The agent principal-agent problem
Summary
The article argues that introducing AI agents into code review disrupts the traditional 'review-then-commit' model, creating a principal-agent problem where developers' signals of effort are diluted. It outlines how this can double review cycles and reduce productivity unless a smaller, high-trust team retains human oversight and adds integration testing and agent-based workflows. It also provides historical context and practical advice for small teams to ship safely with agent-enabled processes.