Agency-Focused Reflection on AI-Assisted Writing

TLDR: Students feel most in control while editing AI text but the output may still reflect AI stylistic markers. One question (“Where did you feel the most agency?”) reframes how students think about AI-assisted work. What students experience as agency and what registers in the writing itself can diverge, and that gap is where the real teaching happens. What if the most transferable AI assignment can be the right reflection question?

What she’s doing: Stephanie teaches Writing Creatively in the Age of AI, an online asynchronous and in-person course that scaffolds students through increasingly complex engagements with generative AI from analyzing avant-garde literary precedents through prompt engineering exercises to fine-tuning chatbots and designing independent experiments. For each assignment, students submit their chat logs alongside a structured reflection. This semester, informed by her participation in the AI in Teaching & Learning Scholars cohort, Stephanie has sharpened those reflection prompts to center a single concept: agency. Rather than asking broadly about collaboration with AI, she now poses pointed questions: Where in the process did you feel the most agency? What were you doing in that moment? To what extent do you feel like the author or co-author of this piece? Why or why not?

What’s working and what isn’t: The agency questions surface something earlier framings didn’t. Students consistently report feeling the greatest sense of ownership during editing and revision—the moment they’re reshaping AI-generated material into something they recognize as theirs. Several have used the reflections to design ambitious final experiments. For instance, one multilingual student, Angel Cruz Davila, confronted his own tendency to “play it safe” when writing in English due to concerns about grammar and vocabulary. He pushed himself to write sensorially about learning English as a child, then used AI to interrogate how models flatten a multilingual voice. The reflections made visible a pattern he might not otherwise have named.*

But Stephanie has identified a persistent tension: even when students report strong feelings of control, their final outputs may still retain unmistakable AI stylistic markers. Felt agency, she’s finding, doesn’t always translate into writing that reads as solely human-authored. She describes reading student work after enthusiastic revision reports and thinking, “I’m not sure that matches what you think it does.” All of this raises a question: to what extent has AI “voice” already become part of how humans compose text, even when writing and editing independently with strong authorial control?

What’s next: Stephanie is developing follow-up questions designed to analyze the gap between felt agency and the degree to which AI stylistic patterns may remain legible in students’ writing. The next move is to ask students to identify where AI-inflected language remains legible in the text itself, and to identify what they, or AI, “smoothed” during iteration and revision that might have been distinctively their own. She’s also considering loosening scaffolding in her first-year writing course: simply inviting students to use AI however they want for part of an assignment, then reflecting on what happened. The driving focus is that structured metacognitive reflection can do more pedagogical work than prescriptive AI-use policies.

*Example and student name listed with permission from the student.

Have you tried something new with AI? How did it go? Send an email to [email protected] to let us know!