Skip to content

CLOUDFLARE 2026-08-06 Tier 1

Read original ↗

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

Summary

Cloudflare frames the "Agentic Internet" — the web reshaped for a new kind of visitor, the AI agent — around four properties it must acquire to stay open: readable, discoverable, callable, and payable. The motivating data point is that a large share of well-behaved-bot traffic is re-fetching pages that have not changed — billions of requests of machine effort with no outcome, "the signature of a web built for humans being visited by something else." Every agent request now has a paying human on the other end, so blocking it blocks a customer. Cloudflare's thesis is that these four primitives should be built on open standards anyone can implement (x402, MCP, Web Bot Auth, PACT) so no single company owns discovery, identity, or payments — and positions itself as the neutral, high-performance layer and Customer Zero of the same rails, "with no privileged path or early-access API that only we can reach."

Key takeaways

  • The Agentic Internet has four properties (Source: sources/2026-08-06-cloudflare-building-an-open-agentic-internet):
  • Readable — agents read content natively with fewer tokens/bandwidth. Server-side: Markdown for Agents. Client-side: Kitesurf, a browser lean enough to run on Workers, spun up per request and thrown away.
  • Discoverable — before an agent can read/call/pay, it must find the resource. Two halves: search (AI Search) and being discovered (Agent Engine Optimization / AEO measures brand visibility across the models and agents customers use). "If you are not measurably visible to the agents your customers use, then you are effectively offline for them."
  • Callable — agents do things (book a table, renew a subscription, pull a report). WebMCP lets a site expose actions directly to agents through the browser via document.modelContext.registerTool, making the tool contract explicit (no HTML parsing / button-guessing) and reusing the user's existing session/state. Code Mode goes further — agents call tools by writing code, faster and more accurate than prose, and gives content owners a clear signal of what is actually used.
  • Payable — ad-based and seat-based models break when the user is a program. A recipe site can charge a fraction of a cent per fetch; a local paper can license articles at read time without a login. x402 lets agents pay merchants directly; Wallets let agents pay for content/APIs; the Monetization Gateway lets domain owners accept agent payments in a few clicks. Every paid interaction leaves a receipt — the publisher can prove which agent fetched which page, the agent can prove it paid.
  • Identity is the starting point. Web Bot Auth lets a bot cryptographically identify itself to any site, ending guessing and spoofed user agents. Sites that already know the human behind a request (login, in-app behavior, purchase history) can issue Private Access Tokens (PACT) — announced with Mozilla, Google, Microsoft, and Shopify — to vouch anonymously, so a legitimate agent presents the token elsewhere and gets in with less friction.
  • "Bots are dead, long live bots." Cloudflare argues the bot-vs-human binary is outdated: agents are a new class of actor whose actions can be desirable (reading resources efficiently, respecting owner rules, paying for use) or undesirable (scraping without compensation, circumventing blocks, ignoring robots.txt). Giving both sides the right tools converts undesirable behavior into desirable behavior.
  • Open standards over a walled garden. All four primitives ride open standards (x402, MCP, Web Bot Auth, PACT) so the rails stay neutral. An open Agentic Internet needs diversity on both sides — publishers and agents: "If the demand side converges, it doesn't matter how open the supply side is." (See standards-convergence-over-vendor-extension.)
  • Selective agent policy per domain owner. A dev tool may want to become agent-ready to be discovered/recommended/paid; a publisher may block extractive agents but allow ones that license or compensate; a nonprofit data provider may block over-rate-limit traffic but let it pay to get unblocked to cover excess resource consumption — a concrete example of content-use signaling plus x402 pay-to-unblock.

Notable numbers / claims

  • "Billions of requests" of well-behaved-bot traffic are re-fetching unchanged pages — machine effort a domain owner paid to serve and an agent paid to make, with no outcome for either side.
  • Cloudflare has spent ~15 years doing the neutral-layer job for the human web and intends to do the same for the Agentic Internet.

Caveats

  • This is a strategy / philosophy post, not a system-internals deep-dive. It is an index into deeper posts (WebMCP, Code Mode, AI Search, Markdown for Agents, Web Bot Auth, PACT, x402, Kitesurf), each of which carries the architecture detail. Treated as in-scope because it defines the umbrella model (agentic-internet) that the other Cloudflare agent-week pages plug into, and contains concrete protocol/API surface (the WebMCP registerTool contract).
  • PACT and Kitesurf are lightly specified here; details would come from their dedicated announcements.

Source

Last updated · 766 distilled / 2,225 read