Deep learning has already "worked" on many tasked and has also pushed the state of the art in a large range of fields. It's not going to solve every problem but I don't really understand what you mean by not being sure whether it "works".