Building an open Agentic Internet: readable, discoverable, callable, and payable

The web’s new visitor

The web was built for humans who render pages, click buttons, and scroll. But a growing share of traffic now comes from agents—programs that fetch pages on behalf of a person, read the content, and never see the layout. Cloudflare’s data shows billions of requests from well-behaved bots re-fetching pages that haven’t changed. That’s wasted machine effort attached to no outcome.

Unlike a human visitor, an agent doesn’t render CSS, see hero images, or click ads. But there’s a paying human behind every request. Block an agent and you block your customer; treat it like a scraper and you lose them. The web is being reshaped around this new kind of visitor, and Cloudflare argues it will need to become readable, discoverable, callable, and payable to stay open.

Identity and access: who’s asking

The first problem is knowing who’s knocking. Web Bot Auth lets a bot cryptographically identify itself to any site it visits, replacing guesswork and spoofed user agents. Many sites already know the human behind a request through login state or purchase history. Those sites can issue Private Access Control Tokens (PACT), a protocol announced with Mozilla, Google, Microsoft, and Shopify. PACT lets a site vouch for a visitor anonymously, so legitimate agents carry that proof elsewhere and hit less friction.

Domain owners keep control: a developer tool can become agent-ready to encourage discovery and payment, while a publisher can block extractive agents that consume resources without giving back, or allow them to pay to get unblocked. The old binary of bot versus human no longer applies—what matters is whether an agent’s actions are desirable.

Readable: cutting token waste

An agent burns tokens and bandwidth on every HTML tag rendered for a human who never looks at it. That waste also pollutes the agent’s context window. Markdown for Agents addresses this server-side by serving content in a format native to AI agents.

On the client side, Kitesurf is Cloudflare’s new browser, lean enough to run on Workers. It’s spun up per request and thrown away after, delivering content and features agents need without the weight of human-oriented browser features.

Discoverable: being found by machines

Before an agent can read a resource, call a tool, or pay, it has to know the resource exists. Search for agents means interfaces built for them, not keyword boxes designed for slow-typing humans. AI Search is available today, making any public site searchable by agents.

The other half is being discovered. Agent Engine Optimization (AEO) measures brand visibility across the models and agents that matter. If you’re not measurably visible to the agents your customers use, you’re effectively offline for them.

Callable: exposing actions directly

On the human web, booking a table or adding a to-do item means parsing HTML, guessing which button is “Add,” and synthesizing a click—fragile against any DOM change. WebMCP lets a site expose its actions directly to agents through the browser:

document.modelContext.registerTool({
  name: "add-todo",
  description: "Add a new item to the user's active todo list",
  inputSchema: {
    type: "object",
    properties: {
      text: { type: "string", description: "The todo item text" }
    },
    required: ["text"]
  },
  async execute({ text }) {
    await addTodoItemToCollection(text);
    return { content: [{ type: "text", text: `Added: "${text}"` }] };
  }
});

The tool contract becomes explicit. No HTML parsing, no guessing at form fields. Because tools run inside the page, they reuse the user’s existing session and state. Code Mode goes further: agents think in code, and calling tools by writing code is faster and more accurate than prose. When agents call endpoints rather than scrape pages, content owners also get a clear signal of what’s actually being used.

Payable: metering the agentic economy

Ad models are breaking, and seat-based licensing doesn’t work when the user is a program. Publishers can’t fund themselves on pageviews that never happen. The answer, per Cloudflare, is fine-grained payment at the point of use: a recipe site could charge a fraction of a cent per fetch and be profitable at agent scale; a local paper could license articles at read time without a login or licensing deal.

Every paid interaction leaves a receipt—the publisher can prove which agent fetched which page, and the agent can prove it paid. Wallets let agents pay for content and APIs; the Monetization Gateway lets domain owners set up agent payments in a few clicks.

These four primitives are built on open standards—x402, MCP, Web Bot Auth, PACT—that anyone can implement. Cloudflare positions itself as Customer Zero on those rails, with no privileged path or early-access API. Domain owners choose their own identity providers, payment processors, and agent partners.

Keeping the Internet open

The alternative future is one where a handful of stacks own discovery, identity, and payments, and everyone else routes through them. An open Agentic Internet needs diversity on both the supply and demand side—not just diverse publishers, but diverse agents. If the demand side converges, the supply side’s openness won’t matter.

Cloudflare’s agentic tools—the dashboard for getting sites agent-ready, AEO measurement, and the AI Playground for experimenting—are meant to push toward that open outcome. Whether the Internet stays open or closes will depend on how agents read, discover, call, and pay.