From Chain of Thought to Network of Action

The latest advancement in AI models centers on reasoning capabilities, embodied in Chain of Thought (CoT). Through hashtag#CoT, models break down complex problems into logical steps, making their reasoning process explicit, which aids in analysis and debugging. This technique has improved the performance of AI models as of early 2025, notably GPT-o1, GPT-o3, and DeepSeek.

We introduce Network of Action (hashtag#NoA), a concept that represents how actions within agents can be structured and exposed in a way that makes them accessible and usable externally. While CoT focuses on improving a model's internal reasoning process, keeping its learned knowledge private and enclosed, hashtag#NoA describes how hashtag#Agents orchestrate actions through openness and external communication.

This distinction can be understood through four key contrasts:

- Models vs. Agents: CoT is centered on models whereas NoA is centered on agents that use models.

- Chain vs. Network: A chain represents a linear sequence of reasoning steps, whereas a network connects actions dynamically, allowing for interdependent decision-making.

- Thought vs. Action: CoT is "thought" expressed in text. Text does not change the world; action does. NoA builds on this by enhancing action, ensuring that agents don't just execute predefined steps but continuously adjust, collaborate, and refine their actions based on real-world feedback.

- Closed vs. Open: CoT operates strictly within model boundaries, preserving knowledge as an internal capability, while NoA enables agents to expose and coordinate actions externally, fostering collaboration and creating a shared ecosystem of capabilities.

However, the true value of NoA is not just in making individual actions available but in enabling a network of actions to adapt dynamically based on execution results. This network-level capability ensures that actions do not operate in isolation; instead, they influence and reference each other, adjusting in real-time based on outcomes. This allows the system to not only deliver results of real-world change but also continuously refine and optimize state changes approaching the real goal desired by human.

We will discuss more about this in Agent Network.

hashtag#Agent hashtag#theNetworkIsTheAI hashtag#AgentNetwork