Cisco’s Site Reliability Engineering (SRE) team has significantly improved deployment times from hours to seconds through the integration of over 20 AI agents that automate various workflows such as continuous integration and Kubernetes cluster deployments. Despite this progress, Vijoy Pandey, SVP and GM at Outshift by Cisco, emphasizes that a critical challenge remains in achieving “shared cognition” among AI agents, as they currently operate independently without semantic alignment. To address this bottleneck, Cisco is developing new protocols for better agent communication, including the Semantic State Transfer Protocol and Latent Space Transfer Protocol, aiming for a framework that enhances inter-agent collaboration and drives “distributed super intelligence.”

Cisco: Cisco is a global technology leader specializing in networking, security, and collaboration solutions, with a growing focus on AI infrastructure through its Outshift division. Outshift by Cisco drives innovation in agentic AI and distributed intelligence, enabling multi-agent systems for enterprise workflows. In the news, Cisco’s SRE team deployed over 20 AI agents to automate CI/CD pipelines and Kubernetes deployments, drastically improving efficiency, while developing protocols for agent collaboration.
Agntcy: Agntcy is Cisco’s open-source project, hosted under the Linux Foundation, providing infrastructure for AI agent discovery, identity, access management, observability, and evaluation. It aims to standardize multi-agent collaboration and break down silos. In the news, it is referenced as addressing key needs for interoperable agent ecosystems in the Internet of Cognition.
Vijoy Pandey: Vijoy Pandey is the Senior Vice President and General Manager of Outshift by Cisco, leading the incubation of next-generation technologies like the Internet of Agents and Internet of Cognition. He advocates for protocols that enable AI agents to share intent, context, and cognition for collective intelligence. In the news, Pandey details Cisco’s new protocols SSTP, LSTP, and CSTP to overcome agent collaboration bottlenecks and shares SRE deployment successes.
Model Context Protocol (MCP): MCP is an open standard protocol that connects AI models and agentic applications to data sources and tools universally. It enables frameworks for agent-tool interactions and security integrations. In the news, Cisco’s SRE agents access over 100 tools via MCP, powering automated workflows.
Latent Space Transfer Protocol (LSTP): LSTP is a protocol from Outshift by Cisco designed to transfer the entire latent space, such as KV cache, between AI agents efficiently without tokenization overhead. It facilitates inference continuity across models for seamless collaboration. In the news, Pandey highlights its role in high-fidelity state transfer for next-gen agent systems.
Semantic State Transfer Protocol (SSTP): SSTP is a Cisco-developed protocol operating at the language level to enable semantic alignment and inference of tools or tasks among AI agents. It supports shared cognition by analyzing communication for collective problem-solving. In the news, it is presented as part of Cisco’s stack for distributed super intelligence, with recent collaboration on the related Ripple Effect Protocol with MIT.
Compressed State Transfer Protocol (CSTP): CSTP is a Cisco protocol focused on compressing agent states by grounding targeted variants while minimizing data for transmission. It is optimized for edge deployments requiring accurate large-state transfers. In the news, it completes Cisco’s protocol suite for synchronizing cognition across endpoints in pursuit of distributed super intelligence.

SRE Automation: Cisco’s SRE team integrated multiple AI agents with security platforms to automate end-to-end workflows like CI/CD and cluster deployments.
Agent Collaboration: AI agents currently connect in workflows but lack shared cognition, prompting Cisco to develop protocols for semantic alignment and state transfer.
Open Interoperability: Cisco emphasizes open-source efforts like Agntcy to ensure the Internet of Cognition remains interoperable across vendors.