That’s most definitely not the same thing, as „improving a codebase” is an open ended task with no reliable metrics the agent could work against.