TFA is a bit rambling and readers are getting distracted by specific claims, like the bit about 99.9%+ reliability. TFAs main point is that productive use of AI agents requires tightly specified context and frequent human intervention, which is what folks have been saying for a while.