Episode 133 - GluuFederation/identerati-office-hours GitHub Wiki
Title: Governing the Agentic Web
- Host: Mike Schwartz, Founder/CEO Gluu
- Guest: Andor Kesselman, Founder DIF Labs
Channels
Description
The agentic web is coming — a world where AI agents act on our behalf, coordinate with one another, and make decisions at scale. But with new autonomy comes new risk: How do we know who or what to trust? In this episode, we’ll explore the foundations of agentic governance: verifiable identity, proof of personhood, and runtime policy enforcement using tools like Open Policy Agent (OPA). We’ll also dive into multi-agent orchestration, where agents must delegate, collaborate, and operate within shared constraints. Drawing on lessons from Project NANDA and decentralized identity work, this conversation will unpack what it takes to build a trustworthy, human-aligned agentic ecosystem — before it builds itself.
Homework
-
Project NANDA Overview (MIT): https://nanda.media.mit.edu
-
Presentation for Global Digital Collaboration meeting: The 🤖 AI Agents Identity Landscape (50 Slides)
Takeaways
-
⚡ From a personal level: How do I define the guardrails for my autonomous agents? How do I know how an agent will act on my behalf with certainty? How can I verify the agent's behavior to make sure it's within my lines?
-
⚡ From an enterpise level: How do I comply with regulatioins and manage risk for the swarm of agents acting on behalf of all the people in my firm?
-
⚡ Agents need "advanced directives." They need to be able to assert the legal authority to do something on behalf of a person, even if the person is dead. For example, "Close all my social media accounts when I die."
-
⚡ Our current tokens are inadequate, we don't even know how to name them, and we certainly can't make policies that should govern their behavior. What kind of JWT will it be?