As AI agents increasingly transact, shop, and act autonomously online — a market that can reach $3 trillion to $5 trillion by 2030 — a key issue comes into focus: how to verify that a real person is behind the activity.

Sam Altman–backed identity project World (formerly WorldCoin) says it has the solution.

On Tuesday, the company rolled out AgentKit, a developer toolkit that allows AI agents to carry cryptographic proof that they are backed by a unique human, using its World ID system. The product works with x402, a protocol developed by Coinbase and Cloudflare that enables “agentic payments” by embedding stablecoin micropayments into the internet’s communication layer so AI Agents and software can pay each other without human intervention.

“Payments are the ‘how’ of agentic commerce, but identity is the ‘who,’” said Erik Reppel, head of engineering at Coinbase Developer Platform and founder of x402. “This is a massive step toward a web where agents aren’t just seen as automated traffic, but as legitimate economic participants.”

The move comes as AI agents are rapidly evolving, handling time-consuming and often frustrating tasks from booking reservations to surfing e-commerce marketplaces for the best deals. Some estimates suggest agentic commerce could reach $3 trillion to $5 trillion by 2030, with agents accounting for up to 25% of U.S. e-commerce, World said.

Coinbase founder Brian Armstrong said he believes “very soon” there will be more AI agents than humans making transactions. Binance founder Changpeng Zhao went further, predicting agents will make one million times more payments than people, “and they will use crypto.”

The missing piece

However, as the agentic commerce market expands, its widespread use creates a problem that payments alone cannot solve: identity.

“One person could run thousands of agents that all pay small fees,” said DC Builder, a research engineer at the World Foundation. “Proof of Human addresses this gap.”

The World spokesperson explained that AgentKit addresses this by linking multiple agents to a single verified human, which allows platforms to impose limits at the identity level.

“AgentKit allows developers to link multiple agents to the same verified human,” the spokesperson said. “This means a platform can allow someone to run several agents while still enforcing limits based on the underlying person.”

That could enable services to cap usage, such as one free trial or a fixed number of bookings per day per human, regardless of how many agents are deployed, the spokesperson added.

Another problem with agentic commerce is that most websites treat automated traffic as suspicious and even block bots outright. That approach, designed to stop abuse, is increasingly at odds with a world in which legitimate software agents are gradually acting on user’s behalf.

AgentKit allows users to delegate their World ID, a privacy-preserving proof that they are a unique human, to AI agents acting on their behalf. And World is positioning this not as a replacement for other identity systems, but as a foundational layer.

“This isn’t necessarily an either-or choice,” a World spokesperson told CoinDesk. “World ID is designed to be a proof of human layer that developers can use on their own or alongside other identity systems.”

The system uses zero-knowledge proofs so platforms can verify that an agent represents a real person without collecting or storing personal data, a design World claims is required for scaling identity in an AI-driven web.

Beyond Orb verification

AgentKit, which is currently in beta version, relies on Orb-based biometric verification, the World’s most controversial component.

But the company says it plans to expand the system to include additional credentials. That will include NFC-enabled passports and IDs via “World ID Credentials,” allowing users to prove attributes about themselves without revealing personal information.

“Beyond the beta, we plan to expand AgentKit alongside the next generation of the World ID protocol,” the spokesperson said.

With the world’s real-time human verification meter reading at 17,912,203 at the time of writing, its networks rank among the largest proof-of-personhood globally. It also makes their broader ambition clear: to become the identity layer for an internet increasingly populated not just by people, but by the AI agents acting on their behalf.

Read more: Visa is ready for AI agents. So is Coinbase. They’re building very different internets