# Evaluation

• AI agents get office tasks wrong around 70% of the time, and a lot of them aren't AI at all