Keep pulling the thread on Vijoy Pandey.
The discovery and identity components of the Agency.org architecture are built on decentralized hash tables (DHTs) to prevent control by a single entity.
A new access control model called "tool, task, and transaction-based access control" (T-back) is proposed for AI agents, operating on a principle of least privilege and ephemeral permissions.
Vijoy Pandey proposes adding two new layers to the OSI networking model to support agentic AI: Layer 8 for syntactic communication and Layer 9 for semantic/cognition state protocols.
The proposed "Internet of Cognition" architecture consists of three layers: a protocol layer (L8/L9), a cognition fabric for many-to-many communication, and cognition engines for acceleration and guardrails.
Cisco has developed an internal multi-agent system called Community AI Platform Engineer (CAPE) to manage complex cloud computing environments.
Cisco's internal CAPE system has automated approximately 40% of tasks for its site reliability engineers, reducing their workload and improving user response times.
Cisco is leading an open-source project called Agency (A-G-N-T-C-Y) to create a foundation for how AI agents from different entities can collaborate.
Vijoy Pandey believes that to accelerate progress towards Artificial Superintelligence (ASI), the industry must develop the horizontal axis of scaling, enabling collective intelligence among multiple AI agents.
The "Internet of Cognition" is a proposed infrastructure that would enable specialized AI agents from different vendors to share intent, coordinate, and innovate collectively.
The open-source project Agency.org, which is part of the Linux Foundation, provides the foundational infrastructure for multi-agent systems across four pillars: discovery, identity, communication, and observability.
To support agent observability, Cisco worked with Microsoft to develop extensions for OpenTelemetry, the de facto observability stack.
Effective guardrails and the ability to manage multi-vendor agent ecosystems are prerequisites for widespread enterprise adoption of agentic AI.