AutoTutor helps students learn by holding a conversation in natural language. AutoTutor is adaptive to the learners’ actions, verbal contributions, and in some systems their emotions. Many of AutoTutor’s conversation patterns simulate human tutoring, but other patterns implement ideal pedagogies that open the door to computer tutors eclipsing human tutors in learning gains. Indeed, current versions of AutoTutor yield learning gains on par with novice and expert human tutors. This article selectively highlights the status of AutoTutor’s dialogue moves, learning gains, implementation challenges, differences between human and ideal tutors, and some of the systems that evolved from AutoTutor. Current and future AutoTutor projects are investigating three-party conversations, called trialogues, where two agents (such as a tutor and student) interact with the human learner.