The agent is the buyer now. Your checkout isn't ready.

Yesterday I wrote that signup is dead, the agent is the signup form now. The argument was simple. Humans aren't filling out your onboarding flow anymore. Their agents are. If your site can't talk back to an agent, you don't get the user.

Today the question is bigger, because checkout is where the money lives. And there's a specific kind of executive who is about to lose it on purpose.

The executive treating product data like a trade secret

You know the type. Search behind a login wall. Pricing fetched by an authenticated GraphQL call inside a useEffect. Inventory hidden until you add to cart. WAF rules that block anything that isn't Chrome on a MacBook. The product catalog locked up like it's a list of nuclear launch codes.

They'll tell you it's about protecting the catalog. Competitive intelligence. Scraper defense. Brand control. The real reason is older and dumber than that. Most retail leaders still think their product data is the asset. It isn't. The transaction is the asset. The product data is the signal that attracts the transaction.

Hide the signal and the agents shop somewhere else. That's not a prediction. That's already happening.

The thing nobody is quite saying out loud

In September 2025, Stripe and OpenAI co-developed and released the Agentic Commerce Protocol. ChatGPT shipped Instant Checkout the same day, powered by ACP, letting US users buy from Etsy sellers and over a million Shopify merchants directly inside the chat. Perplexity has its own shopping product, Buy with Pro, settled through PayPal. Visa shipped Intelligent Commerce. Mastercard shipped Agent Pay. Stripe's Shared Payment Tokens now wrap both of those plus Affirm and Klarna. Anthropic, Microsoft Copilot, Google, and others are building toward the same model from different angles.

Read that paragraph again, slowly. The money rails for autonomous AI purchases are live. Not "coming in 2027." Live. Today. There is a credit card sitting inside an agent right now, and it works.

The most common take on this is some version of "yeah but our customers aren't buying through ChatGPT." That's not the question. The question is whether your competitor's catalog shows up when an agent asks "find me a CRM for a 12-person sales team under $80 per seat" and yours doesn't.

If your search and your checkout aren't open to agents, you are losing transactions you don't even know are happening. There's no analytics event for "an agent tried to buy from you and gave up because your prices were rendered in JavaScript." It's just a flat line where revenue should be.

What "agent-readable" actually means

This is where I get to be the frontend guy for a second, because the failure modes are deeply frontend failure modes.

Here is what agents need from your site to transact.

A catalog they can read without rendering your app. JSON-LD Product and Offer schema on every product page. A sitemap that lists products. Ideally an MCP server or a clean public read-only API that exposes search, variants, and stock. If your prices are populated by a useEffect that hits an authenticated GraphQL endpoint, the agent sees a blank. Congrats, you're invisible.

Pricing and inventory that are facts, not interactions. "Call for quote." "Sign in to see price." "Add to cart to see shipping." Every one of these is a wall. A human will tolerate one of them. An agent will route around all of them and pick the vendor that didn't make it weird.

A checkout an agent can actually drive. ACP gives you the structural path. Shared Payment Tokens give you the payment primitive. An agent presents a scoped token, your backend confirms the charge through your existing processor, the human approved the spend out of band in their wallet. The plumbing is done. You have to opt in.

A door that opens for legitimate agents. Half of you are reading this thinking "we already block bots at the edge." Yes. And you're going to block your future customers' agents the same way. Cloudflare is shipping agent passes. Signed agent tokens are a real thing. User-agent allowlists are a real thing. "Block everything that isn't a real browser" is a 2019 strategy and it's about to cost you.

The deep shift. Your product catalog is now an API surface. Stop treating it like a marketing page where the goal is to capture an email. The goal is to be the structured answer when an agent asks a question with a purchase intent attached.

The mobile-web moment, again

I'm old enough to remember the room full of executives in 2010 saying "our customers don't use phones to buy things." Every one of them lost a decade. A few of them lost the whole company.

This is the same shape. The executives saying "our customers don't use agents to buy things" in 2026 are going to say it right up until their organic traffic flatlines and they can't figure out why. The traffic didn't disappear. It got intermediated. An agent answered the question and recommended somebody else.

What's different this time is the timeline. Mobile took six or seven years to fully eat retail. Agent commerce is going to move faster because the infrastructure is being shipped together, on purpose, by the same handful of companies. Stripe, the model providers, the card networks, the protocols. The flywheel is intentional. You have maybe 12 to 18 months before "agent-readable" stops being an edge and starts being table stakes.

What to do this quarter

I've been building MCP integrations for the better part of a year, watching agents become the new browser. Here is what I'd do if I owned an e-commerce site right now. Three things. This quarter. Not a roadmap item, not a Q3 OKR.

1. Make your catalog machine-readable today. Add JSON-LD Product and Offer schema to every PDP. Server-render the price. Publish a product sitemap. If you have engineering bandwidth this quarter, stand up an MCP server or a public read-only catalog API. This is a week of work and it's the highest-leverage week your team will spend in 2026.

2. Open a lane for agent checkout. Look at the Agentic Commerce Protocol and the Shared Payment Token flow. You don't have to rip out your existing checkout. You just have to expose an alternate path that an agent can drive. A stable cart endpoint, a confirm-with-token endpoint, a webhook for fulfillment. Stripe wrote the hard part. You're writing the glue.

3. Stop blocking the wrong robots. Audit your WAF rules, your Cloudflare config, your rate limits. Distinguish "scraper trying to copy my catalog" from "agent trying to buy something on behalf of a paying human." The first one is the threat. The second one is the customer. If your current setup can't tell the difference, that's the bug.

That's it. Three things. None of them are speculative. All three are based on infrastructure that is already shipped, already documented, already running in production at companies you've heard of.

The honest close

Your catalog isn't an asset you protect. It's a signal you broadcast. The executives who still think product data is a competitive moat are about to learn the same lesson their 2010 predecessors learned about mobile, except faster and more expensively.

In 18 months you're going to be in a board meeting wondering where the organic traffic went. The answer is going to be that an agent answered the question, picked the vendor whose data was public, completed the checkout that was open, and you weren't on the list.

You're not protecting the catalog. You're killing it. Every locked endpoint, every JS-rendered price, every "we don't allow scraping" rule is a sign on your front door that says "agents not welcome here." The agents will respect it. They'll go next door.

Open it up this quarter, or watch the agents route around you for the rest of the decade. Yesterday the agent filled out your form. This week it's pulling out a credit card. If the only thing standing between that credit card and your revenue is a JS-rendered price tag and an overzealous bot blocker, fix it before the quarter ends.

The buyer changed. The checkout has to change with it.