AI Coding Agents Show Performance Decline After Task Interruption

New research, "Switch-SWE," reveals that AI coding agents experience an 11.5 percentage point performance drop when resuming tasks. This finding underscores a notable challenge in developing reliable AI coding agents, indicating that maintaining context and continuity across interrupted workflows is a critical area for improvement in AI development. Addressing this could enhance the practical utility of AI in software engineering.