Thanks, I was confused by the article's induction proof which seems to completely ignore new tasks coming in. The random walk argument is very clear.