Key Stage 5 teachers: Join our new research study!
We’re launching a new study to explore developing a pedagogical framework and teaching resources for using LLM tools in programming tasks. The study will take place in collaboration with A-level/Scottish Highers teachers (or equivalent) in England, Scotland and Wales, and look at:
- What students need to know about how LLM tools work before using them in programming tasks
- How to safely and effectively use LLM tools for help-seeking requests in programming tasks
- How students can develop AI interaction literacy skills to understand and critically evaluate outputs from LLM tools
Our goal is to find practical ways to help teachers teach programming safely, ethically and responsibly with LLM tools in Computer Science lessons. The study will be collaborative, with two in-person workshops in 2026 and 2027, and we’re inviting A-level/Scottish Highers (or equivalent) teachers to join us.
You can apply to participate by filling in this form:
Why is this research important?
Large language models (LLMs) are already shaping how many learners encounter programming, both in and beyond formal education. A 2025 position paper established why young people should still learn to code in the age of AI, and in our research studies we’ve been exploring what LLM-assisted programming means for teaching and learning. This work aims to develop an evidence base to support educators and policymakers navigating rapid change. In this blog, we give an overview of the questions driving the work and the patterns we’re beginning to see.
In an ongoing project, we are conducting a series of studies to explore how LLMs can support learning to program in text-based programming in secondary schools. We are working with teachers of students aged 11–18 who are developing foundational programming skills, and we are thinking carefully about the guardrails which need to be in place to support ethical, safe and responsible use of this new technology. Our focus is on the pedagogical approaches for teaching and learning, rather than evaluating how effective a particular tool is. We aim to generate insights that will remain useful even as AI tools continue to evolve, and that will be directly relevant to classroom practice.
A sociocultural approach to using LLM tools in the classroom
At this point in our research project, we have completed two studies. In our first study, we asked secondary teachers to evaluate the usefulness of AI-generated explanations of programming error messages (PEMs). When novice learners encounter a bug in their program, they often find it difficult to decipher the PEM and get help to fix the error. This can lead to learner frustration and hinder progress, and so creates an opportunity to explore whether LLM-generated explanations can help learners understand the PEM. Eight expert secondary computing teachers participated in activity-based interviews, where they gave their views on LLM-generated feedback about a PEM in a piece of Python code. Teachers’ views correlated to feedback literacy, a set of competencies that support the social interaction of feedback in the classroom. For example, educators preferred the LLM-generated explanations to guide students and help them develop understanding, rather than telling students the answer. In this paper from the study, we concluded that in the sociocultural context of a computing classroom, supporting students and teachers to practice their feedback literacy skills matters as much as the content of the PEM explanations.
Our next study has also been conducted with secondary computing teachers, but this time we have begun to introduce LLM tools into the classroom, with careful consideration for teacher and student wellbeing and responsible use. We originally intended to focus on using LLM tools to explain PEMs when students were writing programs. However, we saw that in real-world settings, teachers and students were keen to leverage the additional capabilities of the LLM tools. For example, students asked for help to understand chunks of code, and to get ideas for how to write functions. Findings from this study are due to be published later in 2026, but we’re already seeing that in the sociocultural context of a computing classroom, LLM tools are not neutral – they can enrich or diminish learning. This highlights the importance of understanding how LLM tools interact with classroom norms, teacher practices, and learner expectations, not just individual students’ use of the technology.
Adding to the evidence base on using LLMs in programming education
The research conducted at the Raspberry Pi Computing Education Research Centre is carefully designed to contribute towards the wider conversations happening in computer science education research. A recent literature review highlighted the importance of teaching students how to use LLM tools, including the limitations of current tools. In our 2024 research seminar series, we heard from Majeed Kazemitabaar about his research using generative AI with students aged 11–17, and how their teachers should support novice programmers to write better quality prompts to produce better code snippets and suggestions. Our research seeks to contribute to this growing conversation by adding empirical evidence from secondary classrooms in the UK.
For educators and policymakers, this research matters because students’ use of LLM tools is already a reality, often developing faster than formal guidance or policy. Robust evidence is needed to inform decisions about classroom guidance, curriculum design, and assessment approaches. Our early findings suggest that the educational impact of LLM tools in programming depends less on their presence and more on how they are introduced, framed, and used within teaching and learning. As more personalised forms of artificial intelligence develop, such as agentic AI, understanding their role in learning—rather than reacting to the technology itself—will become increasingly important.
Thoughtful decisions about the use of LLMs depends on robust, independent evidence rather than assumptions or headlines. Through this research, we aim to support educators and learners as they navigate a rapidly changing landscape, grounding discussion about AI in what is safest and most useful for classrooms. We’ll be sharing further findings as our work continues, and we invite educators, researchers, and policymakers to follow and engage with this evolving evidence base.
