In my experience it depends a lot on what you give the agent: agents that work on a project well structured, well documented and the agent is given very good co

In my experience it depends a lot on what you give the agent: agents that work on a project well structured, well documented and the agent is given very good context (that documentation, project structure and a clear pattern/style to reuse) then yes, it can be very productive and provide good quality code, especially if the ask is well contained, explained, has clear goals. If you give no boundaries, no good context, references, pattern and style then yes, they write a huge amount of useless code.