Decision-making is a dynamic process that demands intricate interplay between perception, memory, and reasoning to discern choices and formulate optimal policies. Conventional approaches to sequential decision-making face challenges related to low sample efficiency and poor generalization. In contrast, the success of foundation models in language and vision domains has showcased their ability of rapid adaptation to diverse new tasks. Therefore, we advocate for the construction of foundation agents as a transformative shift in the learning paradigm of agents. This proposal is underpinned by an extensive discussion on the current state of the field, encompassing recent advancements that concern learning from large-scale interactive data through offline reinforcement or imitation learning, self-supervised pretraining and adaptation for sequential decision-making, and the utilization of large language models to create autonomous agents. In conclusion, we pinpoint critical challenges and delineate trends for foundation agents, addressing both technical and theoretical aspects to propel the field towards a more comprehensive and impactful future.