Computational thinking and modeling is essential to promote students’ STEM+C learning.

CoSMoAgent is a knowledgeable, collaborative peer agent to enhance students’ computational thinking and coding activities. It works alongside students as they design, build, and debug computational models through contextual awareness.

CoSMoAgent provides a balance of emotional support and domain expertise while students build physics and computing models in the C2STEM platform. Designed to overcome the “one-size-fits-all” limitations of many AI tutors, CoSMoAgent operationalizes the theory-driven Evidence-Decision-Feedback (EDF) framework. It uses four specialized sub-agents that continuously analyze student action logs and dialogue, infer problem-solving strategies, and deliver highly personalized scaffolding. In a recent classroom study, CoSMoAgent demonstrated significant positive effects on learning dynamics. The agent successfully adapted its guidance by fading support over time. As students achieved higher task mastery, CoSMoAgent shifted from probing for basic understanding to pushing students to extend their reasoning. Crucially, students’ verbalized conceptual understanding strongly aligned with their actual performance, indicating true knowledge construction rather than superficial “gaming” of the system. Furthermore, students demonstrated increasing independence by relying less on the agent as their competence grew, effectively mitigating the risk of overreliance commonly seen with Large Language Models.

Central Goals of CoSMo Agent

• Provide Adaptive Scaffolding: The agent dynamically personalizes support by fading its guidance as students gain task mastery, smoothly shifting from probing foundational understanding to expanding on complex concepts.
• Operationalize Learning Theory: CoSMoAgent explicitly grounds its interactions with the student pair in established educational frameworks, such as Social Cognitive Theory and Social Constructivism, using the Evidence-Decision-Feedback (EDF) model.
• Foster Independent Problem-Solving: By adopting a collaborative peer persona, CoSMoAgent encourages inquiry and active knowledge construction rather than simply supplying direct answers, thereby preventing harmful student overreliance.
• Ensure Interpretability and Transparency: The system maintains a clear, traceable reasoning chain from raw student input data to the agent’s pedagogical decisions and feedback to build trust among teachers and stakeholders.

Promote Conceptual Understanding over Task Completion: CoSMoAgent ensures that a student’s progression through a learning task is meaningfully aligned with their ability to verbally articulate and reason about their computational models.