A study examined the integration of AI agents with deployment infrastructure, finding that agents often modify both application code and infrastructure, leading to integration overhead and security risks.