Understanding human-diary agent collaboration through an activity-theoretical framework

Abstract Background: Conversational AI systems based on large language models (LLMs) are increasingly used for reflection, planning, and everyday decision-making, emerging as potential instruments for human-agent collaboration. However, how to ensure relevance and quality, and collaborative behaviour of such systems is an open question, particularly in health-related contexts. Objective: The objective was to identify design requirements for developing an LLM-based collaborative diary agent supporting reflection and planning through dialogue grounded in a person’s activities and motives, which could serve as an instrument for diary studies in the future. Methods: An activity-theoretical framework was used for interpreting the collaborative diary construction activity. A participatory design methodology was applied involving six occupational therapy experts. Data were collected in co-design sessions through interviews and observations of prototype use, and data was analysed using inductive content analysis. Results: The findings highlight the importance of grounding, personalisation, explainability, emotional safety, ownership of diary content, and adaptive interaction in human-agent collaboration. The experts identified potential roles for the diary agent in revisiting experiences, supporting reflection, and planning everyday activities. The study also identified tensions between reflection and planning, structure and flexibility, and guidance and autonomy, interpreted as contradictions within the activity-theoretical framework. Conclusion: The findings suggest that diary-agent design should support the person in actively influencing and questioning generated narratives rather than simply automating narrative generation. Grounding, explainability, user control, and adaptability emerged as important requirements for future human–diary agent collaboration.