The web was built around a fairly simple assumption: there is a human at the other end of the request.
We created HTML to present information to people. Browsers rendered it. Search engines indexed it so people could find it. Ecommerce added forms, logins, carts and checkout so people could transact.
AI agents break that assumption.
An agent does not necessarily want to look at a website. It wants to understand what an organisation offers and, increasingly, do something.
That may sound like a subtle distinction. I think it could become one of the more important changes to the web since search.
The web is getting a second type of user
AI traffic is already growing quickly. DataDome reported processing 17.7 billion AI-agent requests in the second quarter of 2026, up 45 percent quarter-on-quarter. HUMAN Security has separately reported enormous growth in traffic from AI agents and agentic browsers. Cloudflare says automated agents and bots now account for more than half of requests across its network.
The measurements are different and should not be compared directly, but the direction is clear: an increasing share of the web is being consumed by machines rather than people.
That alone is not new. Search-engine crawlers have been reading websites for decades.
What is new is intent.
A crawler reads a page so that a human can find it later. An AI agent may be reading the page because it wants to act now.
It might be trying to find a hotel, obtain an insurance quote, move a flight, buy a replacement part, compare suppliers or submit an application.
That is a very different type of internet user.
Humans want interfaces. Agents want capabilities.
Imagine I tell an agent:
Find me a good hotel near Circular Quay for Thursday night, make sure it has parking, and book it if it is under $450.
Today an advanced agent can attempt this by using a website much like a person. It can inspect the page, identify fields, click buttons, enter dates, compare prices and work its way through checkout.
That is impressive.
It is also slightly absurd.
The hotel's systems already know how to search rooms, check availability, return prices, make reservations and take payment. Yet the AI may be reading HTML or interpreting pixels to work out which rectangle on the screen means Book.
We are making computers pretend to be humans so they can communicate with other computers.
That feels transitional.
Cloudflare has described the emerging agentic internet in terms of sites becoming readable, discoverable, callable and payable by agents. That framing makes sense to me.
The long-term answer is unlikely to be ever-better robots clicking websites.
It is more likely to be businesses exposing their knowledge and capabilities directly.
A website currently does two jobs
Most company websites combine two things.
Information: what the company offers, what products exist, what they cost, what the policies are, what is available and what the customer needs to know.
Interface: menus, buttons, forms, navigation, search, checkout and all the other mechanisms humans use to interact with that information.
People need both.
Agents mainly need the information plus machine-callable actions.
That suggests an increasingly plausible split:
Human: Browser → Website UI → Business systems
AI: Agent → Machine interface → Business systems
The organisation remains the same. The underlying data and systems remain the same. But the interface is completely different.
This is already starting to happen
This is no longer just a theoretical architecture.
Anthropic introduced the Model Context Protocol as a standard way for AI systems to connect to external data and tools. MCP has since expanded well beyond Anthropic, with support from major AI and cloud companies and thousands of public servers now available.
Cloudflare is working on WebMCP and related infrastructure intended to let websites expose capabilities to agents directly rather than requiring them to reverse-engineer a human interface.
Commerce is moving in the same direction. OpenAI and Stripe have developed the Agentic Commerce Protocol. Google has introduced its Universal Commerce Protocol with retailers and commerce platforms, and has positioned it as infrastructure for agentic transactions.
We are starting to see the outline of a parallel web stack.
HTML and traditional web interfaces for humans.
Protocols, APIs, structured knowledge and tools for agents.
That is a much bigger change than adding a chatbot to a website.
We now need to know why the machine is here
Cloudflare's recent changes to AI crawler controls make another important distinction.
Machine traffic is no longer simply a question of bot or human. Infrastructure increasingly needs to understand what the machine is trying to do.
Is it indexing a page for search?
Is it collecting information for AI training?
Is it acting in real time on behalf of a user?
Those are very different activities.
A company might happily allow an AI assistant to read a product page on behalf of a customer while refusing permission for the same page to be copied into a training dataset.
It might allow an authenticated agent to obtain a quote but not to place an order.
It might allow another agent to purchase something, but only up to a spending limit set by the user.
The web therefore starts needing answers to a new set of questions.
Who is this agent? Who authorised it? What is it allowed to know? What is it allowed to do? Can it spend money? Can it accept terms? Can it represent a person legally? Can its actions be audited?
That begins to look less like web browsing and more like delegated digital authority.
SEO may not be enough
For more than twenty years businesses have worried about one central digital question:
Will Google find us?
The next question may be:
Will AI agents understand and choose us?
Cloudflare and others are already talking about Answer Engine Optimization as organisations try to understand how AI systems consume and cite web content.
But I think that only describes the first stage.
If agents move from answering to acting, businesses will care about three things.
Discoverability: can an agent discover that the business provides the product or service?
Comprehension: can the agent correctly understand the products, pricing, policies, restrictions and availability?
Actionability: can the agent actually do something once it has decided this is the right business?
A company could rank beautifully in conventional search while being almost unusable by autonomous agents.
The website is not dead
It is tempting to take this argument too far and declare that websites are disappearing.
I do not think they are.
People will still want to browse travel, restaurants, fashion, journalism, entertainment, property and thousands of other experiences where presentation matters.
But a large amount of internet activity is utilitarian.
Renew my insurance.
Find the cheapest replacement toner.
Move my flight to Friday.
Pay this invoice.
Find three suppliers that meet these requirements.
Order the same filters I purchased last time.
For these tasks, I do not necessarily want to visit a website at all.
I want the outcome.
My agent may initially visit the sites for me. Eventually even that may be unnecessary if it can interact directly with the businesses.
The relationship then changes from:
Human → Website → Business
to:
Human → Agent → Business
The website remains, but it is no longer necessarily the primary digital interface.
This creates a problem for the economics of the web
Much of the commercial internet is built around human attention.
A search engine sends a person to a publisher, where the person sees advertising. A retailer gets the person onto a product page where it can recommend related products. A booking site carefully designs an interface to influence what the customer chooses.
Now imagine I tell an agent:
Find me the best washing machine under $1,500.
The agent researches dozens of retailers and comes back with one recommendation:
Buy this one. $1,279, delivered tomorrow.
I never visit any of the websites.
I never see the advertising, sponsored placement, newsletter popup, upsell or carefully optimised product page.
That is not a small change.
It potentially alters an internet economy built around page views and attention.
Cloudflare is already experimenting with mechanisms such as pay-per-crawl and stronger controls over machine access to content. Over time, the economic unit of the web could start shifting from the page view toward machine access, citation, transaction or action.
Brands may have to convince an agent
There is another strange implication.
Companies spend enormous amounts of money creating brand preference in humans.
But what happens when an AI mediates the purchase?
If I say:
Buy me decent dishwasher tablets. Do not spend more than $30.
What determines the winner?
My previous purchases? Price? Reviews? Product quality? Availability? Whether I have historically cared about the brand?
The company is no longer only trying to influence me. It also needs to be correctly understood by the decision-making system representing me.
That could create an entirely new discipline somewhere between marketing, structured data, reputation management and AI optimisation.
And, inevitably, people will try to game it.
SEO spam may eventually be joined by attempts to manipulate agents into selecting particular products, sources or services.
Identity becomes critical
An anonymous browser can read a public page.
An agent taking actions needs something stronger.
Businesses will need to distinguish a search crawler from a training crawler, a commercial agent, a malicious bot and an authenticated agent genuinely acting for a customer.
That requires identity, permissions and auditability.
We might eventually give agents very specific authority.
My agent can read my electricity usage. It can obtain competing quotes. It can negotiate. It can spend up to $500. But it cannot terminate my existing contract without asking me.
Those are not browser permissions.
They are delegated authority.
Building that safely may prove to be one of the harder parts of the agentic internet.
The machine version of a business may actually be simpler
There is an interesting irony here.
The modern web became extraordinarily complicated partly because people require rich interfaces.
An AI agent does not care about the hero image, animated navigation, responsive layout or carefully designed checkout button.
Its view of a company may ultimately be much simpler:
Here is who we are.
Here is what we know.
Here is what we sell.
Here are the actions you can perform.
Here are the permissions required.
Here is how you pay.
That is remarkably close to the direction emerging agent protocols are already taking.
What does your business look like to an AI?
For thirty years businesses have asked:
What should our website look like?
Then:
How do we rank on Google?
More recently:
How do we appear in ChatGPT?
But those are still questions from a world in which AI mainly helps people find information.
Agents introduce a different question:
What does our business look like to a machine that wants to do something?
The web is not going away.
But it may be acquiring something it has never really had before: a machine-native interface sitting alongside the human one.
And if agents become an important intermediary between people and businesses, building that interface may eventually matter as much as building the website itself.
Sources
DataDome: AI Traffic Report, Q2 2026
Cloudflare: Your Content, Your Rules
Cloudflare: The Agentic Internet
Anthropic: Model Context Protocol