Recent advances in generative AI have rapidly transformed knowledge work, yet prevailing interaction paradigms still frame effective human-AI collaboration primarily as a matter of crafting better prompts. This framing of optimizing a single prompt risks reducing the role of human judgment and authorship, particularly when AI outputs appear convincing. Recent work has therefore proposed alternative frameworks that structure collaboration as an iterative cycle of human oversight rather than prompt optimization alone. One such framework, FERC (Frame-Explore-Refine-Commit), conceptualizes human-AI interaction as a governed sequence of metacognitive functions: framing intent and constraints, exploring alternative possibilities, refining the solution space through evaluation, and committing to accountable decisions. To translate the FERC framework into an actionable interaction method, this paper introduces FERC-Bot, a conversational software system that embeds the FERC cycle directly into a generative AI interface. Crucially, this operationalization evaluates process quality (i.e., human contribution/authorship) rather than output quality. Across four design iterations, FERC-Bot progressively evolved from surface-level scoring based on word counts and keywords toward context-sensitive phase classification and phase-specific authorship metrics. We evaluated the alignment between human-judged authorship and each FERC-Bot version using selected WildChat interactions and adversarial TrickyBot interactions, designed to exploit the FERC-Bot scoring by using knowledge of the scoring metrics to score highly while providing limited authorship. Results show that the fourth version achieved almost perfect agreement with human FERC-phase classification and moderate alignment with human authorship ratings across FERC phases. These findings provide a preliminary validation of FERC-Bot as an operationalization of the FERC framework. Importantly, the scoring metrics still require refinement before the FERC-Bot can serve as (1) a productivity interface for AI-assisted work, (2) a metacognitive scaffold supporting reflective collaboration practices, and (3) an empirical platform for studying structured human–AI interaction.
Suggested citation: Rindle, M., Kjeldsen, F. B., Berg, M., Rafner, J., Kenett, Y. N., Goecke, B., Weiss, S., & Sherson, J. (2026). From framework to interface: FERC-Bot as a metacognitive training system for governed human-AI collaboration [in press]. 5th International Conference on Human and Artificial Rationalities, Paris, France. https://www.researchgate.net/publication/415081302