How reliable are AI agents for fixing frontend bugs?

I’ve been testing AI coding agents on React projects, especially for issues involving state management, broken API calls, and unexpected UI behavior. They can identify the problem quickly, but I still find myself checking whether the fix actually addresses the root cause. How reliable have AI agents been for you when debugging React or Next.js projects?