What does the research show about students’ pair programming with AI? In Pair programming, one person acts as the “driver” and the other as the “navigator” as they work together to write a program. Research shows some learning benefits, although there can be some disadvantages as well.
With the advent of generative AI, it is now possible to replicate pair programming with one student working with an AI partner. But how does this approach compare to “traditional” pair programming? How do students feel about it? Because AI is new, there is not yet much research about the practice. That’s why we are conducting a research study focused on AI pair programming.
Our Approach
In our paper at ASEE’s 2026 annual conference, we report a systematic literature review for AI pair programming. Our research question was, What evidence exists regarding the outcomes of student-AI pair programming? 
We identified 10 relevant studies where students used AI as a pair programming partner.
Findings
Almost all of the studies involved college students. They used a variety of methods and approaches. Some studies analyzed interview transcripts, for example, and others gathered data via surveys. The most commonly used tool was ChatGPT. While findings varied, there was some evidence of positive learning outcomes; however, there was often less social connection when students partnered with AI.
Implications
This study confirmed that there is very little research on AI pair programming. While this is not unexpected, more research is needed to determine how AI pair programming impacts students.
For example, in traditional pair programming, when a partnership includes one student with more prior experience and one with less, it is sometimes the case that the less experienced student does not benefit as much from the practice. We might wonder whether and how this finding applies to AI pair programming: should the AI be prompted in such a way that it mirrors the students’ experience level? Or does the dynamic change when one of the partners is an AI tool?
There are other gaps in the research as well. For example, almost all studies were of post-secondary students. But since traditional pair programming is used with students as young as elementary school, it would be helpful to study AI pair programming with younger students.
We found that studies of AI pair programming identified some advantages for cognitive learning outcomes but had more mixed results for other outcomes, such as engagement. Explorations of how to capture the cognitive learning advantages of AI pair programming while avoiding other disadvantages is an important next step.
This study used AI search tools in addition to traditional literature search tools. We found that the AI tools yielded helpful results, but concerns remain about their replicability and transparency as part of the research process.
For more information about this literature review, you can read the full paper.
Future Work
The Institute for Advancing Computing Education is engaged in a three-year study to assess the impact of AI pair programming in high school computer science classes. You can visit the project’s website, AI Pair Programming, to learn more about this project.
Citation
Smith, J. M. (2026, June). Pair Programming with AI: A Systematic Literature Review. In 2026 ASEE Annual Conference & Exposition.
Acknowledgement
This material is based upon work supported by the National Science Foundation under Award No. 2524429. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.