Selfish behavior can often lead to suboptimal outcome for all participants. This is especially true in dynamically changing environments where the game or the set of the participants can change at any time without even the players realizing it.  Over the last decade we have developed good understanding how to quantify the impact of strategic user behavior on overall performance via studying stable Nash equilibria of the games.  In this talk we will consider the quality of outcomes in games when the population of players is dynamically changing, and where participants have to adapt to the dynamic environment. We show that in large classes of games (including traffic routing), if players use a form of learning that helps them to adapt to the changing environment, this guarantees high social welfare, even under very frequent changes. Joint work with Thodoris Lykouris and Vasilis Syrgkanis.