Can Your Product Take an Order From a Robot? The Agent-Payable Gap in B2B

Every purchase your company has ever made was made by a person. Someone read the page, weighed the options, clicked the button, and signed the form. The whole apparatus of B2B selling, the demos and the decks and the "let's grab 30 minutes," was built on that single assumption: at the moment money moves, a human is in the room.

That assumption quietly stopped being safe sometime in the last year.

For revenue leaders, RevOps, product and pricing owners, and anyone who runs the path between "interested" and "paid": the software that decides what to buy and completes the purchase is increasingly not a person, and most B2B companies have built no way for it to transact with them. Your competitor's product might be one an agent can buy in ninety seconds. Yours might be one it gives up on.

The rails got built while everyone was arguing about chatbots

While the industry spent 2025 debating whether AI agents were real, the payment networks went ahead and built the plumbing for agents to spend money.

OpenAI and Stripe published the Agentic Commerce Protocol in the fall of 2025, an open standard for how an AI agent completes a purchase with a business on a buyer's behalf. Shopify wired it in at the platform level, which switched on more than a million merchants at once. Mastercard shipped Agent Pay. Visa followed with its Trusted Agent Protocol and expanded its Intelligent Commerce pilots into 2026. Google put out the Agent Payments Protocol. In March 2026, Stripe and Tempo launched the Machine Payments Protocol with a partner list that read like a summit invite: Visa, Mastercard, Anthropic, OpenAI, and Shopify among them.

None of these are demos. They handle tokenized credentials scoped to a specific agent, a specific merchant, and a specific spending policy, so a model can complete a checkout without ever touching a raw card number. Adyen rolled out a modular suite, Agentic Feed, Agentic Cart, and Agentic Payments, so enterprises could sell through conversational platforms without rebuilding their commerce stack for each one.

Read that list again and notice who is on it. These are not startups chasing a trend. This is the entire settlement layer of global commerce agreeing on how machines pay for things.

B2B told itself it was exempt. It was half right.

The reflex in most B2B companies is to wave this off as a consumer story. Somebody's AI assistant buying sneakers or booking a flight. Enterprise software, the thinking goes, involves buying committees, security reviews, and six-figure contracts that no agent is going to autonomously approve.

That instinct is correct about the top of the market and wrong about the rest of it.

Gartner expects up to 40 percent of enterprise applications to include task-specific AI agents by the end of 2026, up from less than 5 percent a year earlier. It has also put roughly 234 billion dollars of enterprise application spending at risk of what it calls agentic arbitrage between now and 2030, meaning spend that could shift as agents, rather than people, decide where it goes. Its broader estimate of embedded agentic AI spending for 2026 sits near 202 billion dollars. Ramp is already applying Visa's Intelligent Commerce to corporate bill pay. The direction is not subtle.

The mistake is picturing an agent signing your master service agreement. That is not where this starts. It starts at the boring end of the buying spectrum, and it works its way up.

Three places it lands in B2B first

Self-serve and product-led renewals. If a customer can buy, expand, or renew your product without talking to a rep, they can also point an agent at that flow. An operations manager who used to add seats through your billing page can now hand that errand to software. The purchase still happens. Nobody from your side is in the conversation, and nobody from theirs is either. Whether your expansion revenue survives that shift depends entirely on whether your self-serve flow is legible to a machine.

Tail and indirect spend. Every company has a long list of smaller recurring purchases that no human enjoys managing: the monitoring tool, the data enrichment credits, the third seat on some utility nobody remembers signing up for. This is exactly the spend procurement teams want to automate, and exactly the spend where an agent completing the transaction creates no controversy. It is also spend that used to renew on autopilot precisely because switching was annoying. Take the annoyance away and the switching gets easier for everyone, including your customers leaving you.

Marketplaces and resale. The cloud marketplaces and reseller channels were already becoming a primary purchase path for B2B software. Layer agent-driven buying on top of a structured catalog and you get a channel where being present, priced, and machine-readable matters more than being persuasive. An agent comparing three tools in a marketplace does not read your brand story. It reads your feed.

What "agent-payable" actually requires

Here is the uncomfortable part. Being ready for this is not a marketing project. It is closer to a plumbing project, and it has four pieces.

The first is a catalog a machine can read. Under the emerging standards, merchants publish structured product feeds, in formats like JSON or CSV, carrying names, descriptions, SKUs, prices, availability, and terms, refreshed frequently so pricing stays current. If your pricing lives in a slide a rep emails after the second call, an agent cannot see it, cannot compare it, and cannot buy it. "Contact sales" is not a price. To an agent, it is a closed door.

The second is an actual transaction endpoint. Somebody on the buying side sending an agent to purchase needs a machine-to-machine way to place the order, confirm it, and pay. The protocols spell this out as a small set of checkout endpoints plus a tokenized payment step. Most B2B companies have a lead form where a checkout should be. A lead form asks a machine to wait for a human to call it back, which is the one thing an autonomous agent will not do.

The third is identity and authorization you can trust. When an agent shows up to buy, you need to know which agent, acting for which company, under what spending limit, with what approval behind it. This is the piece the payment networks are actually solving with scoped tokens, and it is the piece that lets you say yes to a machine without opening yourself to fraud. It is not optional. Deepfake and impersonation fraud has been climbing fast, with reported losses now in the billions, so an unverified agent asking to transact is a threat, not a lead.

The fourth is terms an agent can accept. Your contract, your usage limits, your acceptable-use policy: if a human has to read and countersign all of it, the purchase stalls. The purchases that will flow through agents first are the ones where the terms are simple, standard, and machine-parseable. Everything you have wrapped in negotiation you have also wrapped in friction, and friction is now a reason an agent routes around you.

The metrics you rely on start lying

This shift breaks measurement in a way that will feel familiar to anyone who watched agent traffic wreck their website analytics.

Attribution goes first. When a person buys, you can usually reconstruct what influenced them: the webinar, the email, the peer recommendation. When an agent buys, that trail thins out or vanishes. The agent read your feed, compared it to two others, and picked one on criteria you never see. Your marketing team keeps optimizing content for a reader who was not there.

The "contact sales" wall stops being a qualification tool and becomes a leak. For years, hiding pricing behind a form was a defensible way to force a conversation and capture a lead. Against an agent-driven buyer, it is a filter that removes you from the comparison set before anyone human ever weighs in. The buyer's agent does not fill out your form to unlock a number. It notes that the number is missing and moves to a vendor whose number it can read.

And your win-loss data develops a blind spot. You will lose deals you never knew were being run, to comparisons you never saw, on a page you never got to argue on. The loss looks like nothing. It looks like a renewal that quietly did not expand, or an evaluation that never became an inbound.

What to do before your buyers' agents arrive

None of this argues for throwing out human selling. The large, complex, high-trust deals will stay human for a long time, and that is genuinely where sellers earn their keep. The work is narrower and more concrete than a strategy offsite.

Start by finding your agent-buyable surface. Walk your own purchase paths and mark the ones a machine could complete today without a human on your side: self-serve signup, in-product expansion, marketplace listings, simple renewals. That list is your exposure and your opportunity, and it is almost always bigger than teams expect once they actually look.

Publish a price a machine can read on those paths, even a starting price or a clear range. You do not have to expose your enterprise negotiation. You do have to stop being invisible to a buyer who now shops with software that treats a missing price as a missing product.

Treat your product feed as a channel, not a chore. The catalog data that agents read, the SKUs and descriptions and terms, is becoming as important to distribution as your homepage was in 2010. Someone should own its accuracy the way someone owns your website.

Get ahead of agent identity with whoever runs your billing and security. Decide now how you will verify an agent, honor a spending limit, and log an autonomous purchase, before the first one shows up and forces an improvised answer.

Then instrument the machine buyers you already have. You almost certainly have some agent-initiated activity in your signup and billing flows right now that you are counting as human. Separating the two is the difference between managing this shift and being surprised by it at a renewal.

The buyer changed shape, not just channel

The last two decades of B2B taught everyone to obsess over the buyer's experience: the content they read, the demo they sat through, the trust you earned across a dozen touches. That work still matters for the deals that stay human.

But a growing share of purchasing is moving to a buyer that does not sit through anything. It does not remember your ad or warm to your founder's LinkedIn post. It reads a feed, checks a price, verifies it can pay, and completes the order, or it does not, and you never hear why. The companies that notice this early get a structured, machine-readable, agent-payable front door built while it still counts as foresight. The ones that wait will find out the way everyone finds out these things now. From a chart, a quarter too late.