941: Multi-Agent Human Societies, with Dr. Vijoy Pandey

Episode Summary

Executive Summary: Dr. Vijoy Pandy of Cisco’s Outshift outlined a vision for an open, interoperable “Internet of Agents” where humans and AI agents collaborate across business, scientific, social, and physical domains. He introduced Agency, an open-source Linux Foundation project designed to solve agent trust, identity, discovery, and interoperability, including zero-trust access via task/tool/transaction-based controls and future semantic understanding.

Main Topics: Outshift and Cisco’s innovation model: Outshift is Cisco’s internal incubator, structured like a startup to bring ideas from ideation to customer traction across adjacent areas like networking, security, observability, and collaboration. Internet of Agents and Agency: Agency is Cisco’s open-source manifestation of an Internet of Agents: a platform for interoperable collaboration among agents, humans, and enterprises across vendors and clouds. Multi-agent human societies: Pandy described a future where humans and agents form societies to accelerate scientific discovery, physical work, and everyday tasks, from drug discovery to doing dishes. Neurosymbolic and hybrid AI: He argued that deterministic systems and probabilistic AI should be blended, not replaced, with digital twins and neurosymbolic approaches improving accuracy, speed, and reliability. Trust, identity, and zero-trust access for agents: Traditional IAM breaks down for agents because they are probabilistic and operate at machine speed; Agency proposes TBAC (task/tool/transaction-based access control), ephemeral permissions, and sandboxed runtimes. Interoperability with emerging standards: Agency is positioned to work alongside MCP and A2A, with Cisco and partners pushing for open standards rather than fragmented protocols. Future direction: semantic understanding: Beyond syntactic interoperability, the next challenge is semantic interoperability—agents must understand meaning, intent, and shared definitions like ‘best’ or ‘90%.’

Key Arguments: Agents will not just automate business software; they will participate in scientific discovery, physical robotics, social interaction, and consumer workflows. Open source is the modern path to shaping standards, because openness and interoperability are required for an effective agent ecosystem. Current IAM systems were designed for humans or deterministic software and fail for probabilistic agents that change behavior and act at machine scale. TBAC is a better fit for agents because access should be granted only for a specific task, tool use, or transaction, then revoked immediately after. Identity, discovery, communication, and evaluation are the four pillars needed for trustworthy agent deployment at enterprise scale. Hybrid systems combining deterministic infrastructure with probabilistic AI provide better speed and accuracy than relying on LLMs alone. Digital twins of humans, systems, networks, and the physical world can safely validate agent workflows before real-world deployment. The next frontier is semantic, not merely syntactic, interoperability; agents must share meaning, not only message formats. Fragmentation among standards and frameworks would slow adoption, so the industry should coordinate around interoperable layers rather than compete on isolated stacks.

Data Points: Outshift structure: Internal incubator within Cisco - Described as a startup-like unit spanning engineering, product, marketing, support, and sales Agency founding ecosystem: 4 formative members plus ~80 additional organizations - Cisco, Dell, Google, Red Hat, and Oracle were named as founding members; around 80 other orgs contribute or deploy it Linux Foundation home: Agency moved to the Linux Foundation - Presented as the open-source home for the Internet of Agents project Enterprise workflow needs: 4 pillars - Identity, discovery, communication, and evaluation were identified as the core pipeline needs Zero-trust access model: TBAC / T3BAC - Task, tool, transaction-based access control proposed for agents Human-in-the-loop duration: Long time to come - Used to emphasize humans will remain part of workflows even as agents proliferate Example access duration: 10 seconds - Illustrative example of granting limited access to withdraw $10 from a safe for only a short task window Vendor confidence example: 90% confidence - Used to show that cross-vendor meaning of confidence scores may differ and needs standardization Community reference app: Coffee Agency - Open-source reference application demonstrating supply-chain complexity and plug-and-play components

Pivotal Quotes: "The openness comes from open source these days. Standards, yes, but open source are the new standards." — Dr. Vijoy Pandy: Explaining why Cisco chose to open-source Agency and shape Internet of Agents standards "I don't trust any agent, and I just trust it once it's proven that it's supposed to do X, Y, or Z for the duration of that X, Y, or Z." — Dr. Vijoy Pandy: Describing zero-trust for agents and the rationale for TBAC "The bigger problem, especially if we are true to our vision, which we are, is to understand each other semantically." — Dr. Vijoy Pandy: On the next frontier after syntactic interoperability

Implications: For builders, the message is clear: enterprise-grade agent systems need open standards, zero-trust identity, hybrid AI, and semantic interoperability. For the industry, Agency aims to lower adoption friction and make multi-agent workflows safer, more scalable, and more portable across vendors.