moments they judge appropriate.
Participants planned prospectively. They configured support in advance for situations they expected, rather than reacting to interruptions, which is a usage pattern most proactive designs do not anticipate.
Suggestions served two distinct purposes, idea generation and self-monitoring, and the second is rarely designed for.
Intrusiveness was managed by presentation. Participants valued lightweight visual representation and non-directive rhetorical framing, so how the intervention is phrased mattered as much as when it arrived.
Abstract
Writing involves diverse cognitive activities, from ideation to revision, and writers' needs vary across individuals and moments. Proactive AI promises to provide the right support at the right time, yet existing proactive tools largely focus on generic textual assistance, such as autocomplete. This paper studies the design space of proactive thought partners: AI agents that proactively offer customizable, higher-level cognitive support during writing. We instantiated this concept in a technology probe and deployed it with 16 participants for one week. The probe allows users to create partners by configuring their roles and proactivity. As users write, relevant partners take the initiative at appropriate moments to offer suggestions. Our findings show that participants configured proactive support through prospective planning, used suggestions for both idea generation and self-monitoring, and valued lightweight visual representations alongside non-directive rhetorical framing for non-intrusive interventions. We derive implications for designing proactive writing assistants around customization, timing, engagement, and representation.