MIT’s Ramesh Raskar offers a fascinating vision of the future. Suppose a 70-year-old prediabetic woman living in rural India wants to attend a festival in a nearby city. She communicates this to her AI agent. In this future, everyone will have a personal AI agent. And then the AI agent comes into play. Find and buy her a train ticket. Negotiate the lowest hotel rates within walking distance of your clinic. Then, we will establish a menu according to her medical condition. Also, arrange for her to attend events that you think will be of most interest to her.
Trips are planned individually through an interactive web of AI agents. An AI agent is software that can understand goals, make decisions, complete transactions, coordinate with other agents, and act on behalf of individuals, organizations, objects, or systems.
Raskar leads Project NANDA, a research initiative that addresses agent identity, discovery, addressability, registries, validation, interoperability, and decentralized coordination. As a participant in the MIT Sloan speaker series “AI + X: How AI is changing management practice,” he described a digital world made up of billions or trillions of agents working together and competing to solve problems.
“Each of us will have our own agent, and each of us may have five or 10 agents, and every organization, every city, every refrigerator, every light, every car, every financial institution, every stock, every IPO, every baseball team will have its own agent,” Lasker said. “Will it take 10 years? Will it take five years? Will it take two years? I don’t know, but that’s the way it is.”
Raskar likened the evolution of AI agents to the evolution of computing. The mainframe era of large, centralized computer systems used by large organizations evolved into the PC era, where individuals could own their own personal machines.
“We are now in the mainframe era of AI,” says Lasker, an associate professor at the MIT Media Lab. This era is characterized by large data centers, centralized data and computing power. However, the cost of computational work is falling rapidly, and AI efforts will become accessible to more organizations and individuals. “We’re moving into the PC era of AI, and that’s going to change everything,” Rascal said.
An economy built around agents
One of Raskar’s key takeaways was that as we move into the next era of artificial intelligence, we need to reconfigure the way we think about the application of technology. He said most people and companies are building “agents of X” — agents that trade stocks, book travel, provide customer service, etc. The biggest opportunities aren’t there.
“The economy of the future is not one of, ‘Let’s create an agent for company X,'” he said. “It’s an economy that instead says, ‘Let’s create X for the agent.'”
This reversal will lead to a sprawling new market infrastructure designed to leverage billions or trillions of AI agents.
Rascal highlighted several categories that describe the economy he has in mind.
- Agent ID and detection system. These are similar to the Internet Corporation for Assigned Names and Numbers, which manages domain names and IP addresses. This type of service is essential for clear and free communication between agents.
- Trusted and reputable service. Similar to human passports, these services verify that agents are who they say they are and prevent malicious parties from interfering with transactions.
- Insurance, repair and legal services. Rascal pointed out that agents make mistakes, are sued, and deteriorate over time, so there will be a need for online services to repair agents and mediate issues and disputes.
- Stablecoin-based micropayments. Rascal foresees the emergence of a new economic infrastructure based on stablecoins that will support transactions between trillions of agents.
Business leaders need to stop thinking of agents as another app and not just focus on building agents, he said. Instead, you should focus on the marketplaces, protocols, and services your agents need.
Agentic AI: Business impact and applications
Face-to-face at MIT Sloan
Register now
Keeping the AI Agent Marketplace Open
The central question today is whether this evolution to an agency economy will remain open to the public or will be dominated by a small number of private companies. As a historical point of comparison, Raskar cited the World Wide Web. The World Wide Web was democratized by an underlying layer designed to communicate identity and trust between web pages, enabling interoperable protocols beyond the control of any single company.
Without efforts to keep this future version of AI open and decentralized, Lasker said, this future version of AI will look similar to most software today, or even mobile phones. Microsoft’s PowerPoint and Apple’s Keynote are not compatible, nor are iOS and Android. Even if people have great new ideas for how to design phones, they are bound by the current telecommunications architecture. Following the same logic, if the agent economy evolved in this way, people would have little or no choice of agents available to them.
Rascal said he is less optimistic about avoiding a future of centralized agents. “Honestly, I think nine out of 10 paths we walk will lead us.” [toward AI agents consolidated under corporate control]On this path, we will miss out on the vast potential of AI agents and will likely face the same problems we are currently seeing in areas such as social media, once promising but now dominated by half a dozen companies.
Still, Rascal is optimistic enough to pour his energy into NANDA. “The window to keep this agent’s net open is closing soon,” Lasker said. “But we haven’t closed yet.”
Learn more about Project Nanda
Ramesh Rascal He is an associate professor at the MIT Media Lab, where he Camera culture research group. he also Networked AI agent Build an “Internet of AI agents.” His focus is on distributed AI agent architectures and machine learning, and imaging for health and sustainability. His research spans physical (e.g. sensors, health technologies), digital (e.g. automated and privacy-aware machine learning), and global (e.g. geographic maps, autonomous mobility) research.
