It also seems like because of this pattern a lot of the tools wouldn't actually be that useful in the field. I've played around with a few now that 1. it was almost impossible to test and debug behavior because you can't tell which elements of the prompt cause it to do what and 2. you can rarely get it to do the same thing each time it's triggered with 100% confidence, which make them pretty useless as workflow tools. This definitely feels like bandwagoning rather than actually coming at it from the perspective of what anyone would find useful.
No comments yet.