Digest

This podcast introduces Cisco's Outshift and VJoy Pandey, discussing the "Internet of Cognition" – a framework for AI agents to collaborate and share context. Cisco's core business in networking, security, observability, and collaboration is highlighted, emphasizing their distributed systems DNA. The OSI model and machine learning in network management are explained. Project Jarvis (CAPE) is presented as an agent-based SRE automation tool. The discussion then delves into multi-agent systems, AGI, ASI, and the shift to horizontal scaling of intelligence. The "Internet of Cognition" is further elaborated as an AI cultural evolution, enabling specialized agents to innovate collectively. Foundational infrastructure for multi-agent systems, including discovery, identity, communication, and observability, is detailed. Challenges in decentralized identity, trust, and accountability for agents are addressed, with T-Back proposed as a new access control model. The podcast introduces two new network layers (syntactic and semantic communication) and the overall "Internet of Cognition" architecture. Finally, the practical demonstration of this concept is highlighted, showcasing its potential for enhancing SRE productivity and enterprise AI adoption.

Outlines

00:00:00

Introduction to Cisco's Outshift and the Internet of Cognition

Cisco's Outshift and VJoy Pandey discuss agentic AI systems and the "Internet of Cognition," a framework for AI agents to collaborate and share context, leveraging Cisco's network infrastructure expertise.

00:04:25

Cisco's Core Business and Distributed Systems Foundation

VJoy Pandey outlines Cisco's core pillars: networking, security, observability, and collaboration, emphasizing their identity as a distributed systems company that enables horizontal scaling and distributed computing.

00:06:52

Understanding Network Communication: The OSI Model and ML Applications

The OSI seven-layer network model is explained, detailing its layers and relevant protocols. Machine learning's role in network management, particularly in control planes for anomaly detection and traffic prediction, is also discussed.

00:13:37

Project CAPE: Agent-Based Automation for SRE

Project Jarvis, now known as CAPE (Community AI Platform Engineer), is introduced as a multi-agent system designed to automate Site Reliability Engineering (SRE) tasks, aiming to reduce SRE workload and improve response times.

00:21:08

The Frontier of Multi-Agent Systems and AI Evolution

The conversation explores the capabilities of multi-agent systems, defining Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI), and highlighting the shift towards horizontal scaling of intelligence.

00:27:22

The Internet of Cognition: AI Cultural Evolution and Collaboration

VJoy Pandey elaborates on the "Internet of Cognition," drawing parallels with human language evolution. The vision includes specialized agents collaborating, sharing intent, and innovating collectively, fostering a form of AI cultural evolution.

00:35:21

Enabling AI Agent Collaboration: Infrastructure Needs

The podcast addresses the current isolation of AI agents and the critical need for infrastructure that enables seamless interaction, shared context, and collaboration to overcome limitations in intent alignment and collective problem-solving.

00:39:21

Foundational Plumbing for Multi-Agent Systems

The essential infrastructure for multi-agent systems is detailed, covering agent discovery, identity and access management, robust communication protocols (MCP, A2A), and comprehensive observability for reliable operation.

00:56:42

Decentralized Identity, Trust, and Access Control for Agents

The importance of decentralized identity and discovery mechanisms for agents is emphasized, contrasting with centralized platforms. Challenges in establishing trust, responsibility, and accountability are discussed, leading to the proposal of T-Back for access control.

01:10:08

Extending the Network Stack for Agent Communication

Two new layers, Layer 8 (syntactic communication) and Layer 9 (semantic/cognition state communication), are proposed to extend the OSI model, enabling effective agent-to-agent interaction and understanding of intent and context.

01:16:09

The Internet of Cognition Architecture and Practical Demonstration

The "Internet of Cognition" architecture is presented with its protocol, fabric, and cognition engine layers. A practical demonstration is introduced, showcasing a multi-agent system designed to enhance SRE productivity and handle crisis situations through collective intelligence.

Keywords

Q&A

• What is the "Internet of Cognition" and why is it important?

  The "Internet of Cognition" is a vision for a future where AI agents can seamlessly share context, understand each other's intent, build reputation, and collaborate to solve complex problems. It's crucial for scaling intelligence horizontally and fostering collective innovation, moving beyond individual AI models.

• How does Cisco's Outshift approach agentic AI systems?

  Outshift leverages Cisco's deep expertise in distributed systems and networking to address the challenges of agentic AI. They focus on building foundational infrastructure and protocols that enable AI agents to interact, collaborate, and operate reliably within enterprise environments.

• What are the core pillars of Cisco's business?

  Cisco's core business is built on four pillars: networking, security, observability, and collaboration. They are fundamentally a distributed systems company enabling scale-out and horizontal scaling of computing resources.

• How is machine learning currently used in network management?

  Machine learning is used in network management primarily within the control plane for tasks like anomaly detection, traffic prediction, and failure prediction. While data flow is deterministic, ML optimizes routing, policy enforcement, and outage management.

• What is CAPE (Community AI Platform Engineer)?

  CAPE is a multi-agent system developed by Cisco that automates Site Reliability Engineering (SRE) tasks. It comprises around 20 agents working together to manage cloud environments, reducing SRE load by up to 30% and automating 40% of tasks.

• What are the proposed new layers for the network stack in the context of AI agents?

  Two new layers are proposed: Layer 8, syntactic communication, to ensure agents built on different frameworks can communicate, and Layer 9, semantic or cognition state communication, to extract meaning and structure from agent interactions for better coordination and trust.

• What is T-Back and why is it needed for AI agents?

  T-Back (Tool, Task, and Transaction-Based Access Control) is a proposed access control model for agents. It grants minimal, ephemeral permissions based on the specific action an agent needs to perform, moving beyond human-centric RBAC to manage the dynamic and non-deterministic nature of agents.

• How does the "Internet of Cognition" architecture facilitate agent collaboration?

  It comprises three layers: a protocol layer for semantic communication, a cognition fabric for scalable many-to-many interactions with pluggable memory, and cognition engines acting as accelerators or guardrails to ensure constructive emergent behavior within defined boundaries.

• What are the key challenges in enabling AI agent collaboration?

  Key challenges include discoverability, identity and access management, secure communication, and observability. Additionally, enabling shared intent, context, and collective innovation requires new protocols and infrastructure beyond current deterministic systems.

• What is the significance of decentralized identity and discovery for agents?

  Decentralized identity and discovery are crucial for an open and interoperable agent ecosystem. They prevent single entities from controlling access and reputation, fostering trust and enabling permissionless participation, unlike traditional centralized platforms.

• What are the main hurdles for generative AI adoption in enterprises?

  Enterprises require well-defined and effective guardrails for AI solutions. Additionally, the integration of diverse agents from multiple providers necessitates a system where these agents can connect, share cognition, align on intent, and work together, as a single agent performing all functions is not feasible in the short term.

• How can one explore the practical application of the "Internet of Cognition"?

  A live, clickable demo is available at outshift.sysco.com/InternetofCognition. This demonstration allows users to interact with a multi-agent system, observe agent activities, intents, contexts, and the emergent collective reasoning, making the concepts more concrete.

Show Notes