Your Next Customer Is an AI Agent

We built AI capable of extraordinary intelligence, and then taught it to click websites like a human.

Imagine a multimillion-dollar humanoid robot sitting at a desk.

It leans toward a monitor, focuses its cameras on the screen, lifts a mechanical arm, grips a mouse and slowly moves a cursor toward a button.

Click.

The image is absurd.

We have built an extraordinarily sophisticated machine, then forced it to imitate one of the least efficient parts of being human: sitting at a computer and navigating a graphical interface.

Yet this is essentially what we are doing with AI today.

The internet was built for humans. Websites are designed around our eyes, our hands, our attention spans and our psychology. Navigation menus tell us where to go. Buttons tell us what to click. Colours establish hierarchy. Videos keep us engaged. Forms collect our information. Countdown timers create urgency.

Now AI agents are arriving in that environment.

And because the environment wasn’t built for them, we are teaching machines to behave like us.

They render pages. Interpret layouts. Search menus. Read text. Fill out forms. Click buttons.

We are entering what might be called the humanoid phase of the web. And it probably won’t last forever.

Why We Build Humanoid Robots

There is a useful parallel in robotics.

From a purely engineering perspective, the human body is often a terrible machine design.

Standing upright on two legs is unstable. Walking requires constant balance adjustments. Hands are extraordinarily complicated. If your only objective is to move pallets around a warehouse, you would never design a human.

You would build a forklift.

If you wanted to weld car frames, you would build a stationary robotic arm.

So why are companies spending billions developing humanoid robots?

Because the world already exists.

Our buildings have stairs built for human legs, doors built for human hands and shelves positioned for human reach. If a robot is going to operate inside an environment designed around people, one solution is to make the robot resemble the people who built it.

The same thing is happening online.

The modern web is a human environment. So today’s AI agents are becoming digital humanoids. They are being trained to use browsers, navigate websites and operate interfaces designed for people because rebuilding the entire internet overnight is impossible.

But history suggests that once machines become important enough, we eventually stop forcing them to adapt to human environments. We redesign the environment around the machines.

Amazon Already Showed Us What Happens Next

Consider the modern automated warehouse.

When Amazon began automating fulfillment at enormous scale, it could have built thousands of humanoid robots capable of walking through aisles, pushing carts and reaching onto shelves.

Instead, it changed the warehouse.

Small robots move underneath storage pods, lift them and bring the inventory to workers.

The important innovation was the environment itself, redesigned around the machine.

Aisles existed because humans needed somewhere to walk. Once the primary movement of inventory no longer depended on humans walking through the warehouse, the assumptions underlying the warehouse could change.

The internet may be approaching a similar moment.

Today, AI agents still navigate websites because that is where the world’s information and commercial infrastructure lives.

But if a growing proportion of digital activity is performed by machines, why should machines continue interacting through interfaces built for human eyes?

Eventually, they won’t.

Instead of navigating pages, an agent could connect directly to structured data. Instead of searching for a pricing page, it could query pricing. Instead of filling out a checkout form, it could initiate a transaction. Instead of reading fifty vendor websites, it could evaluate fifty vendors programmatically.

That is the transition from the human web to the agentic web. And it changes much more than website design. It changes what a customer is.

Three Kinds of Bots

Businesses already see enormous amounts of automated traffic, which has created a fairly understandable reaction: Block the bots.

But that framing becomes dangerous if every machine visitor is treated as the same thing.

1. The crawler

A crawler collects information. Its purpose may be indexing, model training or large-scale data acquisition. It requests pages, extracts their contents and moves on.

From the perspective of a publisher, this can become an intellectual-property problem: the machine consumes information without necessarily sending a human visitor back.

2. The retriever

A retriever is acting in response to a particular user’s request.

Ask an AI system to research the latest results from a company, for example, and it may retrieve information from relevant sources, extract what it needs and synthesize an answer.

The important difference is that the human may never visit those websites. The information travels to the user rather than the user travelling to the information.

3. The actor

The third category is much more consequential for commerce.

An actor acts.

It might compare products, check availability, make a reservation, schedule a meeting, submit information, purchase something or execute part of a procurement workflow.

And that means the machine visiting your website may be your customer’s representative.

Blocking it indiscriminately could be the digital equivalent of locking the front door when a buyer arrives.

The Customer Is Moving to the Two Ends

Human beings stay at the heart of commerce.

Humans still have the goals. Humans still hold the budgets. Humans still decide what matters. Humans still possess taste, judgment, relationships and accountability.

What changes is where the human sits in the buying process.

For most of the internet era, the person was in the middle of almost everything. They searched. They visited websites. They read articles. They downloaded reports. They compared features. They looked at reviews. They sat through demonstrations. They assembled spreadsheets. They narrowed ten or twenty possibilities into three. Then they decided.

AI lets the human move toward the two ends of that process.

At the beginning, the human expresses intent:

At the end, the human exercises judgment:

The enormous amount of work in between can increasingly be delegated to the machine.

That seemingly small change has enormous consequences for marketing and sales.

The Invisible Shortlist

Imagine you sell enterprise software.

Under today’s model, a prospective customer researching your category leaves footprints everywhere.

Someone from the account searches Google. They visit your website. They read three articles. They download a report. They attend a webinar. They return to the pricing page.

Your marketing automation system notices. Their lead score rises. Sales receives an alert. A rep reaches out.

Modern revenue organizations have spent decades getting better at reading these digital footprints.

Now imagine that discovery is performed by an AI agent.

The buyer asks it to evaluate the market against a detailed set of requirements.

The agent examines fifty vendors. It checks integrations. Security certifications. Product specifications. Contract parameters. Customer evidence. Pricing.

Forty-seven don’t meet the requirements. Three survive. Those three are presented to the human.

The entire process might happen without any of the fifty vendors knowing that an active buying process existed.

You may lose deals you never knew you were in.

There may be no abandoned form. No anonymous visitor spending four minutes on your pricing page. No webinar registration. No MQL. No intent signal for sales to pursue.

The machine evaluates you, fails to find something important, excludes you and moves on.

Your analytics may record little more than a fleeting automated request.

That is the invisible shortlist. And getting onto it requires a different kind of marketing.

From Persuasion to Proof

The modern commercial website is, among other things, a persuasion machine.

Every component has been optimized for human psychology.

Brand imagery creates emotional associations. Testimonials provide social proof. Visual hierarchy directs attention. Countdown timers introduce urgency. Calls to action encourage progression. Conversion-rate optimization teams test everything from headline language to button placement.

But an AI agent is an entirely different audience.

Your orange button, the fourteen other people looking at the same hotel room and the expensive brand video on your homepage all leave it unmoved.

When acting against explicit buying criteria, it wants something much simpler: Evidence.

Can the product perform the required function? Does it integrate with the required system? Does it hold the required certification? What does it cost? What contractual restrictions apply? What evidence supports the vendor’s claims? Can those facts be discovered, parsed and verified?

Branding stays.

Humans still make consequential decisions. Trust, reputation, identity, design, narrative and relationships still matter, especially once a small number of options reaches the person making the decision.

But the role of branding may shift.

The machine increasingly influences who gets considered. The human increasingly determines who gets chosen.

Why Gated Content Becomes Dangerous

This also creates a problem for one of B2B marketing’s favourite inventions: the form.

For years, companies have intentionally created friction around valuable information.

Want the research report? Give us your email address.

Want the detailed comparison? Fill in the form.

Want pricing? Talk to sales.

The bargain made sense when a human researcher was on the other side. The vendor sacrificed some reach in exchange for identifying the buyer and initiating a sales process.

An autonomous agent may interpret the same experience very differently.

It is trying to complete a task.

If it has been instructed to compare ten products under $150,000 and your website says only Contact Sales, the agent has a problem.

It has no reliable number.

Perhaps it searches third-party sources. Perhaps it estimates. Or perhaps it simply chooses another vendor whose pricing it can verify.

The same problem applies to technical specifications trapped inside difficult PDFs, security information behind registration walls or product capabilities that only become visible after a demonstration.

The information may exist. But to the machine evaluating you, it may effectively be missing.

In the agentic web, friction can become invisibility.

The Great Business Paradox

At first glance, the solution appears obvious.

Make everything agent friendly.

Publish structured product information. Expose specifications. Make pricing accessible. Build APIs. Allow trusted agents to transact programmatically.

But there is a problem.

For many businesses, friction is profitable.

Consider airlines.

The airline website sells you a seat from Toronto to London. It is also an environment designed to sell upgrades, credit cards, insurance, hotels, rental cars and additional services.

The interface itself is monetized.

Now imagine telling an AI travel agent:

The agent needs inventory, price, seat availability and payment capability. The homepage and six pages of upsells are in its way.

If it can transact directly with the airline’s systems, the airline receives the booking, and may lose much of its opportunity to influence the customer’s behaviour around that transaction.

The product becomes easier to compare. Suppliers become easier to substitute. And the AI platform controlling the customer relationship becomes more powerful.

This creates the central strategic tension of the agentic web.

Become agent friendly

Make your information easy for machines to access and understand. You increase the probability of being evaluated and included. But you may also make your product easier to compare and commoditize.

Become agent hostile

Force traffic through the human interface. Protect your customer relationship, your data and potentially your higher-margin business model. But you risk being excluded when customers delegate discovery and purchasing to agents.

Neither extreme is particularly attractive.

What should agents be allowed to discover, compare and execute? And where should the human relationship remain uniquely valuable?

The Web May Split in Two

The likely destination is a web with two increasingly distinct layers.

One layer remains designed for people.

It communicates ideas. Builds trust. Tells stories. Educates. Creates desire. Explains complexity. Supports relationships.

The other layer is designed for machines.

Structured product data. Pricing. Availability. Technical specifications. Identity. Permissions. Transaction protocols. APIs. Machine-readable evidence.

A company may eventually need both.

The human layer answers: Why should I trust you?The machine layer answers: Do you satisfy the requirements?

And businesses that confuse those two audiences may struggle with both.

The Infrastructure Is Starting to Matter

A truly agent-native economy also requires capabilities the traditional web was never designed to provide elegantly.

An agent needs identity.

A merchant needs to know what system is requesting access and, potentially, on whose behalf it is acting.

It needs permissions.

A business may want an authorized purchasing agent to access negotiated pricing while preventing the same software from accessing proprietary internal information.

And it needs financial authority.

Instead of handing an AI an unrestricted corporate credit card, future payment infrastructure can theoretically give agents tightly scoped authority: spend up to a certain amount, with approved vendors, for a particular purpose, before a specified deadline.

Those mechanisms transform agents from software that can merely browse into software that can safely act.

Once that infrastructure becomes reliable, clicking around websites begins to look increasingly unnecessary.

In Some Ways, We’ve Been Here Before

Ironically, the agentic web may be a return to an older idea.

Long before the web became dominated by visual interfaces, businesses were already connecting computers directly to other computers.

Electronic Data Interchange allowed corporate systems to exchange purchase orders, invoices and inventory information automatically.

Procurement systems used punchout catalogs to connect buyers directly with supplier databases.

Machine-to-machine commerce is decades old.

What is new is the intelligence sitting on top of it.

Traditional systems needed rigid workflows and predefined transactions.

AI agents can interpret natural-language goals, research unfamiliar environments, reason across alternatives and potentially decide which workflow to execute.

We may therefore be combining two eras of computing: the machine-readable infrastructure of the old internet with the flexible intelligence of modern AI.

So What Should Businesses Do Now?

It is too early to tear down every website and replace it with an API.

Humans still browse. Search still matters. Sales still matters. Brands still matter. Relationships certainly still matter.

The humanoid phase could persist for years precisely because so much economic infrastructure is built around human interfaces.

But companies should begin asking a second set of questions alongside the traditional ones.

• Does our homepage convert?→Can an agent understand what we sell?
• Can we capture the visitor’s email address?→Can a buyer’s agent retrieve enough information to consider us?
• Does the website tell a compelling story?→Are our important claims backed by machine-readable evidence?
• How much traffic are we generating?→Are we present when machines construct the shortlist?

That last question may become one of the most important metrics in marketing.

Because your next customer may never visit your website.

Their AI might.

And by the time the human arrives, the most important part of the sale may already have happened.

The invisible buyer is already arriving.

The question now reaches past the people who understand and trust your digital presence, to the agents acting for them: can they discover, verify and transact with it?