World is pushing deeper into the emerging “agentic web” with the launch of AgentKit, a developer toolkit designed to let AI agents prove they are backed by real humans. The beta release integrates with the x402 protocol, an open standard backed by Coinbase and Cloudflare, combining identity verification with payment infrastructure for automated systems.

The timing is not accidental. As AI agents begin handling tasks like booking services, comparing prices or interacting with APIs, the internet is running into a basic problem: how to tell the difference between useful automation and abuse at scale.

Right now, most platforms treat automated traffic as suspicious by default. That approach blocks malicious bots — but it also limits legitimate agents that are supposed to act on behalf of users. AgentKit is an attempt to fix that without sacrificing privacy.

What AgentKit actually does

AgentKit allows developers to create what World calls “human-backed agents”. These are AI systems that can prove, cryptographically, that a real and unique person is behind them — without revealing who that person is.

The system works through World ID, which provides a privacy-preserving identity layer. A user can delegate their identity to an AI agent, and that agent can then present proof of uniqueness when interacting with websites, services or APIs.

Instead of exposing personal data, the system relies on cryptographic verification. Platforms get a simple signal: this agent represents one real human — not a bot network.

That distinction is becoming increasingly important as AI agents move from passive tools into active participants in online transactions.

Investor Takeaway

Identity is emerging as a missing layer in the AI economy. Projects that can prove “one human = one agent” without sacrificing privacy may become critical infrastructure as agent-driven commerce scales.

Why payments alone are not enough

The integration with x402 highlights a key point: payments solve access, not identity.

The protocol already allows AI agents to make micropayments to access APIs or services. Since launching in 2025, the ecosystem has processed over 100 million payments, showing early traction in agent-based interactions.

But payments do not answer a more fundamental question: how many real people are behind those agents?

One user could theoretically deploy thousands of agents, each capable of paying small fees. From the platform’s perspective, that looks like demand — but it may actually be coordinated activity.

AgentKit adds a second layer of verification. Websites can now request proof of a unique human alongside — or instead of — payment before granting access.

This gives platforms a way to limit abuse based on real users rather than raw activity volume.

Where this could be used

The model opens up a range of practical use cases where identity matters more than speed.

Reservation platforms could allow agents to book on behalf of users while preventing bots from hoarding slots. Ticketing systems could ensure purchases are tied to real individuals instead of automated scalping networks. Services offering free trials could allocate access per human rather than per wallet.

In each case, the goal is the same: separate legitimate activity from mass automation without blocking useful agents.

AgentKit also supports more advanced identity signals, such as age or country of residence, using zero-knowledge proofs. That means platforms can verify specific attributes without collecting sensitive personal data.

Investor Takeaway

If agent-based commerce grows as expected, identity layers could become as important as payment rails. Combining both in one stack gives developers a clearer path to building real-world applications.

The bigger shift toward agent-based commerce

Industry estimates suggest that AI-driven commerce could reach between $3 trillion and $5 trillion globally by 2030, with agents potentially accounting for a significant share of online transactions.

If that happens, the internet will need new rules for participation. Today’s model — where bots are blocked and humans are assumed — does not scale when software starts acting on behalf of users.

World’s approach is to introduce a middle ground: agents are allowed, but they must prove they represent real people.

The company says its network already includes nearly 18 million verified users across more than 160 countries, providing a base layer for this identity model.

AgentKit is still in beta, and the current version builds on existing World ID infrastructure. A more advanced version is expected as the protocol evolves.

For now, the direction is clear. As AI agents become economic actors, identity is moving from a background feature to a core requirement — and the platforms that solve it may define how the next phase of the internet operates.

AgentKit beta is available now to developers building AI agents who hold a verified World ID. Documentation and developer access are available here. 

The current beta version is built on the existing World ID architecture, with a more advanced version planned as the next generation of the protocol is released. The initial rollout is intended to gather feedback from developers and demonstrate how proof of unique human can extend to the growing ecosystem of autonomous agents.