Towards an Appropriate Level of Reliance on AI: A Preliminary Reliance-Control Framework for AI in Software Engineering

Researchers propose a reliance-control framework for AI tools in software development, based on interviews with 22 developers using LLMs. The study addresses the tension between overreliance (risking skill atrophy) and underreliance (missing productivity gains), offering guidance for developers, educators, and policymakers on appropriate AI tool usage.

This research addresses a critical emerging challenge in AI adoption: determining the optimal balance between leveraging AI capabilities and maintaining human expertise. As LLMs become increasingly integrated into software development workflows, the question of appropriate reliance has moved from theoretical concern to practical necessity. The study's reliance-control framework provides a structured approach to evaluating how developers should interact with AI tools, moving beyond simplistic adoption metrics toward nuanced understanding of tool effectiveness.

The research builds on growing recognition that AI tool adoption isn't binary. While productivity gains from LLM-assisted development are documented, concerns about developer skill degradation and over-automation have gained traction in engineering communities. This study quantifies those concerns through qualitative research, establishing control as a measurable dimension for assessing reliance levels. By framing the problem as a spectrum rather than a binary choice, the framework acknowledges that different development contexts may warrant different reliance strategies.

For the software engineering industry, this work has implications across multiple stakeholder groups. Tool developers can use control-level insights to design better user interfaces that calibrate assistance appropriately. Organizations managing development teams face new considerations about AI tool rollout and training protocols. Educational institutions require guidance on curriculum changes as AI becomes standard in professional practice. The framework's emphasis on "appropriate" rather than "maximum" reliance suggests a maturing conversation about AI integration, moving from enthusiastic adoption toward sustainable, responsibility-focused deployment.

Future research should empirically validate the framework across diverse development contexts and measure long-term skill outcomes. As LLM capabilities evolve, the specific control mechanisms that constitute appropriate reliance will require continuous reassessment.

• →Researchers propose a reliance-control framework identifying optimal balance between AI tool usage and maintaining developer expertise.
• →Overreliance on LLMs risks skill atrophy while underreliance forgoes productivity and quality improvements in software development.
• →Control mechanisms emerge as the key variable for assessing appropriate AI reliance levels across different development scenarios.
• →The study provides actionable guidance for tool developers, organizations, educators, and policymakers managing AI adoption.
• →Framework emphasizes sustainable, responsibility-focused AI integration rather than maximizing automation levels.