Stuart_Armstrong comments on The subagent problem is really hard