AI Agent Distribution: The New Front Door

Meta told investors that its next act runs through a consumer AI agent that has not shipped yet. On its most recent earnings call the company put personal agents at the center of the plan and said billions of people will soon lean on one. Revenue for the quarter came in around 60 billion dollars, and Meta is spending north of 30 billion dollars every three months on the machines meant to make that agent real. Read the transcript closely and the bet is not the model. Zuckerberg was not promising the smartest model. He was promising the agent that billions of people route their apps through. Whoever owns that agent owns the front door.

That is the part most founders are sleeping through. We spent 20 years learning to reach humans. We got good at search, at ads, at storefronts, at building an audience. All of it assumed the same thing: a person on the other end who searches, clicks, compares, and chooses. That person is starting to hand the choosing to a machine. When they do, every distribution muscle you built points at a buyer who left the room.

This is not a post about optimizing for AI search. It is a post about what happens to your business when the customer stops being the one who picks. It sits inside my larger AI-native founder playbook and connects to the broader AI opportunity map, because distribution is where most of that opportunity gets won or lost. I have built two companies where AI does most of the production work, and the scary shift is not on the supply side. It is that the demand side is quietly being intermediated by software you do not own. Here is the durable version of the story, the map I use, and the one move that survives it.

Table of Contents

• The front door is moving and you are optimizing for a reader who stopped reading
• The framework: the three distribution eras
• The agent shelf
• The intermediation tax
• The selection surface versus the persuasion surface
• Shelf versus infrastructure
• The direct-relationship reserve
• The disintermediation map
• The contrarian take: stop trying to win the shelf
• What to do Monday morning: the front-door audit
• FAQ

The front door is moving and you are optimizing for a reader who stopped reading

Every distribution playbook you know assumes a human at the end of the funnel. You write copy to persuade a person. You rank in search because a person types a query and scans results. You run ads because a person sees them and feels something. You build in public because a person follows along and remembers you when the need shows up. The entire craft rests on one buyer: a human who looks, judges, and decides.

That buyer is being replaced by a proxy. Not everywhere, not all at once, but fast where it counts. People are starting to tell an agent what they want and letting it do the looking. The signals are already loud. Gartner projected that traditional search volume would fall about 25 percent by 2026 as answer engines take over queries that used to run through a search box. By 2026, roughly 65 percent of Google searches ended without a single click to any website, and for searches that triggered an AI answer summary, more than 80 percent ended in zero clicks. The person got the answer and never visited the source. When there is no click, your beautifully ranked page persuaded no one, because no one read it.

Commerce shows the same pattern harder. By 2026, AI-referred retail traffic was growing several hundred percent year over year and converting meaningfully better than traditional search, and a single assistant was fielding tens of millions of shopping queries a day. OpenAI shipped a commerce protocol that lets people buy inside the chat window, with Shopify, Etsy, and Walmart wired in at launch. The analysts who size these things put agentic commerce at roughly 7 to 8 billion dollars in 2026 and project it into the tens of billions within a decade. The direction is not subtle. The person who used to arrive at your storefront is increasingly sending a delegate.

Here is the uncomfortable math. If a growing share of your buyers hand the decision to an agent, then a growing share of your distribution spend is aimed at someone who is no longer in the room. You are still optimizing the persuasion of a human who has outsourced persuasion to a machine that does not feel your brand, does not read your landing page, and does not care about your story. The front door moved. Most founders have not noticed because their dashboards still show humans, and the agent traffic is hiding inside referral buckets nobody has bothered to split out.

The framework: the three distribution eras

To see where this goes, it helps to name where it came from. Distribution has moved through three eras, and each one put a new gatekeeper between you and the person who pays. The pattern is not random. In every shift, the buyer you optimize for gets one step further from the human whose money you want.

Look at what moved. In Era 1 the front door was the person’s own attention, and you competed for it directly. You built a storefront, you ranked in search, you earned word of mouth. The tax was your own sweat. The buyer and the payer were the same set of eyes.

In Era 2 a platform slid in between. The App Store, the ad network, the social feed. Now a person still made the final call, but a ranking decided whether they ever saw you, and the platform took a cut of the transaction for standing in the middle. Apple has taken roughly 30 percent of App Store sales since 2008, and every founder who built on that surface learned that the aggregator sets the terms. You optimized for the algorithm, not the human, because the algorithm decided which humans reached you.

Era 3 finishes the move. The human delegates the decision itself. They tell an agent what they want, the agent queries a shelf of machine-readable options, and it returns one. The person may never see the alternatives. They may never see you even when you win, because the agent hands them a result, not a list. You are no longer optimizing for a human, or even for a ranking a human scrolls. You are optimizing to be the thing a machine reaches for on behalf of a person who trusts it to reach well. The front door is now inside a piece of software you do not own.

The through-line is the law that runs the rest of this piece. When an agent stands between you and your user, distribution stops being who you reach. It becomes whether the agent reaches for you. Everything else is a consequence of that sentence.

The agent shelf

Start with the mechanics, because the mechanics decide the strategy. When a person asks an agent to book, buy, compare, or handle something, the agent does not wander the open web the way a human browses. It calls a set of tools and data sources it already knows how to reach, ranks the options by signals it can read, and picks. That set of reachable options is a shelf. Your product is either on it or it is not, and if it is on it, your position is set by the agent, not by you.

This is aggregation in its purest form. The agent owns the demand, because it owns the moment of choice. You supply one option among many, and you have almost no control over how you are displayed, ordered, or described. The economics are the same ones the App Store taught a generation of founders. The party that stands between supply and demand skims margin, controls visibility, and monetizes participation, and the suppliers who depend on it cannot influence it. The only thing that changed is the shape of the shelf. It used to be a grid of icons a human scrolled. Now it is a list of callable tools an agent queries, and the human never scrolls it at all.

Being on the shelf is not automatic. To be reachable by an agent, your product has to exist in a form an agent can find and call. That is why the number of tools agents can reach has exploded. The registry of callable tool servers that agents use went from roughly a hundred entries in late 2024 to nearly ten thousand by the middle of 2026 and into the tens of thousands by the end of it. Those are not marketing pages. Each one is a product an agent can discover, authenticate against, and use without a human ever visiting a website. If your product is not on that layer, it is not slow to find. It is invisible. A whole class of founders is about to learn what app developers learned in 2008: the shelf is where the demand is, and the shelf has an owner who is not you. The same swarm of agents that creates this demand surface is the one I wrote about in AI agent sprawl, only here the sprawl is on the buyer’s side of the table.

The trap inside the shelf is that it feels like distribution. You get listed, you get calls, revenue shows up, and it looks like you cracked a channel. What you actually rented is a position that the shelf owner can reprice, reorder, or revoke the moment it decides to monetize placement or launch a competing option of its own. A thin option on the shelf is the demand-side twin of the AI wrapper trap: easy to list, easy to replace. That is not a hypothetical. It is the standard arc of every aggregator that ever existed. First it needs supply, so the terms are generous. Then it has demand, so the terms tighten. I write more about that dependency dynamic in my piece on AI platform risk for founders, and the same logic now applies to the demand side, not just the supply side.

The intermediation tax

Every layer between you and the person who pays charges rent. Call it the intermediation tax. It is not always a fee. Sometimes it is a cut, sometimes it is lost visibility, sometimes it is a customer relationship you never get to form. But it is always there, and it always grows, because the intermediary’s power over you increases as you depend on it more.

In Era 2 the tax was legible. Apple and Google name a number, and you can do the arithmetic on a 30 percent cut. What made it brutal was not the headline rate. It was that the tax rose with your success and you had no seat at the table when it changed. The App Store dropped to 15 percent for small developers and for subscriptions past a year, but only after years of pressure and lawsuits, and always on the aggregator’s timeline, not yours.

In Era 3 the tax is worse because it is invisible. When an agent picks a competitor over you, there is no line item. You simply do not get the sale, and you never learn you were in the running. Zero-click search already showed the shape of this. When 65 percent of searches end without a click, the tax is not a percentage of your revenue. It is the revenue that silently never arrives because the answer engine satisfied the person before they reached you. An agent that completes a purchase inside its own window does the same thing at the level of the whole transaction. You might be the fulfilling merchant and still never own the customer, never see their email, never get the chance to sell them the second thing.

The reason this matters for strategy is simple. A visible tax you can price into your model. An invisible one you cannot, because you cannot manage a number you never see. The first job in the agent era is to make the tax visible again by measuring how much of your demand already runs through intermediaries you do not own. Almost no founder has that number. It is usually the single most important number they are not tracking.

The selection surface versus the persuasion surface

Humans and agents buy on completely different signals, and most products are built for only one of them. A human buys on persuasion. Story, design, the feeling your brand gives, the social proof of other people, the way the page makes them trust you. An agent buys on selection signals. Can it call you programmatically, is your data structured and verifiable, is your price legible, are you reliable, do you have the permissions it needs, do the reviews parse as data rather than vibes. The human reads a page. The agent reads a schema. This is the flip side of a point I made in AI agents are your customer: there the agent is the buyer you sell to, here it is the buyer that stands between you and the human, deciding on their behalf.

I think of these as two different surfaces your product presents to the world. The persuasion surface is everything a human perceives. The selection surface is everything a machine can parse and act on. For 20 years the persuasion surface was the whole game, so that is where all the craft went. The selection surface barely existed, because there was no machine buyer to serve. Now there is, and the weighting between the two flips as a larger share of your demand arrives through agents.

The mistake in both directions is real. A product built entirely for persuasion, all brand and no callable interface, is invisible to an agent no matter how good it is. The machine cannot feel your brand. A product built entirely for selection, all API and no story, is forgettable to the humans who still choose directly and to the humans who will decide which agents to trust. You need both. What changes is the mix, and the mix should track one number: the share of your demand that now arrives through a machine. Most founders have that dial jammed all the way left because they have never had a reason to move it. The reason arrived.

Shelf versus infrastructure

There are two ways to be present in the agent era, and they have opposite risk profiles. You can be an option the agent ranks, or you can be a tool the agent calls. I call these the shelf position and the infrastructure position, and the difference decides how much power the intermediary has over you.

On the shelf, you are one of many interchangeable choices. The agent compares you against alternatives and picks the one that scores best on its signals. Your position is entirely at the agent’s discretion, and the moment a cheaper or better-integrated option appears, you drop. This is the high-risk spot. It looks like distribution and behaves like a rented ad slot. As infrastructure, you are the thing the agent depends on to do the job at all. The agent does not compare you against a list. It calls you because you are wired into the workflow, hold the data, or complete the action nobody else can. That dependency is your defense. Ripping you out is expensive, so the agent does not casually swap you.

Most products get pushed onto the shelf by default, because being one option among many is the easy thing to build. Becoming infrastructure takes a deliberate choice about what you own that the agent cannot get elsewhere. This is the same question I ask about model dependency in my piece on AI vendor lock-in and switching costs, only pointed the other way. There, the danger is your dependency on a vendor. Here, your safety is the agent’s dependency on you. The strategy is to convert yourself from a choice into a component. A choice gets reranked. A component gets kept.

The deeper reason infrastructure wins is that models keep getting cheaper and more capable, so anything an agent can do on its own or buy interchangeably drifts toward free. I wrote a whole piece on that absorption dynamic in the AI commoditization clock. Distribution safety in the agent era is the flip side of that clock. The parts of you that are interchangeable get absorbed into the shelf and repriced to zero. The parts that are load-bearing for the agent’s job survive, because the agent cannot afford to lose them.

The direct-relationship reserve

The single most valuable asset in the agent era is demand that reaches you without passing through an intermediary you do not own. I call it the direct-relationship reserve. It is the fraction of your customers who come to you directly, know your name, log into your product, sit on your email list, or belong to a community you run. That reserve is the one thing an agent cannot reprice, reorder, or disintermediate, because it never touches the agent at all.

Think about why the reserve is worth so much. Everything that flows through an agent is rented. The agent can drop you, tax you, or replace you, and you have no recourse because you do not own the relationship. Everything in your reserve is owned. Those customers chose you, remember you, and can be reached again on your terms. In a world where intermediaries are multiplying, the owned relationship is the only compounding asset. It is the difference between a business that rents its demand and one that holds it.

This is where the old distribution wisdom comes back around, upgraded. Building an audience you own was always good advice, and I made that case in my piece on distribution before product. In the human era it was an edge. In the agent era it is a survival trait. An audience that follows you directly is a channel no agent stands in the middle of. A brand strong enough that people ask for you by name forces the agent to reach for you specifically rather than for the category. That named pull is the distribution expression of the taste moat: the human edge a model cannot copy and an agent cannot average away. Brand pull, in the agent era, is not soft. It is the thing that converts you from a shelf item into a named request.

The reserve also changes how you should read your own growth. A founder celebrating a spike in agent-driven sales without a growing reserve is celebrating rented demand. It can vanish on the next reranking. A founder with slower growth but a thick, owned reserve is building something that survives the intermediary. When I audit a business now, I weigh the reserve more heavily than the top-line number, because the reserve tells me what the founder actually owns. This is the same instinct behind my piece on the one-person company and the incompressible core: figure out what cannot be handed off or taken away, and build there.

The disintermediation map

Two variables decide how exposed you are. One is the share of your demand that flows through agents you do not own. The other is the strength of your direct-relationship reserve. Plot them against each other and you get a map that tells you exactly where you stand and which way to move.

Safe is where you have a strong reserve and agents are not yet a big part of your demand. Most durable brands live here today, and the temptation is to assume it lasts. It does not, because the horizontal axis only moves right over time. Complacent is the quiet danger. Your agent share is still low, so nothing hurts yet, but your reserve is thin, which means you are one shift in buyer behavior away from trouble and you will not see it coming. This is where most software companies actually sit, and they mistake the calm for safety.

Exposed is the pain quadrant. A large share of your demand runs through agents, and you have no owned relationship to fall back on. The agent owns your customer, and you are one reranking away from a revenue cliff. Companies that built their whole business on a single platform’s algorithm have lived this before, and it is worse with agents because the drop is invisible until it happens. Hedged is the goal. You participate fully in agent distribution, you get the volume, but you have deliberately built an owned line that the agent cannot sit between. The arrow on the map is the only strategy that matters: move up before you are forced right. Build the reserve while your agent share is still low, because building it after you are already exposed is building a lifeboat in a storm.

The contrarian take: stop trying to win the shelf

Here is what most people are getting wrong right now. The entire market is racing to optimize for AI agents. Answer engine optimization, being the cited source, structuring your content so the model quotes you, getting listed on every agent surface. All of it is aimed at winning a better position on someone else’s shelf. And all of it is Era 2 thinking wearing Era 3 clothes.

Winning the shelf is a rented victory. The moment you become the top-ranked option in an agent you do not own, you have made yourself dependent on a placement the agent can reprice or hand to a competitor the instant it decides to. This is exactly the lesson the App Store taught, and the founders rushing to game agent rankings are volunteering for the same trap a second time, faster. Being the best option on a shelf you do not control is not a moat. It is a lease with no renewal clause.

The non-obvious move is not to win the shelf. It is to reduce the share of your demand that has to pass through a shelf at all. That splits into two plays. First, build the direct-relationship reserve so that a meaningful chunk of your demand never touches an intermediary. Second, convert yourself from a shelf item the agent ranks into infrastructure the agent depends on, so that even your agent-mediated demand is held by dependency rather than by ranking. Both plays reduce the intermediary’s power over you. Optimizing your shelf position increases it, because it deepens your dependence on the shelf while calling it a win.

Now the honest counterweight, because I do not want to sell you a clean story. The bet that agents will disrupt the incumbents may be wrong. Gartner predicted search would fall 25 percent, and Google adapted, shipped its own AI answers, and held more than 90 percent of the market. The likely outcome is not that agents free you from aggregators. It is that the same giants own the agents too, and the chokepoint gets more concentrated, not less. That does not weaken the argument. It sharpens it. If the front door is going to be owned by a handful of enormous players no matter what, then the only defensible position is the demand they cannot route through their door: the customers who come to you directly and the workflows that call you because they have to. You are not going to out-aggregate the aggregator. You are going to own a line it cannot cut.

What to do Monday morning: the front-door audit

This is not theory you file away. It is five moves you can start this week. I call it the front-door audit, and it turns an invisible shift into numbers you can manage.

1. Measure your agent-mediated share. Pull your acquisition data and split out everything that arrived through an AI surface: an assistant, an answer engine, an agent, an AI referral. Most analytics setups bury this in referral or direct buckets. Dig it out and put a percentage on it. That number is the most important one you are not tracking, because it tells you how much of your demand already runs through a door you do not own.

2. Score your selection surface. Ask the blunt questions. Can an agent call your product programmatically, or does it require a human to click through a website? Is your core data structured and verifiable, or is it locked in marketing prose? Is your price a clear number a machine can compare? If an agent cannot reach you, order you, and trust you, you are not on the shelf at all, and being invisible is worse than being ranked low.

3. Build one direct-relationship reserve line. Pick a single owned channel and grow it deliberately this quarter. An email list, an account login that brings people back, a community you run, a brand people ask for by name. The test is simple: can you reach these customers again without an intermediary’s permission? If yes, it counts toward your reserve. If no, it is rented.

4. Decide shelf or infrastructure. Look at how an agent would use you and ask whether you are a choice it ranks or a component it depends on. If you are a choice, find the thing you could own, the data, the action, the integration, that would make removing you expensive. Moving even one step from shelf toward infrastructure lowers your disintermediation risk more than any ranking optimization will.

5. Set a tripwire. Decide the agent-mediated share number at which you stop pouring effort into someone else’s shelf and start pouring it into your reserve. Maybe it is 20 percent, maybe 40. The point is to choose it now, while you are calm, so that when the number crosses it you act on a plan instead of a panic. Founders who set the tripwire early move up the map. Founders who wait get moved right by the market and discover the reserve they needed took a year they did not have.

The businesses that survive the agent era will not be the ones with the best shelf position. They will be the ones that measured the front-door shift early, built an owned line before they needed it, and made themselves the thing the agent cannot do without. Distribution stopped being who you reach. It is now whether the agent reaches for you, and whether you kept a door of your own.

Frequently asked questions

What is AI agent distribution?

AI agent distribution is how products get discovered, chosen, and used when an AI agent, rather than a human, makes the decision. Instead of a person searching, comparing, and clicking, the person delegates to an agent that queries a set of machine-readable options and picks one. Winning shifts from persuading a human to being the option an agent reaches for, which means being callable, structured, reliable, and either the best-ranked choice or the tool the agent depends on.

How is this different from SEO or answer engine optimization?

SEO and answer engine optimization both aim to win a better position on a surface you do not own, a search results page or an agent’s shelf. That is useful but rented, because the owner can reprice or reorder you at will. AI agent distribution is broader. It includes optimizing your selection surface, but the durable half of it is reducing the share of your demand that must pass through any intermediary, by building an owned audience and by becoming infrastructure the agent cannot easily swap.

What is the selection surface?

The selection surface is everything about your product that a machine can parse and act on: a callable interface, structured and verifiable data, a legible price, proof of reliability, and the permissions an agent needs. It is the counterpart to the persuasion surface, which is everything a human perceives, like story, design, and brand feeling. Human buyers read the persuasion surface. Agent buyers read the selection surface. As more of your demand arrives through agents, your investment should shift toward the selection surface without abandoning the persuasion one.

What is a direct-relationship reserve and why does it matter?

A direct-relationship reserve is the share of your demand that reaches you without passing through an intermediary you do not own: customers on your email list, people who log into your product, a community you run, or a brand strong enough that people ask for you by name. It matters because everything routed through an agent is rented and can be repriced or disintermediated, while everything in the reserve is owned and compounds. In the agent era the owned relationship is the only asset an intermediary cannot take from you.

Should I still invest in brand and content if agents are buying?

Yes, more than ever, but for a sharper reason. Brand pull is what forces an agent to reach for you by name instead of for the category, which turns you from an interchangeable shelf item into a specific request. Content and brand also build the direct-relationship reserve, the owned audience that never touches an agent. The mistake is treating brand as the whole game. The move is to pair a strong persuasion surface for humans with a strong selection surface for machines.

What does it mean to be infrastructure instead of a shelf item?

A shelf item is an option an agent ranks against interchangeable alternatives, so you drop the moment a better one appears. Infrastructure is a tool the agent calls because the job requires it, because you hold the data, complete the action, or are wired into the workflow. Infrastructure has low disintermediation risk because removing you breaks the task, while a shelf item has high risk because swapping you costs the agent nothing. The strategic move is to convert yourself from a choice into a component.

How do I measure how exposed my business is?

Measure two numbers. First, the share of your demand that already arrives through agents, assistants, or answer engines you do not own. Second, the strength of your direct-relationship reserve, the demand you can reach again without an intermediary. Plot them on the disintermediation map. High agent share with a weak reserve is the exposed quadrant, where the agent owns your customer. The goal is the hedged quadrant, where you participate in agent distribution but keep an owned line, and the move is to build the reserve before your agent share climbs.

Is agent distribution only relevant for consumer products?

No. It applies anywhere a buyer can delegate a decision to software, which now includes business software, developer tools, and services. In business settings the agent is often an internal assistant or a workflow that calls tools directly, and the same logic holds: be callable, be reliable, and be the component the workflow depends on rather than one option among many. If anything the infrastructure position is easier to reach in business software, because workflows reward the tool that is wired in and hard to remove.