
It’s a good question, likely has scoped some permissions to build code, in order to debug the code that it writes. But the real question should be:
“How much can we restrict the agent from ‘doing’ before it is no longer an agent and just a chat interface?”
Agent by definition means “authorized to act on behalf of another” so if it cannot act, then what use is it?
English
































