Who will own the AI agent economy?

MIT’s Ramesh Raskar offers a compelling vision of the future. Suppose a 70-year-old pre-diabetic woman from rural India wants to attend a festival in a nearby city. She tells this to her AI agent — every person in this future has a personal AI agent — and it gets to work. It finds and purchases her train ticket; it negotiates the best rate on a hotel within walking distance of a health clinic; and it establishes a menu sensitive to her medical condition. It also arranges for her to attend the events that would most interest her.

The trip is planned with individual specificity through an interacting web of AI agents — software that can understand goals, make decisions, complete transactions, coordinate with other agents, and act on behalf of a person, organization, object, or system.

Raskar leads Project NANDA, a research initiative working on agent identity, discovery, addressability, registries, verification, 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 comprising billions or trillions of agents collaborating and competing to solve problems. 

“Every one of us will have our own agent, but every one of us could have five or 10 agents, and every organization, every city, every fridge, every light, every car, every financial institution, every stock, every IPO, and every baseball team — they’ll all have their own agents,” Raskar said. “Is it going to take 10 years? Is it going to take five years? Two? I don’t know, but that’s where it’s going.”

Raskar compared the evolution of AI agents to that of computing: The mainframe era of large, centralized computer systems used by large organizations evolved into the PC era, when individuals could own personal machines. 

“Right now, we are in the mainframe era of AI,” said Raskar, an associate professor at the MIT Media Lab. This era is marked by big data centers and centralized data and computing power. But the cost of computational work is falling quickly, which will make AI efforts accessible to more organizations and individuals. “We are moving to the PC era of AI, and that changes everything,” Raskar said. 

An economy built around agents

One of Raskar’s key points was that as we move into the next era of artificial intelligence, we need to reorient how we think about the technology’s application. Most people and companies, he said, are building “agents for X” — agents for stock trading or booking travel or providing customer service. That is not where the biggest opportunities will be.

“The economy of the future is not an economy where we say, ‘Let’s create agents for X,’” he said. “It’s an economy where we instead say, ‘Let’s create X for agents.’”

This inversion will lead to a sprawling new market infrastructure designed to take advantage of billions or trillions of AI agents. 

Raskar highlighted a handful of categories that illustrate the economy he has in mind:
 

• Agent identity and discovery systems. These would be analogous to the Internet Corporation for Assigned Names and Numbers, which manages domain names and IP addresses. This kind of service is essential to clear and unfettered communication between agents.
• Trust and reputation services. Like human passports, these services would confirm that agents are who they say they are and prevent bad actors from injecting themselves into transactions.
• Insurance, repair, and legal services. Raskar noted that agents will make mistakes, get sued, and degrade over time, so online services for repairing them and mediating problems or disputes will be needed.
• Stablecoin-based micropayments. Raskar foresees the emergence of a new economic infrastructure based on stablecoin that supports transactions between trillions of agents.

Business leaders need to stop thinking of agents as another app, and they shouldn’t focus only on building agents, he said. Instead, they should be focusing on the marketplaces, protocols, and services those agents will require.

Agentic AI: Business Implications and Applications

In person at MIT Sloan

Register now

Keeping the AI agent marketplace open

The central question today is whether this evolution toward an economy of agents will remain open to ordinary people or be dominated by a handful of private corporations. As a historical point of comparison, Raskar pointed to the World Wide Web, which was democratized though a foundational layer designed to communicate identities and trust between web pages and allowed for interoperable protocols beyond the control of any single company.

The alternative version of this AI future — absent efforts to keep the agentic web open and decentralized — would look more like most of today’s software, or like cellphones, Raskar said. Microsoft’s PowerPoint and Apple’s Keynote are not compatible, nor are iOS and Android. Even if people have terrific new ideas about how to design a phone, they are locked into current telecommunications architectures. Following the same logic, if the agent economy evolves in this way, people will have little to no choice in the agents available to them.

Raskar said he is not terribly optimistic about avoiding a centralized agentic future. “If I’m honest, I think nine out of 10 paths we take will lead us [toward AI agents consolidated under corporate control],” he said. On this path, we will lose the vast potential of AI agents and likely see the same problems we now see in sectors like social media — realms that, once promising, are now controlled by a half-dozen companies.

Yet Raskar is optimistic enough to put his energies into NANDA. “The window to keep this web of agents open is closing soon,” Raskar said. “But it’s not closed yet.”

learn more about Project Nanda 

Ramesh Raskar is an associate professor at the MIT Media Lab, where he directs the Camera Culture research group. He also leads work on networked AI agents to build the “internet of AI agents.” His focus is on distributed AI agent architectures and machine learning, and imaging for health and sustainability. His work spans research in physical (e.g., sensors, health tech), digital (e.g., automated and privacy-aware machine learning), and global (e.g., geomaps, autonomous mobility) domains.