Self-Evolving Coding Agents

This survey studies self-evolving coding agents: coding agents that improve future behavior by updating their frameworks, memories, skills, tools, models, or collaboration structures from prior software-engineering interactions. It proposes an object-centered taxonomy and adds two complementary views: when evolution occurs and which software-specific evidence drives it. The authors argue that executable feedback, repository-level context, and coding trajectories make software engineering a distinctive setting for agent evolution. They also identify unresolved issues including unreliable feedback, benchmark overfitting, safety, maintainability, cost, and generalization.