Imagine an AI agent capable of seamlessly navigating a multitude of tasks, from controlling robots to strategizing in complex games, all thanks to a revolutionary learning approach. This is the promise of foundation agents, a paradigm shift in decision-making that draws inspiration from the success of large language models (LLMs). Traditional methods for training AI agents, like reinforcement learning, often struggle with efficiency and adaptability. Foundation agents, however, aim to overcome these limitations by learning from massive datasets of interactive experiences, much like LLMs learn from vast amounts of text. This allows them to develop a unified understanding of various decision-making components, such as actions, states, and rewards, enabling them to generalize their knowledge to new, unseen situations. The development of foundation agents involves three key stages: large-scale data collection (or generation using simulators), self-supervised pretraining, and alignment with LLMs. This process allows the agents to learn complex behaviors, handle uncertainty, and even reason strategically. One of the most exciting aspects of foundation agents is their potential to tackle open-ended tasks, those without a predefined goal or endpoint. This opens doors to a new era of AI where agents can continually learn, adapt, and even exhibit creativity in solving problems. While the field is still in its early stages, researchers are exploring various approaches to building foundation agents, including unified models that handle all tasks and compositional models that combine existing specialized models. Key challenges remain, such as developing robust theoretical foundations for policy optimization and addressing the potential for biases inherited from LLMs. Despite these challenges, the potential of foundation agents to revolutionize fields like robotics, healthcare, and scientific research is immense. As these agents continue to evolve, we can expect to see them playing increasingly important roles in our lives, making complex decisions and driving innovation across various industries.