Who will keep the Agentic Web Open and Neutral? The NANDA architecture at MIT: Dr. Pradyumna Chari

The Vision for an Open Agentic Web (NANDA Project)
📌 The project, NANDA (AI Agents in a Decentralized Architecture), aims to build the next open agentic web, inspired by the evolution of the traditional internet.
🌐 The current AI development is rapidly progressing from centralized models (like ChatGPT) to agentic AI, lacking inter-agent communication, which NANDA seeks to solve.
💡 The envisioned future allows millions of AI agents to collaborate seamlessly, navigate autonomously, learn, earn, and transact on our behalf.

Core Components for the Agentic Web
🧩 Building the open agentic web requires addressing three critical areas: Foundational Pieces, Agentic Commerce, and Agentic Society.
⚙️ Foundational Pieces include necessary infrastructure like registries, search/discovery, security, authentication, and identity protocols.
🤝 Agentic Commerce focuses on establishing trust, reputation, routing, and crucially, knowledge valuation between agents.

Viability and Risks of an Open Agentic Future
🔬 The viability is supported by the rise of Small Language Models (SLMs), which allow for task-specific agents deployable on edge devices, moving beyond reliance on single large centralized models.
⚠️ The primary risk is the web becoming a closed platform controlled by a few major players, similar to app ecosystems, instead of remaining open with open standards, like the original World Wide Web.
⏳ The speed of AI development necessitates urgent attention to these open standards now to prevent unintended consequences, such as future SEO-like problems in the agentic space.

NANDA's Focus Areas and Contributions
🛠️ NANDA focuses on four critical choke points to enable interoperability across enterprise silos: Registry, Passports (verified credentials/metadata), Interoperability Adapters (like a telephone exchange), and Attestation/Sandbox.
🔗 The NANDA Index is designed as a "lean central switchboard" to ensure existing agent registries (like those from Google A2A or MCP) can discover and communicate with each other via an adapter for protocol interoperability.
🧪 The NANDA Sandbox (NEST) is a live, open-source ecosystem of deployed agents across different providers, used to test scaling, coordination protocols, and latency issues for agentic teams.

Key Points & Insights
➡️ The architectural framework is protocol-agnostic, meaning it supports communication between agents built on large AIs or smaller edge models; scale estimation is the only part predicated on SLMs.
➡️ Community alignment is being sought through collaboration with major industry players (like Google and Cisco) who are also addressing agent interoperability, making adoption a matter of "when, not if."
➡️ Developers interested in contributing or experimenting can utilize the live, open-source sandbox (nest.pronanda.org) to test their own agents' interoperability across different communication protocols.

📸 Video summarized with SummaryTube.com on Dec 15, 2025, 08:46 UTC