Advances & challenges in foundation agents: Introduction to Part I – Core components of intelligent agents

This article is the Introduction to Part I of a series of articles featuring Liu and colleagues’ book Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems.

What makes an intelligent agent more than just a clever model? The difference lies not in scale but in structure—in the presence of a mind that can see, feel, remember, plan, and act. This part of the series dives beneath the surface of modern AI systems to examine the internal architecture of intelligent agents, not as abstract maths, but as working minds. Minds that perceive their world, carry memory through time, adapt to change, and chase goals beyond the next token.

At the heart of Liu and colleagues’ framework is a living loop: perception feeds cognition, cognition selects action, and action reshapes perception. But this loop is not empty. Inside cognition, we find a constellation of faculties—memory, world model, emotion, goals, and learning—each contributing to the agent’s ability to adapt and pursue long-term purpose. These are not accessories but necessities, as real to an agent as lungs and eyes are to a body.

This part explores each of these faculties in turn. We begin with cognition, the control center that organizes thought, plans, and decisions. From there, we trace the roles of memory, which binds past to present; the world model, which imagines futures; and the reward system, which teaches from outcomes.

We also examine emotion, not as sentiment but as a powerful biasing force in cognition and behavior. The chapters then lead outward to perception, the gateway from world to mind, and action, the channel through which mind shapes world. Each module is inspired by how biological systems—especially the human brain—solve these problems, but is cast in a form suitable for building digital minds. This part lays the groundwork for everything that follows.

Before we can teach agents to collaborate, generalize, or reason morally, we must understand what makes them tick—what internal machinery gives rise to intelligent behavior. These are the core components of a Foundation Agent. We now open the casing to see how each part fits, functions, and learns.

Next part: Introduction to Chapter 2 – Cognition.

Article source: Liu, B., Li, X., Zhang, J., Wang, J., He, T., Hong, S., … & Wu, C. (2025). Advances and challenges in foundation agents: From brain-inspired intelligence to evolutionary, collaborative, and safe systems. arXiv preprint arXiv:2504.01990. CC BY-NC-SA 4.0.

Header image: AI is Everywhere by Ariyana Ahmad & The Bigger Picture / Better Images of AI, CC BY 4.0.