A model casually says this to me:

I see what happened - when I did git reset --hard HEAD on main, it wiped out all the work we had done. Let me recreate the test files and schema updates, but this time with the correct field names based on what we learned from the backend

This is what is frustrating about these errors (other than the actual code deletion). The model response is so unhuman. Humans could make this mistake sure (although less likely), but their response to this would be different. I'd expect something along the lines of "omg what was I thinking? let me backtrack and fix it and learn from it" or something like bewilderment about how this happened and curious to learn where things went wrong. The model on the other hand, treats this as it does any other turn in the conversation and doesn't backtrack with the panic and fervor that a human would.

Its just weird and something to get used to.