But how well can it do it? I still see a lot of errors.
Very well (gathering information given a good model, producing something) and poorly (substituting for real observation and operating without human oversight).
Those looking to replace people will fail hard. Those looking to enhance their staff will kick tail.