Cloudflare believes that AI agents are changing the nature of the web, not merely as a new type of software, but as a new type of visitor browsing on behalf of people or companies. The company says that much of the traffic generated by well-behaved bots repeatedly fetches pages that have not changed, in billions of requests that consume computing resources without a clear result.
Cloudflare calls the web taking shape around this type of visitor the “Agentic Internet.” In its vision, this internet should be readable, discoverable, actionable, and payable, relying on open tools and protocols that allow domain owners and agents to cooperate rather than collide.
Blocking Agents Could Mean Blocking Customers
An agent differs from a traditional browser: it does not render CSS, see hero images, or click ads. But it operates on behalf of a human who pays for its service, making every request tied to a goal and a financial cost. According to Cloudflare’s argument, treating these requests as conventional scraping could lead to losing the customer, rather than merely stopping a harmful source.
The company warns that today’s web, along with its analytics tools and associated business models, was not designed for this reality. It believes that the way agents read, discover, invoke, and pay for content will determine whether the internet remains open or becomes controlled by a limited group of platforms that own discovery, identity, and payments.
Four Pillars of the Agentic Internet
Cloudflare says its vision is based on open standards that any party can implement, including x402, MCP, Web Bot Auth, and PACT. It emphasizes that domain owners choose the identity providers, payment processors, and agent partners that suit them, and that Cloudflare is one option within the ecosystem, not the entire ecosystem.
Reading: Web Bot Auth allows a bot to identify itself cryptographically to websites, helping owners determine which agents they want to receive instead of relying on a spoofable User-Agent. The company also mentions “Private Access Control Tokens” (PACT), announced in collaboration with Mozilla, Google, Microsoft, and Shopify, which allow websites to vouch for agents anonymously.
Cloudflare also proposes Markdown for Agents to present website content in a way that reduces token and bandwidth consumption. Kitesurf, meanwhile, is a new browser that the company describes as lightweight enough to run on Workers, launching for each request and then being discarded, and providing agents with the content and features they need without browser features designed for humans.
Discovery: The company believes that accessing resources begins with knowing that they exist. It says AI Search is currently available, allowing any public website to be searchable by agents. It also mentions Agent Engine Optimization, or AEO, for measuring brand visibility across important models and agents; if a website is not visible to the agents customers use, it may effectively be invisible to them.
Execution: WebMCP allows a website to expose its actions directly to agents inside the browser. Instead of parsing HTML and guessing fields and buttons, an agent can use a clear “contract” for the tool, while benefiting from the user’s session and current state. Cloudflare says Code Mode goes a step further by enabling agents to invoke tools by writing code, with a clearer indication of the content actually used.
Payment: Cloudflare believes that advertising models and seat-based subscriptions do not always suit a user that is software. It proposes that an agent pay for each interaction, such as fetching a recipe, licensing an article, or using an API. The company says each paid operation leaves a receipt proving which agent fetched the page and what was paid for, while Monetization Gateway allows domain owners to configure payments from agents in a few steps.
Distinguishing Desired Agents from Harmful Ones
Cloudflare does not believe the traditional binary—bots are bad and humans are good—remains sufficient. An agent may be desirable when it reads content in a resource-preserving way, interacts with the website according to defined rules, and pays for what it uses. It may be harmful when it collects millions of pages without compensation, attempts to bypass blocking, or ignores robots.txt.
The company proposes giving domain owners tools to choose which agents they want to support. A developer may want to make it easier to discover and pay for their service, while a publisher may block agents that consume content without payment and allow those that license or compensate them. A nonprofit data provider might also allow use of its services after payment of a fee covering excess resource consumption.
Ultimately, Cloudflare presents this vision as an open alternative whose infrastructure is not owned by a single company. It invites website owners and agents to try the new internet technologies through the ready-to-use agent-ready website setup dashboard and AI Playground, and to sign up for news about the Answer Engine Optimization product.