Recommender systems have evolved from addressing information overload in mailing lists and usenet news in the 1990s to now helping with advertising, engagement, and connection. They have moved from collaborative filtering based on user ratings to being query-less and able to anticipate user interests without searches. While traditional problems include accuracy, scalability, and cold starts, five open problems are predictions over time, balancing algorithms and data, understanding users and ratings, modeling items, and measuring recommendations beyond rankings. The key lessons are that recommender systems draw from many disciplines, solutions must consider the domain, and participating in the community is important.