Making AI annoying on purpose: when helpful tools don't always help

This design-based study examined how four high school students used a constrained AI writing system comprised of teacher-delimited AI chatbots designed to ask questions rather than generate text to build argumentation skills. In a nine-week unit, the TRACE model (Target, Refine, Assess, Cycle, Extend) oriented interactions through specialized chatbots that promoted productive friction by requiring students to defend and refine their thinking. Data included writing samples, AI interaction logs, weekly reflections, and interviews analyzed through thematic analysis. Analysis showed substantial changes in argumentative reasoning as students evolved from passive AI consumers to strategic evaluators, actively pushing back on AI outputs. They developed "prompt literacy," crafting targeted requests and critically evaluating responses, while strengthening counterargument development and evidence integration. These findings challenge recent criticism of AI supported writing instruction and claims that AI's generative potential expands invention and play. In contrast, this study demonstrates that constraint, not unrestricted generativity, was associated with increased self-regulation and sustained engagement with argumentative contexts. Strategic AI limitations made sustained cognitive engagement unavoidable, embedding repeated evaluation into everyday composing decisions through productive friction between tool assistance and intellectual autonomy.