Thinking-through-Writing
in the Age of GenAI

Reflective writing is central to knowledge work: it externalizes knowledge workers’ existing ideas, catalyzes new ones, fosters learning, and deepens understanding of the subject matter. But as LLMs increasingly automate text generation, the thinking and learning processes behind reflective writing are at risk of being automated or undermined, which can harm workers’ learning and long-term growth.

This project re-imagines what deep, reflective writing might look like in the age of LLMs. Two design-research questions anchor the work: How can designers rapidly prototype LLM assistance when its effectiveness for facilitating thinking is hard to evaluate quickly and objectively? How can we imagine beyond Clippy-style suggestions when offering writing-task-agnostic LLM assistance?

The project is ongoing. Below are some initial publications:

• How Problematic Writer-AI Interactions (Rather than Problematic AI) Hinder Writers’ Idea Generation (Umarova et al., arXiv’25)
• CoAuthor: Designing a Human-AI Collaborative Writing Dataset for Exploring Language Model Capabilities (Lee et al., CHI’24 🏅)
• Sketching NLP: A Case Study of Exploring the Right Things To Design with Language Intelligence (Yang et al., CHI’19 🏅)

This project is supported by the Schmidt Futures Foundation, Google, and the Cornell Center for Teaching Innovation. Yuqing Wu and Chelsea Shi are student leads. Mina Lee and Khonzoda Umarova were previous student leads.