The AI Agent Economy: Building Infrastructure, Not Just Apps

The future of artificial intelligence isn't about building better chatbots or smarter assistants. According to MIT researcher Ramesh Raskar, the real transformation will come when billions—or trillions—of AI agents need infrastructure to interact, transact, and coordinate with each other.

Raskar, an associate professor at the MIT Media Lab who leads Project NANDA, presented a vision of this agent-driven economy at MIT Sloan's "AI + X" speaker series. He described a world where every person, organization, device, and even financial instrument has its own AI agent capable of making decisions, completing transactions, and negotiating with other agents autonomously.

Why it matters

Most companies are racing to build AI agents for specific tasks—customer service bots, trading algorithms, travel planners. But Raskar's framework suggests the larger economic opportunity lies in creating the foundational services these agents will need to function at scale. This shift in perspective could determine whether the agent economy remains open and competitive or becomes controlled by a handful of tech giants.

From mainframe AI to personal AI

Raskar drew a parallel to computing history. Just as mainframe computers gave way to personal computers, he believes we're transitioning from today's centralized AI—dominated by large data centers and cloud providers—to a "PC era of AI" where computational work becomes cheap enough for individuals and smaller organizations to deploy their own agents.

"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. The timeline remains uncertain—perhaps two, five, or ten years—but the direction is clear.

The infrastructure opportunity

Raskar's central thesis challenges conventional thinking: "The economy of the future is not an economy where we say, 'Let's create agents for X.' It's an economy where we instead say, 'Let's create X for agents.'"

This inversion points to several emerging categories:

Agent identity and discovery systems analogous to domain name registries, enabling agents to find and communicate with each other across networks.

Trust and reputation services that verify agent identities and prevent malicious actors from infiltrating transactions—essentially digital passports for software.

Insurance, repair, and legal services to handle inevitable agent failures, degradation, and disputes.

Stablecoin-based micropayment systems designed to support transactions between trillions of agents without traditional banking friction.

Raskar illustrated the potential with a scenario: a 70-year-old pre-diabetic woman in rural India tells her AI agent she wants to attend a festival. The agent coordinates with other agents to book transportation, negotiate hotel rates near a health clinic, arrange appropriate meals, and schedule events matching her interests—all through an interacting web of specialized agents.

The centralization risk

Raskar warned that without deliberate effort, this agent economy could follow the path of smartphones or productivity software—locked into proprietary ecosystems controlled by a few corporations. He pointed to the World Wide Web as a counterexample, where open protocols and interoperability standards prevented any single company from dominating.

"If I'm honest, I think nine out of 10 paths we take will lead us [toward AI agents consolidated under corporate control]," Raskar acknowledged. Yet he remains committed to Project NANDA's work on agent identity, verification, and decentralized coordination. "The window to keep this web of agents open is closing soon," he said. "But it's not closed yet."

These details were first reported by MIT Sloan's Ideas Made to Matter.