Use case: agent comparison

[brillout]

Sure, you can use low effort, but you may still wonder if the result could have been different if higher, and vice verse.

Sure, you can tell an agent like claude code to do the same task in two different subagents and show the difference. But this should be an inbuilt feature - a toggle to say the thread should be forked with x/y/x reasoning, and then you check the difference between the three and pick the right one to resume, and discard the rest.

Quoting YT comment.

I wonder if this can just be a skill, or something that comes "for free" with a skill, maybe agent-driver does the trick? The user can then just ask the agent to spin up two agents with different effort level and then compare results.

Actually, I guess creating an agent-driver skill does the trick.