An agent system includes a microphone configured to collect a sound of a vehicle interior, a speaker configured to output a sound to the vehicle interior, a recognizer configured to recognize a meaning of the sound collected, an agent controller configured to generate an agent sound for speaking to an occupant of a vehicle on the basis of the meaning recognized and cause the agent sound to be output to the speaker so that a sound image of the generated agent sound is localized at a prescribed position, and a travel environment acquirer configured to acquire a travel environment of the vehicle, wherein the agent controller causes an aspect in which the sound image of the agent sound is localized to be changed when the travel environment of the vehicle acquired by the travel environment acquirer is changed or is predicted to be changed in the future.