The Internet is About to Change Forever: Why We Need a New Address Book

# The Internet is About to Change Forever: Why We Need a New Address Book *Picture this: It's 2027. You wake up, and before you even reach for your phone, your personal AI agent has already booked your dentist appointment, rescheduled a meeting that conflicts with your daughter's soccer game, ordered groceries for tonight's dinner party, and negotiated a better rate on your car insurance. But here's the thing, none of this is happening on today's internet. It's happening on something entirely new.* --- ## The Day the Internet Broke (Or Almost Did) Let me tell you a story about Sarah, a project manager at a company in Austin. It's a Tuesday morning in early 2026, and Sarah decides to let her new AI assistant handle her routine tasks for the day. "Book a conference room for 2 PM, order lunch for the team meeting, and find a reliable translator for our Japanese client call," she tells her AI. Sounds simple, right? But here's what actually happens behind the scenes: Sarah's AI agent needs to: - Find and verify available conference rooms across multiple booking systems - Coordinate with her colleagues' AI assistants to check their preferences and dietary restrictions - Locate a certified translator who speaks business-level Japanese and is available at 3 AM Tokyo time - Ensure all of this happens securely without exposing sensitive company information The problem? Today's internet has no idea how to handle this. Sarah's AI assistant starts frantically searching the web like a human would, clicking through booking sites, calling APIs that may or may not work, trying to piece together information from dozens of disconnected services. What should take seconds stretches into minutes, then hours. Some tasks fail entirely because there's no reliable way for her AI to verify if other services are trustworthy. By noon, Sarah has a half-booked conference room, lunch ordered for the wrong day, and no translator. This isn't science fiction. This is what happens when you try to run tomorrow's AI agents on yesterday's internet infrastructure. --- ## The Great Mismatch: Why Today's Internet Wasn't Built for This Imagine the internet as a vast city. The internet we built in the 1990s is like a sleepy suburban town where: - Everyone knows everyone else - People call ahead before visiting - The phone book gets updated once a year - If you want something, you walk to the store yourself This worked perfectly when humans were the only residents. But now imagine millions of new residents moving in, except these aren't people. They're tireless workers who: - Never sleep and are constantly active - Don't know each other but need to collaborate instantly - Change jobs and addresses every few minutes - Need to verify each other's credentials before working together Our sleepy suburban infrastructure is about to be overwhelmed. ### The Phone Book Problem When you type "google.com" in your browser, a system called DNS looks up Google's address and connects you. Think of it as the internet's phone book. This phone book was revolutionary in 1983, but it has some very human assumptions: **It assumes you know what you're looking for.** The phone book can tell you where Google lives, but it can't answer "Who can help me translate this document right now?" **It assumes addresses don't change often.** Your local pizza place probably has the same phone number for years. But AI agents might change their "address" every few minutes as they move between servers for better performance. **It assumes you don't care about credentials.** The phone book tells you where someone lives, but not whether they're qualified to do brain surgery. **It assumes you want everyone to know you're calling.** Every lookup is public, there's no privacy. For humans calling businesses, this works fine. For AI agents that need to find, verify, and collaborate with other agents hundreds of times per second? It's like trying to run a modern stock exchange with a 1950s telephone switchboard. --- ## When AI Agents Take Over ### The Scale is Mind-Boggling Now imagine this isn't just Sarah. It's billions of people, each with their own AI agents, all trying to coordinate with each other simultaneously. We're talking about a world where: - Every person might have multiple specialized agents (travel, finance, health, work, shopping) - Every business has agents handling customer service, inventory, scheduling, and more - These agents need to find each other, verify capabilities, and work together in real-time If 80% of the world's population gets personal agents, that's 6.4 billion new "entities" that need to discover and connect with each other, treating each agent like a website, that's six times more entities than exist on the entire internet today. The current internet infrastructure simply isn't built to handle this explosion of intelligent, autonomous actors. --- ## Enter Project NANDA: A DNS for the Agent World This is where [Project NANDA](https://nanda.media.mit.edu/) comes in. An ambitious initiative from [MIT Media Lab](https://www.media.mit.edu/)'s Professor [Ramesh Raskar](https://www.linkedin.com/in/raskar/) to build the foundational infrastructure for what researchers call the "Internet of AI Agents." Think of NANDA as upgrading from a simple phone book to an intelligent directory system. But instead of just storing names and numbers, NANDA creates what's called an "index" that can handle the complex requirements of agent discovery. ### The NANDA Architecture NANDA solves the phone book problem through a three-layer system: **1. The Lean Index Layer** Instead of storing everything in one place (like DNS does), NANDA maintains only essential static information, think of it as a lightweight business card that points to where the real information lives. This card contains: - Agent identity and cryptographic signatures - Pointers to where detailed capabilities are stored - TTL(Time-to-live) settings for how fresh the information is **2. AgentFacts: Dynamic Capability Documents** This is where the magic happens. AgentFacts are like detailed, constantly updated resumes for each agent, stored separately from the index. They contain: - Real-time availability and endpoints - Verified capabilities and performance metrics - Authentication requirements and security credentials - All cryptographically signed to prevent forgery **3. Adaptive Resolution Layer** This layer handles the complex routing—deciding which specific instance of an agent to connect to based on location, load, capabilities, and other factors. ### Why This Architecture Works By separating static identity (the index) from dynamic capabilities (AgentFacts), NANDA can update agent information in seconds rather than hours. The system scales because most queries only need the lightweight index, while detailed capabilities are fetched only when needed. The cryptographic signatures ensure agents can't lie about their capabilities, while the modular design means the system can evolve without breaking existing implementations. --- ## What This Means for Your Future Within the next decade, you'll likely interact with the digital world primarily through AI agents rather than apps and websites. These agents will manage your schedule, handle your finances, book travel, and interface with various services on your behalf. For this to work safely and effectively, these agents need infrastructure like NANDA to find each other, verify capabilities, and coordinate complex tasks across organizational boundaries. Companies are already experimenting with agents for customer service, supply chain management, and more. But the real transformation will happen when agents from different companies can work together seamlessly—when your travel agent can coordinate with hotel agents, airline agents, and restaurant agents to plan your perfect vacation. ## Two Paths Forward We're at a critical juncture. The infrastructure decisions being made right now will determine which future we get: **Path A: The Open Agent Internet** NANDA and similar open protocols succeed. We get an interoperable agent ecosystem where agents from any company can work with agents from any other company. You're not locked into one tech giant's ecosystem, innovation happens rapidly through collaboration, and competition keeps services improving. **Path B: Agent Silos** Open protocols fail. Each tech company builds its own closed agent system. Your Google agents can't talk to your Apple agents. You're forced to choose one ecosystem, innovation slows, and a few tech giants control everything. The stakes are enormous similar to the early internet, but this time we're deciding how intelligent agents will make decisions on our behalf and shape our daily lives. --- ## Coming Up Next The current internet served us brilliantly for three decades, but it's showing its age when faced with the demands of autonomous AI agents. Project NANDA represents our attempt at building the infrastructure this new world requires infrastructure that's secure, private, open, and built to handle agent-to-agent collaboration at unprecedented scale. Over the coming weeks, I'll be diving deeper into different aspects of this transformation, and what are some of the problems that we are solving for other than Index. *The agent internet is coming whether we're ready or not. The only question is whether we'll build the right foundation for it.* {%preview https://nanda.media.mit.edu/ %}