Abstract:
Agentic programming is becoming increasingly central to contemporary software and AI development. Rather than viewing programs as scripted procedures, agentic programming frames them as semi-autonomous systems that perceive their environment, pursue goals, make decisions, and adapt over time.
This paper proposes a pedagogical lens that uses Scratch, a block-based visual programming environment, to make agentic programming ideas more concrete and teachable. We argue that Scratch offers an intuitive space in which learners can conceptually model agentic behavior without relying on advanced AI infrastructure. Sprites function as agents, variables hold internal states, sensing blocks provide perception, broadcasts enable communication, and control structures create sense–think–act cycles that resemble agentic reasoning. Thus, Scratch becomes a shared language through which computer science and software engineering educators and practitioners can discuss and explore agentic systems.
We illustrate how familiar Scratch activities can be reframed to highlight agency, autonomy, decision-making, and interaction, thereby strengthening computational thinking on a higher level of abstraction. This reframing enables students to view programs as purposeful digital collaborators rather than passive executors of instructions, building mental models that connect traditional coding practices with emerging agentic approaches. In a future work, we plan to develop AgentScratch, which will include a Cognition Palette whose basic building blocks simulate key processes of agentic software development (Hazzan and Ragonis, 2026).
This paper proposes a pedagogical lens that uses Scratch, a block-based visual programming environment, to make agentic programming ideas more concrete and teachable. We argue that Scratch offers an intuitive space in which learners can conceptually model agentic behavior without relying on advanced AI infrastructure. Sprites function as agents, variables hold internal states, sensing blocks provide perception, broadcasts enable communication, and control structures create sense–think–act cycles that resemble agentic reasoning. Thus, Scratch becomes a shared language through which computer science and software engineering educators and practitioners can discuss and explore agentic systems.
We illustrate how familiar Scratch activities can be reframed to highlight agency, autonomy, decision-making, and interaction, thereby strengthening computational thinking on a higher level of abstraction. This reframing enables students to view programs as purposeful digital collaborators rather than passive executors of instructions, building mental models that connect traditional coding practices with emerging agentic approaches. In a future work, we plan to develop AgentScratch, which will include a Cognition Palette whose basic building blocks simulate key processes of agentic software development (Hazzan and Ragonis, 2026).