Agentic AI: Rethinking the OSI Model for the Internet of Agents and Cognition

Did you memorize the 7 layers of the OSI model during your networking journey? Application, Presentation, Session, Transport, Network, Data Link, Physical. There are many acronyms to help you remember them, such as: “All People Seem To Need Data Processing.” Fun fact: there are plenty of variations out there. 😄 For decades, the OSI model has been a fundamental reference in networking. Thousands of books, courses, and certifications have taught it as the foundation for understanding how data moves across networks. But now, we are entering a different era. With the rise of Agentic AI and the Internet of Cognition, simply moving data from one endpoint to another is no longer enough. The traditional OSI model was designed for an Internet of Information, an internet where systems primarily exchange data. The OSI model isn't irrelevant. Far from it. But as AI agents begin communicating, collaborating, and exchanging context and meaning, we may need to rethink what sits above the traditional seven layers. That's where two proposed new layers from Cisco Research become particularly interesting: layers focused not just on moving data, but on exchanging meaning between intelligent systems,and that's the idea behind Layer 8 and Layer 9. the two proposed layers by Cisco Research Artificial intelligence. The Agentic AI era is real. It’s not a myth. Companies are already using AI agents to solve increasingly complex problems. Unfortunately, many of these agents are intelligent but operate in silos. With the traditional seven layers of the OSI model, we can send a message from one endpoint to another without understanding the meaning behind that message. The model is designed to move and deliver information not to understand what that information means. The OSI model is excellent at synchronizing deterministic state between endpoints. But it was never designed to synchronize cognitive state between intelligent agents, Once the two new layers are approved, they will improve communication beyond simply exchanging information and transferring data. The Token Economy and Two Proposed OSI Layers Network communication itself primarily transports data; it does not inherently guarantee shared intent or semantic understanding. Imagine agents exchanging information without adequately capturing the intent behind each message. In a token-based economy, unnecessary communication can translate into unnecessary costs for companies. By making intent explicit in agent-to-agent communication, we could move beyond simply exchanging information toward communication that is more meaningful, efficient, and context-aware. Layer 8 - The Agent Communication Layer:This communication layer handles the structure and context of agent-to-agent conversations. Protocols such as MCP and A2A operate at this layer. It allows agents to distinguish whether they are being tasked, queried, informed, or updated. **Layer 9 - The Agent Semantic Layer:**In life, we often start with the “why” before starting a project. This layer focuses on the meaning and intent behind communicating the “what” and “why.” It is designed to establish a verifiable shared understanding before a task is executed. Protocols make the difference Cognition Fabric | Credit :Outshift Collective intelligence is no longer just a theoretical idea. If we want it to scale across the Internet of Cognition, the underlying architecture must evolve beyond simply moving messages between agents. It needs mechanisms for grounding, discovery, resolution, coordination, and negotiation. These capabilities would allow agents to do more than exchange data. They could establish context, interpret intent, coordinate actions, resolve differences, and build a shared understanding of the knowledge exchanged across distributed systems. This is the shift from communication between agents to collaboration between agents. Think about agents communicating like engineers working on the same network. Sometimes, sending the final message is not enough; the other engineer also needs to understand how you arrived there. Latent State Transfer Protocols (LSTPs) could address this problem by preserving more of an agent’s reasoning trajectory while transferring state between endpoints. It is like handing another engineer not only your final configuration, but also the troubleshooting notes that explain how you reached that configuration. The goal is high-fidelity, low-latency communication between agents without losing important context along the way. Compressed State Transfer Protocols (CSTPs) take a different approach: send less, but preserve what matters. Think of compressing a large network capture before sending it to another engineer. The smaller payload consumes less bandwidth and can reach the destination faster, while retaining the information needed by the receiver. Semantic State Transfer Protocols (SSTPs). This is where communication becomes more than moving bits from A to B. Imagine two engineers using different terminology for the same network problem. The data may arrive perfectly, but if they interpret it differently, communication has failed. SSTPs would focus on preserving the meaning of the information so that heterogeneous agents and systems can understand and coordinate around the same semantic context. Cognition Fabric | Credit Outshift Today, we are getting better at connecting AI agents, but connectivity alone is not enough. Just like the Internet can move packets between endpoints without understanding what those packets mean, agents can exchange messages while remaining isolated in their own cognitive worlds.The Cognition Fabric introduces a different vision: connecting cognitive entities so they can exchange not only messages, but also knowledge, intent, and cognitive context. Underneath this vision, technologies such as AGNTCY, MCP, and A2A provide the interoperability and communication foundations. The goal is not simply to make agents talk to each other,it is to give them a way to understand, share, and coordinate context across distributed environments. Shared Intent | Credit Outshift Consider two specialized agents: Prometheus, focused on network engineering, and Themis, responsible for compliance. Instead of operating independently, they need to work toward the same objective. A cognition-state protocol could allow them to compare what each agent knows, detect gaps or contradictions, and feed the resulting knowledge back into a shared cognitive memory fabric. In this model, communication goes beyond sending messages. The agents are able to share context, reconcile knowledge, and coordinate their next actions. Conclusion We are moving from packets to cognition. The OSI model gave networking a common language for moving data between endpoints. But the agentic AI era is asking a different question: What happens when the endpoints are no longer just computers, but cognitive entities? The first seven layers remain fundamental. They move bits, frames, packets, segments, and messages across increasingly complex networks. But connecting agents is not the same as enabling them to collaborate. That is where the idea of Layer 8: Cognition and Layer 9: Semantics becomes interesting. Layer 8 asks: What does the agent know, intend, or need to preserve? Layer 9 asks: What does that information actually mean to another agent? This creates a shift from data transport,state synchronization,cognitive coordination. In this model, technologies such as A2A, MCP, SLIM, and AGNTCY are not replacing the networking foundations we already have. They are becoming pieces of a larger architecture for connecting intelligent systems. The network of the future may therefore not be defined only by how efficiently it moves packets. It may be defined by how effectively it allows machines to exchange context, preserve intent, understand meaning, and coordinate cognition.