In 1996, a Dutch engineer, Geert-Jan Bruinsma, noticed something about hotels. Travellers couldn’t compare them in parallel. A visitor opened tab after tab, scribbled prices on a notepad, and settled for whatever looked reasonable. Bruinsma built Booking.com to fix this. In 2005, Priceline bought it for $133 million. The company today is worth $140 billion. A fortune built on a single assumption: that the traveller would rather pay than compare.

Bruinsma didn’t create value. He harvested friction. Markets have been capitalising this inertia for decades, calling it brand equity, customer retention, and product stickiness. A renewal earned is a product doing its job. A renewal extracted is a tax on the customer’s attention. The ledger records both as loyalty. A better name is friction value.

Three rents

The first rent is laziness. The seller can quietly raise prices without improving the product, because switching is more work than staying. General insurance in Australia offers an example. In 2021, the regulator found that insurers had promised loyalty discounts in their renewal notices and systematically failed to apply them. The gap was revenue earned from the near-certainty that most customers wouldn’t bother to check. Insurers are now repaying $815 million.

The second rent is search cost. It arises when options are hard to see in parallel, and an intermediary charges to do the sorting. Comparison sites are the canonical form. Booking.com takes commissions of 15 percent or more on the rooms it books, a fee justified by the idea that the alternative is a traveller opening seven tabs. Price aggregators in insurance, energy, and credit cards do the same work at thinner margins. The commission is what the customer pays for not looking.

The third rent is cognitive overload. It arises when a product is too complex to use, audit, or leave, and the vendor knows more about that use than the buyer does. Enterprise software is the purest form. Audits of corporate SaaS portfolios find that close to 40 percent of paid licences go unused, because consumption data is fragmented across a dozen vendors and procurement teams cannot work out which to cut. Large Oracle and SAP contracts renew on autopilot, because the cost of auditing them exceeds the cost of paying another year.

Across all three, the trick is to make the rent look like loyalty.

The rails of agentic commerce

An AI agent is software that can search, compare, and pay on the user’s behalf, without handing the task back to a browser. In September 2025, OpenAI and Stripe launched the Agentic Commerce Protocol, or ACP. In January 2026, Google followed with the Universal Commerce Protocol, or UCP. Amazon is building a closed equivalent inside its own ecosystem. The three are racing to become the rails of agentic commerce. Most of the firms affected don’t yet know the tracks are being laid.

The agent does not veg out in the summer or fondly recall the last boozy lunch with the broker. It does not decide it’s too late at night to compare insurers. It does exactly what it is told to do, for as long as the task requires. Friction value feeds on the customer who is, in one way or another, done for the day. The agent is never done.

Consider the average household. There’s a gym membership unused since January. Two streaming services, one forgotten after the latest instalment of Bridgerton. A home insurance premium that quietly rose 12 percent at renewal. A mobile plan two years older than the current equivalent. Each depends on the same assumption: that the customer can’t be stuffed. An agent asked to audit the household’s recurring spend can cancel or renegotiate the lot. That is friction value collapsing in an hour.

Matching versus making

The test reduces to a single question: what does the fee buy? If it buys matching, comparison, or discovery, an agent will kill it. If it buys a physical service the agent cannot perform, the fee survives.

Booking.com’s 15 percent commission is “comparison” rent. The traveller pays a toll to avoid opening seven tabs. A sliver of the fee buys trust: payment guarantees, refund handling, assurance that the hotel is not a scam. An agent operating under ACP or UCP can do most of that. It can search, reserve, pay, and confirm the booking itself.

Uber Eats is a hybrid, with two revenue streams inside one commission. The first is “discovery” rent: restaurants pay to be visible on a crowded marketplace, and customers pay for the privilege of browsing it. An agent strips that margin out. The second is the dispatch network: riders, routing, and the physical act of moving food from kitchen to front door. An agent can optimise a route but it cannot ride a bike. What remains is a logistics business on thinner margins.

Airbnb, by contrast, has no riders. Discovery goes for the same reason it goes at Booking.com. Trust deserves more attention. Airbnb’s trust layer has three parts: guest reviews, host verification, and insurance with a complaints process. The first two are information an agent can read. The third is a commodity dressed as a feature. Travel insurance is cheap and disputes are a process, not a moat. Airbnb bundled them at checkout. An agent rebuilds that checkout on the guest’s side for free. The listings remain. The 15 percent does not.

Fees for matching collapse. Fees for making survive.

Where the rent goes

Rent does not vanish. It migrates.

Whoever builds the dominant agent inherits the choke point Booking.com and the auto-renewing insurer once held. The consumer is now represented by OpenAI, Google, or Amazon: intermediaries with their own commercial incentives. When the agent reports it has found “the best deal,” it will often mean the merchant that paid most to be shown. The consumer discovers, as they did with search, that representation was never free.

The collapse is also an equity repricing. Share markets have long priced human inertia into balance sheets as goodwill, brand value, and customer lifetime value. The share of S&P 500 market value attributable to intangible assets has risen from 17 percent in 1975 to roughly 90 percent today. A material slice of that 90 percent is a long-dated put option on human laziness. The investor assumed the customer would renew, would not compare, would not switch. That assumption was written into every discounted cash flow model that treated retention as free upside. Agents make the option worthless.

Let’s return to Booking.com. A large share of its $140 billion valuation rests on a single recurring event: a traveller arriving at the site, opening the tabs, paying the toll. Strip the commission away and what remains is a directory business competing on listings and price. The market currently applies a commission multiple. Directories don’t earn one. When the commissions shrink, the share price follows.

The repricing will be uneven. Laziness rents go first. Search-cost rents follow. Cognitive-overload rents take longest, because enterprise procurement runs on annual cycles, multi-year contracts, and committee sign-off. The first signal will be goodwill writedowns: companies admitting they paid for customer bases less loyal than they seemed.

The new tollbooth

The rails will not be neutral. For an agent to cancel your insurance, it must prove to the insurer that it is really you, not a bot scraping data. That verification layer will not settle on a single global standard. Apple, Google, and the card networks are each pushing their own, and the prize is the tollbooth between the agent and every transaction. Whoever wins inherits a position richer than Visa’s. Agentic commerce will have a gatekeeper, and the gatekeeper will charge.

The ad market, split

Global ad spend will reach $1.2 trillion in 2026. Not all of it survives. The split is between persuasion at the moment of comparison and persuasion before it begins. Search ads, sponsored listings, and the retargeting that follows a shopper from basket to checkout all sit in the first category. Whoever sold the slot was selling friction value: the chance a customer might waver.

Brand works upstream. It shapes what the customer asks the agent to find. A buyer who wants Nike shoes, a Lexus, or a Diageo whisky arrives with preferences already set. You cannot flatter an API. But you can plant the word Nike in someone’s head before they instruct one.

Marketing splits accordingly. The function that survives is brand: the hard work of planting a story long before the transaction begins. Performance marketing gets repriced. For twenty years brands have paid Google and Meta to catch the eye of a wavering customer. That customer is no longer in the room.

The new SEO

An agent cannot surface what it cannot read. So make your product legible to a machine: structured data, verified certifications, machine-readable pricing, APIs that expose stock. A product without them is invisible to the buyer that will matter.

The ranking layer is not neutral. As products become more machine-readable, an industry will emerge to tune that legibility: to certify, structure, and present data in ways that sway ranking decisions. That industry will be to the agent layer what SEO became to Google search.

Then comes the harder question. For every line of revenue, what share is friction value and nothing else? That share is the exposure. The share that remains is the business.

What markets called brand, loyalty, and customer lifetime value was often something simpler: the customer’s unwillingness to look. The customer is about to delegate the looking. The margin that remains is the margin that was earned.

About the Author

Luke Heilbuth is CEO of sustainability strategy consultancy BWD Strategic, and a former Australian diplomat.

On Substack, Luke writes about the systems we’re breaking and the blindness that lets us — from climate and geopolitics to AI and the future of work. Read & Subscribe on Substack here.