Redesigning Dynamic Assessment for Human-AI Collaborative Writing: A Design-Based Study

In AI-assisted writing, GenAI is typically used to provide direct performance support, reflecting dominant non-dynamic assessment (N-DA) practices and potentially decoupling writing performance from writing development. Such risk can be intensified in Human–AI Collaborative Writing (HACW), where GenAI contributes to meaning-making and text construction beyond tool assistance. In response, we introduce Dynamic Assessment (DA) to enable graduated, dialogic, and contingent mediation in HACW. Adopting a design-based research (DBR) approach, we designed and iteratively refined a DA-informed GenAI agent over three cycles, an expert review and two classroom implementations with Chinese EFL learners (n=74), using a non-DA agent as a comparison condition. Data included written essays, open-ended questionnaires, and interviews. Writing outcomes were analysed using ANCOVA, and qualitative data through reflexive thematic analysis. Across the three cycles, six implementation challenges were identified, leading to five design features that were synthesised into a three-layer DA-based design framework that positions GenAI as a mediational participant in HACW. Following iterative refinement, the redesigned DA-informed agent demonstrated more promising learning outcomes than the non-DA agent, providing preliminary evidence of the educational promise of the proposed design. This study advances a human-first design approach for enabling learners to collaborate with GenAI without compromising writing development.