New Toolkit Aims to Verify AI Agents as Human-Operated, Tackling a $5 Trillion Commerce Challenge
The rapid rise of AI agents handling tasks like booking travel, comparing prices, and making purchases has created a critical new problem for the internet: how can websites tell the difference between a legitimate automated assistant acting for a real person and a malicious bot network? A new developer toolkit, AgentKit, launched in beta today, offers a technical solution by merging identity verification with internet-native payments.
Bridging Identity and Payments for AI Agents
Developed by Sam Altman’s World, AgentKit integrates the company’s World ID verification system with the x402 open protocol, originally created by Coinbase and Cloudflare. The core innovation allows an AI agent to carry a cryptographic proof that it is acting on behalf of a unique, verified human—without revealing that person’s actual identity to the website or service it interacts with.
“Payments alone don’t prove uniqueness,” a World spokesperson explained. “A single bad actor could still operate thousands of fee-paying agents. AgentKit attaches a separate, privacy-preserving signal of human uniqueness to each request.” This signal can be used by developers alongside payment validations to enforce policies like “one human, one account” or to grant premium access, helping to filter out spam and automated abuse.
The Stakes: Trillions in Agent-Driven Commerce
The push comes as analyst firms project explosive growth in agentic commerce. McKinsey & Company estimates this sector could generate between $3 trillion and $5 trillion globally by 2030. Separately, Bain & Company has projected that AI agents could account for up to 25% of U.S. e-commerce sales by the end of the decade. This scale makes solving the identity verification problem not just a technical issue, but a foundational requirement for secure digital commerce.
The x402 protocol, promoted by the x402 Foundation, is designed to make payments a native part of web traffic, often using stablecoins. By combining x402’s payment proof with World’s proof of personhood, AgentKit provides a two-factor trust model for websites: the agent can both pay and prove it represents a distinct human.
How It Works: Delegated Verification
Under the model, a user who has already verified their identity with World (via its Orb biometric scanner or other methods) can formally delegate their World ID to a specific AI agent. When that agent then interacts with an API or website that supports AgentKit, it presents cryptographic credentials. The website can verify these credentials against World’s network without accessing the user’s personal data, confirming the activity stems from a real, unique individual.
This approach addresses a key limitation of current bot-fighting systems, which often rely on behavior analysis or CAPTCHAs that are ill-suited for autonomous agents. It also offers a privacy-centric alternative to traditional login systems, as no personal information is shared with the third-party site.
Scale and Ecosystem Context
World’s potential reach is significant. The company states its network now includes nearly 18 million verified humans across more than 160 countries. This existing installed base provides a ready-made pool of unique human identities for developers to leverage without building their own verification infrastructure.
The launch also fits into a broader industry trend. Major platforms and protocols are actively preparing standards for agent-to-machine commerce. By open-sourcing the integration with x402, World, Coinbase, and Cloudflare are collectively pushing for a more trusted, agent-friendly internet layer.
The beta release of AgentKit signals a move from theoretical discussions about AI agent identity to practical tooling. For developers, it offers a path to accepting beneficial automation without sacrificing security. For the projected multi-trillion-dollar agent economy, it attempts to build a crucial trust layer from the ground up.
Disclosure: This article was edited by Estefano Gomez. For more information on how we create and review content, see our Editorial Policy.
