World's AgentKit Puts a Human Check on AI Shopping

Imagine an AI agent books your flights, orders your groceries, and pays for API calls on your behalf — all without you touching a browser. That future is close enough that it creates a real problem: how does a merchant know there’s an actual human responsible for those transactions?

World launched AgentKit today to answer that question. It’s a developer toolkit that lets AI agents carry cryptographic proof they’re backed by a verified human — using World ID as the identity layer and zero-knowledge proofs as the verification mechanism.

How the Verification Works

World ID verification starts with an iris scan at a World Orb device. The Orb captures the iris, generates an encrypted IrisCode (a numerical representation of the iris pattern), then deletes the original images. What persists is the IrisCode — unique to the person, not reversible to the original image.

AgentKit lets developers link that verified World ID to one or more AI agents cryptographically. Multiple agents can act on behalf of the same verified person, but a merchant or platform can see that they all trace back to a single human identity. That enables per-human usage caps, fraud detection, and accountability — none of which work when you’re dealing with anonymous AI agents.

The privacy piece matters: the verification doesn’t require sharing names, email addresses, or any personal data. The zero-knowledge proof confirms “this agent is backed by a unique verified human” without revealing who that human is.

The x402 Integration

AgentKit integrates with the x402 protocol, developed by Coinbase and Cloudflare, which embeds stablecoin micropayments into HTTP — essentially making internet requests payable at the protocol level. Agents can autonomously purchase API access, pay for compute, acquire premium data, or pay for storage without requiring a human to authorize each transaction. The protocol has processed over 50 million transactions.

The combination of World ID (identity) and x402 (payment) is what World is calling a “complete trust stack” for agentic commerce: agents can pay for things AND the things they’re paying for can verify a real human is ultimately responsible.

Why This Matters More Than It Sounds

The gap AgentKit fills is subtle but structural. As GPT-5.4 mini and nano and similar models make agent deployments cheaper, the volume of autonomous agent transactions will scale rapidly. The accountability infrastructure hasn’t kept up.

Merchants dealing with AI agents today have no good way to distinguish one user running fifty agents from fifty users running one agent each. That distinction matters for fraud, for pricing, for compliance. AgentKit creates the layer that makes that distinction possible without requiring any personal data disclosure.

OpenAI’s agentic models are doing more and more autonomously. The infrastructure for verifying human intent behind those actions — who’s responsible when an agent buys the wrong thing, or buys too much, or gets defrauded — is a genuine gap. World has 17.9 million verified humans with World IDs already. That’s the network that makes AgentKit meaningful rather than theoretical.

AgentKit is in beta for developers now, built in partnership with Coinbase. General availability pricing hasn’t been disclosed. The beta is focused on payment gateways, marketplaces, and agent platform providers — the integrations where the human-verification problem is most acute.