In Brief
- World launched AgentKit beta, a toolkit that lets AI agents carry proof they are backed by a unique human.
- AgentKit aims to help websites distinguish productive agents from bot swarms and other abusive automation.
World has introduced AgentKit beta, a developer toolkit meant to solve a problem that is moving quickly from theory to practice: how websites can tell whether an AI agent is acting on behalf of a real person or as part of a bot swarm. The product ties World ID, the company’s proof-of-human system, to x402, the machine-payments protocol developed by Coinbase and later advanced with Cloudflare, giving websites a way to request payment, proof of a unique human behind the agent, or both.
The timing is not accidental. McKinsey estimates that AI agents could mediate $3 trillion to $5 trillion of global consumer commerce by 2030. Bain projects AI agents could account for 15% to 25% of U.S. e-commerce sales, or roughly $300 billion to $500 billion, by 2030. Those forecasts help explain why payment companies, cloud providers, retailers and identity networks are suddenly racing to build the rails for what is increasingly being called agentic commerce.
World’s pitch is that the web has a missing trust layer. In its announcement, the company said AgentKit allows verified World ID holders to delegate their credential to AI agents so those agents can carry cryptographic proof that a unique human is behind them, without revealing who that person is. World describes this as “programmable proof of unique human,” and says the toolkit is aimed at helping websites distinguish legitimate, human-backed agents from scripts, scalpers and spam.
That framing goes to the heart of why this launch matters. The internet already has tools for blocking bots, rate-limiting traffic and charging for access. What it has lacked is a broadly usable way to express that a software agent represents one real person rather than one operator controlling hundreds or thousands of agents. For many online services, that distinction matters more than whether the request was automated in the first place. A merchant may be willing to let an AI assistant shop on a customer’s behalf, but not to let one actor flood checkout flows, promotions or inventory with coordinated bots.
The product is built to sit on top of x402, not replace it. Coinbase and Cloudflare said in September 2025 that they planned to launch the x402 Foundation to standardize an open internet protocol for digital payments built around the long-dormant HTTP 402 “Payment Required” response code.
Cloudflare said the standard is intended to make it easier for websites, developers and content creators to request and receive payments online, especially in agent-driven and machine-to-machine contexts.
Coinbase’s Erik Reppel, who created x402 and leads engineering for Coinbase Developer Platform, has described agentic commerce as a “once-in-a-generation opportunity to rethink how value moves online.”
Cloudflare’s technical description of x402 makes clear why the protocol has attracted so much attention in AI circles. In its simplest form, a client tries to access a gated resource, the server responds with a 402 and payment instructions, the client sends back a payment authorization header, and the transaction is verified and settled so the resource can be delivered. Cloudflare says the result is programmatic access to resources across the internet without requiring the traditional account, subscription or API-key model. It has also integrated x402 into its Agents SDK and MCP tooling, underscoring that the protocol is being positioned for software agents as much as for human users.
But payment is only one half of the problem. A micropayment can meter usage, yet it says very little about how many people are behind a wave of agent traffic. If the economic upside is large enough, an operator can simply pay to run large volumes of requests. That is the gap World is trying to fill. According to World’s documentation, AgentKit beta extends x402 so websites can admit agent traffic “without falling victim to spam,” distinguish human-backed agents from bots and scripts, and enforce rules such as limited free trials. The reference integration supports payments on World Chain and Base, registers agents on Worldchain through a registry called AgentBook, and includes a default “free-trial” mode with three uses.
That means a verified World ID holder can register an agent wallet, complete the World App verification flow, and then allow that agent to present proof that it traces back to a unique human whenever it interacts with a participating website. The site can then choose to grant access, offer a discount, allow a fixed number of free requests, or require payment.
It is important to note that multiple agents can still trace back to the same human, so a website can see that a hundred agents belong to one person rather than mistakenly treating them as a hundred independent users.
“Payments are the ‘how’ of agentic commerce, but identity is the ‘who,’” Reppel said. “By integrating World ID with the x402 protocol, developers now have a complete trust stack: a way for agents to pay for what they need and a way for platforms to verify there is a real human behind the wallet.”
That quote captures the strategic logic of the launch: World is trying to position proof of personhood as a core primitive of the agentic web, next to payments and communication protocols.
The broader market is moving in a similar direction, though with different architectures. Google said in late 2025 that its Agent Payments Protocol, or AP2, was designed as a universal protocol for secure AI-driven payments and that the company had worked with Coinbase, the Ethereum Foundation, MetaMask and others on an A2A x402 extension for agent-based crypto payments. Google said AP2 was intended to create “a trusted foundation” for AI commerce and specifically pointed to adjacent opportunities such as agent authorization and decentralized identity.
Visa has taken a parallel route. In October 2025, the payments network introduced Trusted Agent Protocol, developed with Cloudflare, as a framework for agentic commerce that helps merchants verify agents and distinguish commerce-intent traffic from malicious bots. Visa said the move came amid a 4,700% surge in AI-driven traffic to U.S. retail sites and cited merchant concerns around guest checkout, bot detection and visibility into the consumer behind the agent. That makes World’s move notable not because it is the only company pursuing agent trust, but because it is bringing a proof-of-human model into a field that has so far been dominated by payments and merchant-authentication approaches.
The commercial pressure to solve this is already visible. Reuters reported this week that a U.S. appeals court temporarily put on hold a lower-court ruling that had blocked Perplexity AI from using its agentic shopping tool on Amazon while the case is reviewed. Amazon has accused Perplexity of covertly accessing customer accounts and disguising automated activity as human browsing, while Perplexity says users should have the right to choose their own AI tools. Regardless of how that litigation ends, the dispute shows that agent traffic is no longer a future scenario for merchants and platforms; it is becoming a live operational and legal issue.
That is the reason World is framing AgentKit as an enablement product rather than merely a security feature. The company’s argument is that many websites will not fully open themselves to agent traffic unless they can control it. A blanket block on automation may be too restrictive in a world where software increasingly acts for real consumers. But a system with no way to tell one person’s assistant from a bot farm is equally unworkable. World is effectively betting that proof of unique human can become the compromise: websites do not need to know a user’s identity, only that there is one distinct person behind the agent
For now, AgentKit remains in beta and is aimed at developers already building inside the emerging x402 ecosystem. But the launch is important because it highlights where the next layer of competition may sit in AI commerce. The first phase was about making models useful. The second is about giving agents wallets and payment rails. The next phase may be about deciding which agents the web will actually trust.
World’s answer is that payments alone are not enough. In the company’s view, the agentic web also needs a way to count humans without naming them. If that logic takes hold, proof of personhood could become part of the basic operating stack for online commerce, much as fraud checks, identity prompts and payment authentication did in earlier phases of the internet.
More From BlockFirms
Disclaimer: This article is for informational purposes only and does not constitute investment advice. Read our Editorial Policy. Parts of this article were drafted/ researched with the assistance of AI tools and subsequently reviewed, edited, and verified by the author and our editorial team to ensure accuracy and journalistic integrity. The final version reflects human editorial judgment and fact-checking. Read our AI Policy.
Image Credits: World, Canva
