A fundamental problem with ai agents right now

A fundamental problem with ai agents right now is that verification is not actually easier than proving (ie, you have to pore over their work to make sure they were comprehensive + you have to do a lot of querying to figure out what they did)