Background As generative artificial intelligence (AI) enters health professions education, learner competence includes partnering effectively with AI for self-directed learning. This pilot study examined how AI-supported environments influenced engagement, learning behaviors, and competency development. Methods Nine first-year medical students were randomized to: (1) independent study without AI, (2) access to a grounded AI learning agent, or (3) access to it plus structured AI competency instruction. During a two-week extracurricular module, participants studied untaught hematology content. The agent applied metacognitive, adult-learning, coaching, and assessment principles to support planning, active learning, and formative self-assessment. The structured group also received instruction in the CRAFT prompting framework, exemplar prompts, and the agent’s pedagogical design. Participants completed study trackers, post-study surveys, and a content-aligned multiple-choice assessment. Analyses were descriptive and hypothesis-generating. Results Assessment performance was not consistently associated with group assignment, study time, AI exposure, or self-rated AI competency. Across groups, higher-performing learners used retrieval practice, reasoning, feedback seeking, knowledge-gap identification, adaptive learning, and multimodal strategies more consistently. Survey responses varied in prior AI experience, trust, verification, prompting practices, and educational roles assigned to AI, suggesting differences in learner–tool fit despite identical AI resources. Conclusions Effective AI-assisted learning may depend less on access than on the competencies learners bring to AI-supported environments, including metacognitive regulation, critical evaluation, adaptive prompting, and active engagement. These findings support a broader conception of learner competence in AI-shaped education and offer a framework for future curriculum design and research.