afaik, no vendor offers any guarantees on the work their agents do.

Therefore sane organisations using those agents currently need to have someone they trust to review and sign off of on anything consequential those agents do on their behalf. Lawyers for legal, Doctors for medical, Software Engineers for Software etc.

Given the amount of money being thrown at AI - and the need for return - it seems unlikely that regardless of domain that lack of guarantees is a situation that will persist.