‘Just a Prototype?’ AI-assisted artifacts and the practical work of assurance in research labs

GenAI coding tools are often discussed as individual productivity aids or as components of software production workflows. In research lab work, however, AI-assisted coding also produces collaborative artifacts: demos, prototypes, scripts, evaluation outputs, visualizations, and integration code that others inspect, discuss, reuse, or build upon. This position paper reports preliminary survey findings from an industrial AI research lab. Respondents described GenAI as helping make ideas concrete and discussable, for example by accelerating prototyping, early experiments, and visualizations. At the same time, they reported difficulties around the status of AI-assisted artifacts: fast prototypes can reorder work before needs are understood, demos can be read as closer to implementation than they are, and generated code can require additional checking, explanation, and coordination. Connecting these findings to classic CSCW concerns with boundary objects, articulation work, and accountability, we argue that assurance is not only a formal governance process applied after AI use. In research lab workflows, assurance is also practical work: making AI-assisted artifacts intelligible, qualifying their status, showing what has been checked, and limiting what others should infer from them. We conclude by briefly outlining follow-up interviews and artifact walkthroughs.