I'm not sure I agree with this. 10x more tokens means leaaving the agent to work for 10x longer, which may lead to bugs and misintepretation of the intention. Breaking the goal into multiple tasks seems more efficient in terms of tokens and getting close to the desired goal. Of course this means more human involvment, but probably not 10x more.