“With a project like this, we need to be 100% accurate, and that’s really difficult with this technology"

Is that a real requirement? If it is, then why did they proceed? It isn't just difficult.

Yes, the results may sound plausible but expecting accuracy from a probabilistic language prediction engine is a misapplication and an exercise in frustration.

Most companies trying to replace workers with AI will discover the same. There are not too many applications were AI slop is perfectly acceptable.

With apparently so little understanding of the difference between difficult and impossible, she should probably not have been given the job.

"Once we get the minimum viable product, let’s get that out there, and then we’ll enhance that as we learn.”

Make that definitely should not.