🧠 AI Ed Wiki

📄 Research Article

Agents That Teach: Designing Incidental Learning Back into AI-Assisted Software Development

● high · 2026-07-08agentic-ai, cs-education, generative-ai, higher-ed, llm, over-reliance, professional-training, self-regulated-learning, ai-literacy, rag

As AI coding agents take over substantial implementation work, developers increasingly lose the informal, effortful problem-solving through which software engineering expertise historically accumulated. The authors argue this "incidental learning" will not return spontaneously and that over-reliance on agentic coding lets unpracticed skills atrophy, accruing a developer-level analogue of Technical Debt they name Knowledge Debt — changes the agent executes that the developer cannot fully understand. They propose six design principles for learning-aware development and operationalize them in SHIELD, a multi-agent system that surfaces contextual, out-of-band learning moments drawn from the coding agent's own reasoning without disrupting developer flow.

  • Incidental learning at risk: Delegating coding to autonomous agents short-circuits the problem-solving pathway that traditionally built SE expertise, risking silent skill atrophy and accumulating Over Reliance.
  • Knowledge Debt concept: A novel framing extending Technical Debt to the developer's understanding gap when agent-generated changes outpace comprehension.
  • Design principles: Six principles guide systems that consciously re-introduce learning into developer–agent interaction, relevant to Agentic Education Coding and Agentic Workflows Education.
  • SHIELD prototype: A "agents that teach" multi-agent system leverages the coding agent's own reasoning to surface learning moments, complementing AI Literacy goals for practitioners.
  • Vision: Learning-aware development environments where productivity and learning are complementary rather than competing — a theme echoed in Self Regulated Learning and Professional Training.

    • Connected Concepts

  • Over Reliance
  • AI Literacy
  • Self Regulated Learning
  • Professional Training

    • Connected Articles

  • Agentic Education Coding
  • Agentic Workflows Education

    • Citation

    Mehra, R., Suri, S., Tagadinamani, P. K., Singi, K., & Kaulgud, V. (2026). Agents That Teach: Towards Designing Incidental Learning Back into AI-Assisted Software Development. arXiv:2607.06101.