The Internet Was Built for Humans; Its Next Billion Users Won't Be

AI Internet User

Imagine waking up on a Monday morning and telling your AI: “I need to be in Seoul on Thursday. Find a sensible flight, book a hotel near my meetings, use my preferred airline if the price difference is reasonable, and keep the entire trip below $1,500.”

Then you go back to work.

Your agent searches flights, compares hotels, checks your loyalty programs, evaluates cancellation policies, confirms that each supplier meets your preferences, books the itinerary, pays for it, updates your calendar, and sends you the final schedule.

You never open an airline website. You never visit a hotel booking platform. You never enter your credit card details. You never even see a checkout page.

For thirty years, we have been designing the internet around a relatively simple assumption: somewhere on the other side of the screen is a human.

That assumption is beginning to break.

AI agents are emerging not simply as another application running on the internet, but as a new class of internet participant—systems capable of discovering services, communicating with software, making decisions, and increasingly transacting on behalf of people and businesses.

Cloudflare recently gave this emerging architecture a useful name: the Agentic Internet. Its vision is an internet that becomes readable, discoverable, callable, and payable by agents. More importantly, this is no longer just a conceptual architecture. Cloudflare has begun introducing infrastructure that gives AI agents persistent identities and programmable wallets, while payment networks including Visa and Mastercard are building mechanisms for agents to initiate real transactions under human-defined permissions [1][2][3].

The internet is getting a new user. And much of today's internet was never designed for it.

A Web Designed for Eyes and Fingers

Consider how much of the modern internet assumes human behavior.

Websites have navigation menus because humans need to find information. Product pages have photographs because humans want to see what they are buying. Search engines rank pages because humans need help discovering them. Advertisers compete for impressions because human attention is scarce. Checkout pages are optimized to reduce the number of clicks between desire and purchase.

Even many of the internet's security mechanisms are built around proving that a visitor is human. CAPTCHAs may be the most obvious example.

The irony is difficult to miss.

For years, websites have effectively asked visitors: “Prove that you are not a machine.”

We may now be entering an era in which businesses increasingly need to ask: “Which machine are you, who do you represent, and what are you authorized to do?”

That is a fundamentally different internet.

Cloudflare says its infrastructure is already seeing billions of requests from well-behaved bots repeatedly fetching webpages that have not changed—machine effort wasted partly because the web remains optimized for human consumption rather than efficient machine interaction. It describes agents as a new kind of visitor requiring different infrastructure [1].

This distinction matters because today's AI assistants largely experience the internet indirectly. They search it, scrape it, interpret it, or interact with interfaces originally built for us. The next generation may increasingly expect the internet to meet them halfway.

From Browsing to Calling

Suppose your AI agent wants to find the cheapest available hotel room.

Today, it might search the web, read pages, interpret listings, compare information, and potentially navigate a booking flow. That is remarkably similar to teaching a machine how to imitate a human browsing the internet.

There is another possibility.

The hotel, airline, retailer, bank, software platform, or data provider could expose structured capabilities that an authorized agent can discover and call directly.

Instead of navigating a webpage to understand a service, the agent asks the service what it can do. Instead of reading pricing from a page, it queries pricing. Instead of clicking through checkout, it invokes a transaction.

The shift of agentic internet

This is why the protocol layer emerging around AI agents may ultimately matter as much as improvements in the models themselves.

Our earlier Tech Epoch article on The Language of AI explored Model Context Protocol and the growing need for standardized ways for AI systems to communicate with tools. That was one part of the problem.

The Agentic Internet expands the question. An agent needs to know what services exist, what they can do, how to communicate with them, who is allowed to use them, what they cost, and whether the agent has permission to pay.

Google's current guide to agent protocols illustrates how quickly this stack is developing. Different protocols are emerging around tool access, agent-to-agent communication, commerce, interfaces, and payments; its Agent Payments Protocol, for example, uses mandates to encode who authorized a purchase, what constraints apply, and when human approval is required [4].

This begins to look less like a chatbot browsing the internet. It begins to look like a new machine-readable economic layer being built on top of it.

When Your Customer Is an Algorithm

Now consider what happens to digital competition.

For much of the commercial internet's history, companies have competed for human attention. Search Engine Optimization helps businesses appear where humans search. Advertising puts products in front of humans. User experience reduces friction for humans. Branding creates preference in human minds. Recommendation algorithms attempt to influence what humans see next.

An autonomous agent has very different priorities.

It does not care whether your homepage has beautiful photography. It does not get distracted by a banner advertisement. It does not impulse-buy because the checkout button is red.

It may instead evaluate price, specifications, delivery time, reliability, contractual terms, historical performance, compatibility, availability, security, and whatever preferences its owner has instructed it to prioritize.

Parts of the internet economy may gradually shift from competing for human attention to competing for machine preference.

The implications are significant. SEO does not disappear, because humans will continue discovering and evaluating products. Branding does not disappear either. Neither does conventional user experience. But another optimization problem begins developing alongside them.

Businesses may need to ask not only whether humans can understand their products, but whether agents can. Is the product information machine-readable? Can an agent verify inventory? Can it understand contractual terms? Can it determine whether a product meets specific technical requirements? Can it authenticate itself? Can it transact? Can another machine verify the provenance of the information it receives?

Mastercard is already arguing that merchants will need machine-readable product information and permissioned preference signals so agents can evaluate offers accurately [3].

The website of the future may therefore have two customers: the human who wants to experience the brand—and the machine that wants to understand the business.

But Machines Need Something Humans Take for Granted: Identity

Giving an agent access to information is relatively easy. Giving it authority is much harder.

Imagine your procurement agent ordering $300 of office supplies. Now imagine it ordering $3 million of manufacturing equipment.

The fundamental question changes from whether the AI can make the decision to whether it has the authority to make it.

Who is the agent? Which individual or company does it represent? What is it authorized to purchase? How much can it spend? Which suppliers can it use? How long does that authority remain valid? What happens when instructions are ambiguous? And who is responsible when the transaction goes wrong?

These questions explain why identity may become one of the most important infrastructure layers of the Agentic Internet.

Cloudflare's wallet architecture addresses part of the problem by giving agents stable identifiers and allowing owners to define spending policies. Mastercard's Verifiable Intent approach similarly attempts to connect identity, the user's instructions, and the resulting action into an auditable record [2][5].

This is not simply payment infrastructure. It is delegation infrastructure.

The internet already knows how you authenticate. Now it needs mechanisms for answering a much stranger question: How does something acting for you prove what you allowed it to do?

Trust May Become More Valuable as Humans Leave the Loop

There is an understandable temptation to assume that increasingly autonomous systems remove friction from commerce. They probably will. But they may simultaneously make trust infrastructure much more important.

When you personally click Buy, intent is relatively obvious. When an agent buys something because of a conversation you had six hours earlier, intent becomes harder to establish.

Did you authorize that exact purchase? Did the agent correctly interpret your request? Did the merchant know it was dealing with an authorized agent? Was the transaction within the permitted budget? Was the AI manipulated by malicious information? If there is a dispute, who can prove what happened?

Consumers appear to recognize this intuitively. Mastercard's August 2026 research found that 85% of consumers were open to collaborating with an AI agent to find the best option and 74% were open to agents completing specific commerce tasks at their request. Yet only one in ten were willing to allow an agent to complete a purchase autonomously [3].

That gap is revealing.

The technological capability may arrive before the social permission. In other words, the bottleneck may not be intelligence. It may be trust.

What Kind of Money Does a Machine Use?

Once agents can discover services, establish identity, and prove authorization, another surprisingly fundamental question appears.

How do they pay?

For a $1,000 hotel booking, existing payment rails can evolve to accommodate agents. Visa says more than 50 issuers and payment partners across Asia-Pacific are already testing agent-led transactions, while its Singapore Agentic Ready program includes 13 banks and fintechs testing agent-initiated transactions in a controlled environment [6][7]. Visa has also announced a collaboration with OpenAI to support secure Visa payments within agentic commerce environments [8].

But machine commerce may eventually include a very different class of transaction.

An agent could pay for an API request. Another agent's specialized analysis. A few seconds of compute. A proprietary dataset. Storage. Inference. Bandwidth. A verified identity check. Perhaps fractions of a cent at a time.

At sufficient scale, the internet could contain enormous numbers of economic interactions that are too small, too frequent, or too automated to resemble conventional human commerce.

That is where programmable money becomes particularly interesting.

Stablecoins are one possible architecture. Tokenized bank deposits are another. Existing card and banking rails may evolve substantially. New payment protocols may bridge several forms of money simultaneously. The winner is far from obvious.

But the deeper idea matters more than the payment technology: the internet may need money that software can understand as naturally as it understands data.

Payments become another programmable instruction: spend no more than this amount; only transact with verified providers; pay automatically if these conditions are met; request approval if they are not.

Now money itself becomes part of the agent's operating environment.

The Internet May Split Into Two Experiences

None of this means humans disappear from the internet.

Quite the opposite.

Humans will continue wanting beautiful experiences, compelling brands, communities, entertainment, stories, and the messy serendipity that makes the internet interesting.

But underneath that human internet, another internet may increasingly develop.

One where machines discover services directly. Where APIs become storefronts. Where identity represents delegated authority. Where protocols replace some interfaces. Where reputation becomes machine-readable. Where money becomes programmable. Where transactions can occur without either party touching a screen.

The two layers will constantly interact.

Humans define intent. Agents translate intent into action. Machines negotiate the execution. Humans intervene when judgment, exception, or approval is required.

The internet becomes less a destination we visit and more an economic environment our software continuously operates within.

The Next Billion Users

The great technology platforms of the internet era were built around successive waves of human users coming online.

First through computers. Then smartphones. Then billions of connected devices.

AI agents represent something different.

They are not another population of people gaining internet access. They are software gaining agency within the internet itself.

That distinction could eventually reshape everything from cybersecurity and payments to advertising, search, APIs, enterprise software, and digital identity.

The infrastructure being built today may look fragmented: an agent protocol here, an identity system there, an experimental payment mechanism somewhere else.

That is usually how infrastructure begins.

The web itself once consisted of comparatively simple protocols connecting comparatively simple pages. Then an economy was built on top of them.

We spent the last thirty years building an internet where machines served humans.

We are now beginning to build an internet where machines also discover, trust, serve, and transact with each other.

The companies that understand that transition early may not simply build the next generation of websites.

They may build the infrastructure for an economy whose newest customers aren't human at all.

References

[1] Cloudflare, “Building an open Agentic Internet: readable, discoverable, callable, and payable,” August 6, 2026. Source

[2] Cloudflare, “Announcing Cloudflare Wallets: the programmable wallet for the agentic Internet,” August 4, 2026. Source

[3] Mastercard, “Building trust for agentic commerce: Mastercard Signals report explores the path forward,” August 2026. Source

[4] Google Developers Blog, “Developer’s Guide to AI Agent Protocols,” March 18, 2026. Source

[5] Mastercard, “When AI starts buying for you, trust becomes the product,” March 5, 2026. Source

[6] Visa, “Visa Launches ‘Agentic Ready’ Program in Asia Pacific with Over 50 Partners,” April 30, 2026. Source

[7] Visa, “Visa Launches Agentic Ready Programme in Singapore with 13 Banks and Fintech Partners,” April 30, 2026. Source

[8] Visa, “Visa Partners with OpenAI to Power the Next Generation of AI Commerce,” June 11, 2026. Source