Computing education has long advanced through the design of tools: from early program visualizers and automated feedback systems to modern online programming environments and learning platforms. Today, generative AI tools represent another major turning point. Large language models and related AI technologies make it possible to generate explanations, hints, feedback, code, and learning materials dynamically and at scale. These capabilities create exciting opportunities for supporting learners in more adaptive and responsive ways. At the same time, they raise serious questions. AI-generated support can be incorrect, misleading, or pedagogically unhelpful, and students may not always be able to recognize when this happens. As a result, designing AI-integrated tools that are grounded in sound pedagogy has become an urgent interdisciplinary challenge.

This Dagstuhl Seminar focuses on the design, evaluation, and responsible use of AI-integrated tools for computing education. AI-integrated tools include, for example, conversational tutors, feedback tools, adaptive problem generators, automated assessment systems, and other intelligent support systems embedded in computing learning environments. Such tools are rapidly proliferating, often driven by new model capabilities, but the field still lacks shared conceptual foundations, common terminology, robust evaluation approaches, and clear reporting standards. In particular, there is limited consensus on how pedagogical goals and learning theories should inform the design and behavior of these systems. Without these, research risks becoming fragmented and difficult to compare, reproduce, or build upon.

The seminar aim to bring together researchers and practitioners from computing education, human-computer interaction, learning sciences, and AI in education to address a set of guiding questions, e.g., How can AI-integrated tools support learning without displacing productive struggle or critical thinking? What pedagogical and design principles help balance automation with learner agency, personalization with transparency, and technical capability with pedagogical intent? How should these systems be evaluated? Through plenary discussions, tool demonstrations, and focused breakout sessions, we will focus on the following interconnected themes. (1) Conceptual foundations and taxonomies: what counts as an AI-integrated tool, how different forms of AI integration can be categorized, and how new systems relate to earlier generations of educational tools. (2) Design principles and pedagogical frameworks, including scaffolding, feedback design, explainability, and learner engagement. (3) Evaluation and research methods, with attention to shared reporting practices, theory-informed study designs, and methods that combine quantitative and qualitative evidence. (4) Ethics and responsible AI in educational contexts, including privacy, bias, accountability, authorship, and sustainability. (5) Infrastructure, reproducibility, and maintenance, including how tools can be sustained beyond pilot studies and connected to shared platforms, datasets, and research ecosystems.

We expect the seminar to contribute to a shared vocabulary and taxonomy for AI-integrated tools in computing education, pedagogically grounded guidelines for their design and evaluation, and a set of reporting standards to improve transparency, comparability, and reproducibility in tool-based research. We also aim to identify key open research challenges and to articulate a broader agenda for future work in this emerging area.

Creative Commons BY 4.0

Juho Leinonen, Paul Denny, Hieke Keuning, Natalie Kiesler, and Lauren Margulieux