Qualitative analysis of design workshops depends on a researcher's trained judgment: reading behaviour in the room and distinguishing genuine engagement from distraction or confusion. We investigate whether AI can take on part of that analytical load without fabricating what it cannot observe. The work draws on six design workshops run to test a startup's new service during an MSc thesis, with aims that included ideation and problem definition. Across the project we used AI at several stages: to help design the workshop games, to plan the sessions, and to analyse what happened in them. Recurring tasks were encoded as reusable AI “skills” with fixed formats and a project memory. We report the analysis skill as the worked example. Its first version clustered all information into gains, frictions and facilitation feedback, which flattened the quality and complexity of insights, so it was rebuilt to carry the observer's reading of behavioural cues and to hold new clusters, including one for what did not fit anywhere, the way researchers leave room for the unknown. This cut analysis time from days to hours and organised the material more like a researcher would. We close with a critical account of what worked and what did not across these uses of AI, what we recommend to other designers, and what we would avoid if we did it again.
We are excited to invite you to the ACM GreekCHI-CYCHI Symposium 2026, held on November 6-7, 2026 jointly in Athens, Thessaloniki, Nicosia and Online.