But usually verifying the answer is cheaper than finding the answer, so it can be done. The prompt given to the sub-agent can be a question with "and explain your answer", the answer to which will still consume far less context than figuring out the answer.
Of course it has to. The subagent might have misunderstood and spent 250k tokens smoking crack. If that output is blindly trusted the parent agent might go on to burn millions of tokens going in the wrong direction, completely negating the point of using a subagent in the first place.
Yes, and back in my school days we all did our tests on paper, except the two kids who had a slip of paper that allowed them to take a digital test instead. Helping people with disabilities doesn't require having everyone make their tests on a computer.
And taking away writing from people with bad handwriting (if not related to disability) is just about the worst thing to do if you want them to ever get better at it.
Idk, I’ve had bad handwriting my entire childhood and I still have bad handwriting now as an adult. I’ve (somewhat jokingly) announced that I am “illegible”. It isn’t a detriment to my day to day activities at all, because I can type sufficiently well
But usually verifying the answer is cheaper than finding the answer, so it can be done. The prompt given to the sub-agent can be a question with "and explain your answer", the answer to which will still consume far less context than figuring out the answer.