Agents in programming educationConnecting learner readiness, collaboration and learning design

Research question

How should a programming course organize collaboration between students and an AI agent, and what evidence would show that the arrangement helps learning? This question connects the agent’s educational function with the activity students complete and the outcomes a course values. The studies in this guide approach it through classroom comparison, learner characteristics, instructional environments and a review of agent roles. Reading them together suggests defining who receives support, what the agent contributes and how learning is assessed. These dimensions provide a practical structure for designing a study of agent-supported programming and for locating the relevant literature.

Classroom evidence

Haoming Wang and colleagues directly examine AI-agent-supported collaborative learning in university programming. Their study describes a quasi-experimental comparison involving 45 undergraduates in an ACM programming competition teaching context over six weeks. The AI-supported group showed stronger achievement, self-efficacy and interest. Mental effort was lower, while the difference in mental load was not significant. These outcomes make the paper relevant to a study that compares collaborative arrangements and evaluates several dimensions of the learning experience. They also supply a concrete reason to report the dimensions of cognitive load separately when discussing the findings.

Learner characteristics receive closer attention in Jing and colleagues’ quasi-experimental study of ChatGPT-assisted Python graphing. AI literacy, programming knowledge and understanding of ChatGPT were associated with effective problem solving. This paper is useful when selecting baseline variables or explaining why the same available tool may be experienced differently by learners. Read alongside the collaborative-learning study, it motivates a specific follow-up: does an agent’s contribution depend on the programming knowledge and AI literacy students bring to the group? A project addressing that question could record these characteristics before instruction and examine patterns in the subsequent learning activities.

MetaClassroom broadens the design perspective from an agent’s immediate assistance to the organization of the learning environment. Its quasi-experimental comparison examines an immersive programming environment based on three-dimensional learning progression and reports improvements in achievement, self-regulated learning, beliefs and motivation. The paper is useful for connecting programming tasks with a sequence of learning goals and assessment. In a proposed agent study, that perspective would encourage a clear account of the surrounding lesson: what students are expected to learn, how they progress through tasks and how the agent’s contribution relates to those goals.

From agent roles to learning design

Liu and colleagues’ systematic review helps name the contribution an educational agent is expected to make. It identifies cognitive-epistemic, self-regulatory, affective and human–AI interaction roles. A researcher could use this vocabulary to distinguish an agent that invites students to explain a solution from one that prompts progress monitoring or supports engagement. The classification is helpful when comparing interventions that all use the label “AI agent” yet assign different educational functions. For a programming project, the proposed next step is to connect the intended role to observable student activity and to an assessment that fits the function.

An illustrative study could organize collaborative debugging around student explanations. Teams would state their interpretation of an error, request agent input and justify which suggestions they adopt. Researchers could examine the resulting solutions, the explanations exchanged and later individual work. This is a design proposal developed from the studies’ complementary focus on collaboration, learner readiness and regulation. It would allow the research question to concern a particular instructional arrangement: how an agent’s participation relates to students’ explanation and evaluation of code. Documenting the task, agent instructions, interaction opportunities and comparison condition would also make the intervention easier for other researchers to interpret.

Research directions

The papers have distinct uses in an introduction or methods rationale. The collaborative-learning study supplies a university classroom comparison; the ChatGPT study motivates attention to learner characteristics; MetaClassroom illustrates the integration of learning goals and outcomes within an environment; and the agent review provides categories for describing educational functions. Bringing these contributions together supports a focused research agenda: investigate how a defined agent role operates within a particular programming activity for learners with documented starting capabilities. Achievement, self-efficacy, interest, regulation and cognitive-load dimensions can then be selected according to that question. This reading route makes the rationale for each measure and instructional choice visible.