Banking software is extremely regulated and controlled, both by governments and within the industry itself (one small mistake can wipe out entire institutions within a day: https://en.wikipedia.org/wiki/Knight_Capital_Group, banks are extremely risk-averse with technology). So likely they were comfortable using agents to _read_ their code but not to write it.

Have you ever discussed something like this with an executive? I think the most honest response I got on a project I was involved in that required some legitimate legwork was:

'We can get some cheap remote resources for this'

And the funny thing now.. those resources will likely be using tokens anyway..