Because LLMs understand language, we can start building algorithms that respond to what users say they want. Instead of reverse-engineering user intent from behavior, you can just tell a system “more of X, less of Y” and it listens. Way more flexible than hard-coded workflows.