Generative artificial intelligence (GenAI) is rapidly entering programming education through tools that can explain code, generate solutions, suggest fixes, provide feedback, and support debugging. These capabilities create opportunities for timely and personalized support, but they also raise concerns about over-reliance, academic integrity, shallow engagement, and the loss of productive struggle. Existing taxonomies classify GenAI tools mainly by functional role or technology type, but these classifications often do not translate easily into instructional design decisions. This conceptual paper proposes a design-space taxonomy for responsible GenAI use in programming education. We argue that two design dimensions are especially consequential: AI entry timing, or when AI becomes available relative to students' independent effort, and AI output scope, or how much of the cognitive work AI is allowed to perform. These dimensions define a two-dimensional design space in which twelve instructional configurations can be positioned. Responsible use is treated not as a binary policy attribute but as a gradient property of the design space: configurations that introduce broad AI output before student effort create high responsible-use demand, whereas configurations that delay AI use, constrain output, or focus on critique and metacognition are more structurally protective of learner agency and cognitive demand. Drawing on recent empirical studies and review literature in computing and programming education, the paper clarifies how tool configuration, scaffolding, disclosure, traceability, and instructor oversight function as associated design features rather than fully independent taxonomy dimensions. The resulting framework offers instructors a practical language for designing AI-integrated programming activities and researchers a structure for comparing interventions more precisely.
Boubaker et al. (Sun,) studied this question.