Overview
Jon’s AI Fellowship explored how generative AI can be embedded into teaching in a way that strengthens, rather than replaces, students’ learning and responsibility. Across two related projects, he focused on using AI to scaffold reasoning, reflection and feedback within a programming course, with carefully designed restraints to support pedagogical aims and academic integrity.
Key work and evidence
A time-aware AI course tutor
Jon developed a course-specific Socratic AI tutor for his Python data science course, fed on his course materials and teaching philosophy. Built as a Claude Project, the tutor uses clear protocols to support different learning scenarios (including assessment support, coding challenges, and catch-up sessions) and is explicitly time-aware, helping students understand where they should be with their learning at different points in the course.
Structured prompts in XML enables the AI to distinguish between different types of course information, improving accuracy and ensuring responses are aligned with the course context. This tutor was first deployed in Autumn Term 2025/26 within one course of around 90 students and has since been adopted on two other courses, reaching around 140 students during Winter Term.
A proactive AI bot
Across the academic year, Jon has been developing an AI tutor bot that contacts students to offer timely learning support. Building on the knowledge base of the original course bot, the tutor aims to offer differentiated learning paths to students at scale, with prompts informed by learning literature and research on cognitive offloading.
The bot diagnoses where students are in their learning at different points in the course and offers tailored resource, such as bespoke quizzes and flashcards, to reinforce course content in ways that align with students’ preferred learning approaches. The bot is designed to counter the isolating effects of working individually with chatbots by acting as a connective ‘social glue’ to connect students, instructors wider course context.
Following a pilot in Autumn Term, the bot was implemented more formally in the Winter Term in a selected course.
Key learnings and next steps
A follow-up study of Jon’s time time-aware AI course tutor is underway analysis chatlogs and version histories of students’ work. Using his GENIAL methodology, Jon is cataloguing and quantifying the different how students use AI, including where it supports or hinders learning and grade outcomes. This work will inform an academic paper and a policy paper setting out recommendations for designing AI-resilient open-book and project-based assessments.
In parallel, Jon’s chatbot approach has been adopted by another AI Fellow, Dori Sallai, for her Management course, with documentation in development to support wider replication and adaptation by other academics.