Co-Author: A Methodology for Developing Agency in Work with AI — Link to CETA and a Protocol for Empirical Testing on the ERIDA Platform

Performance in work with generative artificial intelligence does not, by itself, show whether a person is developing the ability to set tasks, check answers and make decisions independently. This article continues the Co-Author model, introduced in the preprint The Double and the Deserved Interlocutor: A Behavioral Model of Professional Agency in Human–AI Collaboration. Keeping the model’s two axes – proactive goal-setting and epistemic vigilance – we specify it as a methodology for developing observable ways of acting in concrete tasks. We propose a learning cycle: the learner’s own goal, an initial attempt, targeted support, the learner’s action and explanation, gradual removal of support, and a separate test of transfer. The quality of the joint output, independence in managing the interaction and acquisition of the skill are treated as distinct. The link to Cognitive-Emotional Trajectory Analysis (CETA) adds a further research level: analysis of sequences of observable evidence in dialogue and of their relation to subsequent independent action. CETA factors are not equated with the Co-Author axes, and transition analysis does not replace independent measurement of outcomes. The ERIDA platform is considered as the intended environment for testing the methodology in multilingual adult learning in Switzerland. We describe the units of observation, a preliminary rubric, the separation of learning from assessment, the stages of the study and the conditions under which the hypotheses would be refuted. The article does not claim proven effectiveness of the method, validity of automated profiling or a completed integration of the research loop. Its contribution is an operational link between a developmental model, trajectory analysis and a testable educational application. Methodological article and proposed research protocol. Preprint; not peer reviewed. No new empirical results are reported. Version 1.0.