Agentic Web: World ID with Okta, Vercel

The shift from human-driven to agent-driven actions online is accelerating. Agents are already browsing the web, accessing APIs, completing purchases and executing multi-step workflows on behalf of the people who deploy them. The internet is being restructured around this reality, and the companies building the infrastructure for it are moving fast.

The trust layer is not keeping pace. The systems agents operate on were built for humans, not AI. Signing up for a service requires an account. Accessing an API requires credentials. Making a purchase requires a payment method tied to a person. And when there’s an issue, there is no reliable way to prove that a human approved the action the agent took. The missing primitive is proof of human: a way for any service to verify, cryptographically and without collecting personal data, that a real, unique human stands behind an agent and its actions.

World is bringing three new capabilities that expand proof of human into agentic workflows with AgentKit: agent delegation, human in the loop and agentic commerce. Together, they provide developers the primitives to build agents that carry proof of the human behind them, can request verifiable human approval for sensitive actions, and can transact on behalf of verified humans while establishing a new type of trust layer.

World’s human in the loop, featuring verifiable intent powered by Vercel

As agents manage higher-stakes tasks, critical workflows require verifiable human intent. In payments, card networks and payment protocols are building cryptographic proof that an account holder authorized a transaction. But not every consequential agent action is a purchase. Such actions can include accessing sensitive data, signing contracts, modifying infrastructure, and approving deployments. Not every context has a payment network to anchor trust. What has been missing is a universal primitive that proves a real human approved a specific action, interoperable across all workflows and platforms.

The pattern is familiar: an agent makes a large purchase and the user disputes it; the merchant eats the chargeback. An agent modifies production infrastructure at 3am and takes down a service; no one can tell whether an engineer approved the change. In each case, an agent acted, something went wrong and there is no record that a human was in the loop.

World’s human in the loop, powered by World ID, addresses this challenge. It lets any agent workflow request a zero-knowledge cryptographic proof, independently verifiable at any point in the future, that a unique real human authorized a specific action. Because the proof is zero-knowledge, services get high-integrity verification with no tradeoff: no personal data to store, no PII liability, and no risk of data leakage.

World is teaming up with Vercel to bring human in the loop to developers building on Vercel's new open source Workflow SDK. Developers can add a step to any workflow or agent to require human verification, and every verification is viewable inside each workflow execution for full auditability. The result is a provable record that a human was in the loop when it mattered most. This feature is live today, and it can be installed from npm.

Securing proof of human with Okta’s Human Principal

Developer products with free offerings and social media platforms are vulnerable to abuse. AI makes scaled abuse and fraud easier than ever. As agents continue to go mainstream, these problems are expected to exacerbate. Without a way to distinguish the human behind agents, other options can be aggressively rate-limiting or per-product heavyweight KYC processes, both of which add friction, can introduce data privacy risks, and undermine the value of automated agentic workflows.

World ID enables a verified human to delegate their proof of human to an agent so it can act on their behalf. Services receiving requests from that agent can verify that a real, unique human is behind it, without collecting any personal data.

Okta is planning to build a new product: Human Principal. Human Principal will allow API builders to verify whether a human stands behind an agent and its actions, and enforce policies accordingly. Humans will be able to verify themselves using a number of verification methods, and obtain device-bound cryptographic proof that carries across products without requiring cumbersome re-verification. World ID, slated to become one of the first Human Principal integration partners, is set to provide Human Principal a privacy-preserving, user-friendly, and ubiquitous proof of human verification method.*

Once combined, World ID and Human Principal will enable features like rate limits per human for agent traffic, abuse-protected free tiers, and a cleaner onboarding flow for agents that need to access services on behalf of their human principal. The waitlist to join the upcoming Human Principal beta is now open at humanprincipal.ai

Browserbase and Exa integrated proof of human for agent delegation

As enterprise discovery and adoption shifts toward agent-led growth, where AI agents evaluate, compare, and transact on behalf of buyers, companies face a new tension: they need to offer agents robust free tiers and frictionless access to win selection, but that same openness creates a wider surface for abuse. World ID helps resolve this by letting companies distinguish verified human-backed agents from unverified traffic, enabling generous access without the risk. We're excited to be tackling this with two leading AI enterprises today.

Browserbase, which builds a platform for agents to automate the web, has also integrated agent delegation. When an agent using Browserbase carries a World ID, it receives the benefits of verified traffic, reducing blocks and friction as it navigates the web. Without it, the agent encounters standard anti-bot protections. The integration is live for developers to use.

Exa, which builds AI-native search, offers 100 free API requests monthly to agents verified through AgentKit. Via verified agents, backed by World ID, developers can experiment with Exa's search and content endpoints without upfront payment, while being protected against abuse. Once the quota is exhausted, the agent falls back to standard x402 payments. The integration is now live for developers to use.

Agentic commerce with proof of human: one human, one agent, one allocation

Commerce is where the stakes of unverified agent activity are most immediate. Flash sales with limited inventory, exclusive product releases, promotional pricing: these are high-value moments that bad actors already exploit with bots. Shops want to accommodate AI agents acting on behalf of real humans, but it's created a new challenge of distinguishing legitimate agent traffic from unwanted activity.

AgentKit now lets a buyer delegate their World ID to an agent (agent delegation) so it can find products online, claim coupons, and facilitate the purchase process – while human in the loop ensures the agent cannot make purchases above a set threshold without explicit human approval. This lets merchants use World ID to obtain confirmation of one human per allocation, limiting how many units each buyer can purchase and confidently open storefronts to agent traffic knowing it is backed by real people. Agent delegation and human in the loop are designed to prioritize legitimate buyers, drive more revenue, and prevent fraud.

A World demo showed how World ID can enable a fully agentic shopping experience when used with the Universal Commerce Protocol (UCP), the open standard for agentic commerce co-developed by Shopify and Google. The demo highlighted how Shopify merchants could verify that a purchase is backed by a unique, verified human. One person, one agent, one allocation.

Developers interested in trying the agentic commerce skill can explore at: https://github.com/worldcoin/agentkit-shopify-demo.

Building for agents with World ID

Agent delegation, human in the loop, and agentic commerce are available for developers via AgentKit: https://docs.world.org/agents/agent-kit/integrate

The shift from human-driven to agent-driven interaction is accelerating. The question is whether the trust infrastructure can keep pace. World ID provides the foundation: proof that a real human is behind every agent, every action and every transaction that matters.

*Any mention of future products, features, functionalities, or certifications in this blog is for informational purposes only. These items are not commitments to deliver and should not be relied upon to make purchasing decisions.