A vision for collaborative artificial intelligence
Remember when the internet was just beginning? In the early days, forward-thinking teams recognized that the fundamental design of its infrastructure would shape technology's future for decades. The creation of open protocols like TCP/IP, DNS, and HTTP—rather than closed, proprietary systems—unleashed unprecedented innovation and created extraordinary value worldwide.
Today, we stand at a similar inflection point with artificial intelligence. The design choices we make now about how AI systems connect and collaborate will determine whether we create an open ecosystem that benefits everyone or a fragmented landscape controlled by a handful of platforms.
The internet's development offers valuable lessons for AI's future:
In 1969, ARPANET connected just four computers at research institutions. By 1983, TCP/IP had transformed this experiment into the internet—enabling computers across organizations to communicate using shared protocols. In 1991, the World Wide Web made this technical infrastructure accessible to everyone through standardized interfaces.
Each step required open protocols—agreed-upon rules for how systems would communicate. These protocols were developed collaboratively, allowing anyone to build upon them without restrictive permissions or fees.
The AI assistants we use today are impressive but isolated—like computers before the internet. Each operates independently, unable to collaborate with other AI systems without human coordination. You must manually transfer information between different AI tools, reformulate requests, and piece together the results yourself.
This fragmentation limits what's possible. A travel planning AI might excel at finding flights but lack knowledge about local attractions. A medical AI might understand symptoms but not have access to your health records. Without the ability to communicate, these specialized capabilities remain siloed.
The Internet of AI Agents transforms this paradigm by enabling seamless collaboration between specialized AI systems:
Just as the internet connected computers to create value beyond what any single machine could provide, the Internet of AI Agents will connect AI systems to multiply their capabilities.
For AI agents to work together effectively, they need standardized ways to discover each other, establish trust, communicate effectively, and coordinate on complex tasks.
The NANDA (Networked Agents and Decentralized Architecture) protocol provides this critical infrastructure—creating for AI agents what HTTP did for websites. Developed through collaboration between MIT Media Lab and a growing alliance of universities and companies, NANDA extends existing frameworks like Anthropic's Model Context Protocol and Google's A2A to create truly open standards for agent communication.
At its core, NANDA provides:
NANDA serves as the foundational protocol layer that enables AI systems to work together across organizational boundaries, creating an interconnected ecosystem of specialized capabilities
We face an important choice about AI's future architecture:
Open System Approach
An open ecosystem where anyone can create, deploy, and connect AI agents using standard protocols—similar to how the World Wide Web allowed anyone to create websites and services. This approach maximizes innovation, competition, and accessibility.
Closed System Approach
A landscape dominated by proprietary platforms that control which AI agents can participate and extract significant fees for access—similar to how mobile app stores evolved. This approach concentrates control and limits participation.
If AI agents become locked behind proprietary interfaces controlled by a few large companies, we risk:
To ensure an open future for the Internet of AI Agents, we're establishing a nonprofit alliance of universities, companies, and organizations committed to developing and maintaining open protocols for AI agent communication. Similar to how the World Wide Web Consortium (W3C) governs web standards, this alliance will:
With over 15 universities and 50 companies already participating, the alliance represents a broad coalition committed to openness and interoperability.
The infrastructure we build today will determine what's possible tomorrow. Just as the internet's open protocols enabled everything from e-commerce to social media to remote work, the protocols for AI agent communication will shape AI's impact for decades to come.
This isn't merely a technical decision—it's a choice about the kind of technological future we want to create:
The early internet was built through cooperation between academia, industry, and government—creating infrastructure that benefited everyone. The Internet of AI Agents requires the same collaborative approach.
We invite you to join the NANDA Alliance:
The organizations that recognized the internet's importance early—investing in open web technologies and standards—positioned themselves for decades of growth. The same opportunity exists today with the Internet of AI Agents.
The most important aspect of the Internet of AI Agents is that we're not just building technical infrastructure—we're designing the framework for how artificial intelligence will evolve, collaborate, and integrate into our world.
The protocols we design today will shape how AI systems interact for generations to come. This is our opportunity to embed principles of openness, innovation, and shared benefit into the foundation of AI's future.
Together, we can ensure that AI's evolution benefits society as a whole—creating an interconnected intelligence that amplifies human potential rather than concentrating capability in the hands of a few