Agentic Coding Studio: Learner-Directed AI in Informal STEM Environments

Artificial intelligence is changing how computer programs are created. As artificial intelligence agents take on more routine code production, people will increasingly need to define goals, break problems into parts, judge proposed solutions, test results, and revise their work. This project will develop the Agentic Coding Studio, a platform in which young people will work with an artificial intelligence coding agent to create personally meaningful games, stories, simulations, and applications. Instead of treating artificial intelligence as an answer machine, youth will remain responsible for what they make: they will clarify their intentions, review the agent's plan, test whether projects work as intended, and improve them through feedback and reflection. The project will lower barriers created by unfamiliar programming syntax while preserving the creativity, judgment, persistence, and collaboration central to learning computer science. Through partnerships with STEM-oriented out-of-school programs and informal learning centers, the project will broaden access to artificial intelligence and computing education. It will directly engage approximately 2,500 youth and 120 educators, reach approximately 50,000 additional youth through participating sites, and make learning activities, educator training materials, and research findings openly available. The three-year project will co-design, implement, and evaluate four integrated components: a customizable block-based coding platform; a pedagogically constrained artificial intelligence coding agent; guided, project-based learning activities; and professional development for informal science, technology, engineering, and mathematics educators. The agent will ask clarifying questions, propose plans before acting, work in manageable milestones, and summarize changes so that learners can direct and evaluate its work. An embedded Prompt Clinic will give feedback on the clarity and specificity of learners' requests. Using design-based implementation research, an iterative approach in which researchers and community partners jointly refine an educational innovation in authentic settings, the team will conduct pilots at two sites, expand to four sites, and then scale to at least 24 sites. Mixed quantitative and qualitative methods will examine how youth learn to plan, prompt, inspect, test, revise, and reflect; how participation affects computational agency, artificial intelligence literacy, motivation, and identification with science and technology; and how educators, peers, professional development, community settings, and platform features support or constrain learning. Evidence will include youth projects, agent interaction logs, prompt-quality scores, platform activity, observations, interviews, embedded learning activities, and co-design records. Findings across these sources will be integrated to guide repeated improvements and to produce practical resources, design principles, and new knowledge about how young learners can retain purposeful control while collaborating with increasingly capable artificial intelligence systems. This project was funded by the STEM K-12 program, which supports fundamental, applied, and translational research on STEM learning in both formal K-12 education and informal learning environments. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.