AI agents are
your best customers.

Your security team spent ten years teaching your platform to turn bots away. Your fastest-growing source of new revenue is a bot. Both are true this quarter, and only one of them is funded.

The industry's answer to the second fact is a protocol war. OpenAI and Stripe put Instant Checkout inside ChatGPT in September 2025 and retired native checkout by March 2026. Google and Shopify shipped the Universal Commerce Protocol in January. In April, Google donated its Agent Payments Protocol to the FIDO Alliance. Three rival rails in nine months, and every vendor wants you to bet on one.

Do not take the bet. The rails will keep changing. The lane is the asset. If an agent cannot read your catalog, trust your data, and buy from it, no protocol will save you.

You can already feel the pull to pick a standard and defend the choice to the board. That is the trap. Every hour spent handicapping the rails is an hour not spent on the one asset that pays off no matter which rail wins. The protocol is a bet. The lane is an investment. Only one of them still exists in eighteen months.

We called it in 2016. It is on the record.

Everyone is discovering the lane this quarter. We wrote it down in 2016.

Rise of the Bots was a threat guide. It covered account takeover, scalping, scraping, and the four generations of bots draining margin from online retail. That is what makes one section of it impossible to argue with now. Inside a document about blocking the machines, the instruction was to build a lane for the ones that buy.

So read the rest of this as one thing. You are not weighing a vendor guessing at a trend. You are reading a firm that saw the shape of this a decade early and has delivered the systems underneath it ever since.

The bots you blocked grew up and now, they're your customers.

Then the data caught the thesis.

Shopping searches on generative AI platforms grew 4,700% between 2024 and 2025. ChatGPT accounted for 16% of Zara's inbound traffic. Those visitors do not only browse. AI shopping agents convert at 11.4%, against 6.1% for organic search. Over Black Friday 2025, Adobe recorded 805% year-over-year growth in traffic from AI sources to US retail sites. Salesforce reported $67 billion in AI-influenced global Cyber Week sales. Agentic traffic grew 1,300% in the first half of 2025 alone.

Read those numbers against ten years of rules written to turn automated traffic away. The bots retail spent a decade blocking grew up into the customers it now wants, and they convert better than the traffic you pay for.

Here is the part that should hold your attention. Right now, an agent is deciding whether to put your product or the brand beside you in front of a buyer. If your catalog is not readable, that decision is already made, and not in your favor. You will never see the traffic you lost, because it never arrives as traffic. It arrives as a competitor's order. Multiply that across a quarter of category searches and you are looking at a revenue line no dashboard you run today will show you.

A lane is a data decision, not a protocol integration.

This is where the protocol conversation misleads the room. A lane is not a switch you flip when a standard wins. You cannot bolt an agent onto a catalog the agent cannot read. The lane is process architecture, and if you own the data, it is your program to lead.

Four things decide whether an agent can use you.

Structured product data

Clean Product, Offer, and FAQ schema, with real-time price and inventory through an API. An agent that cannot parse your catalog will never recommend it.

Trust an agent can read

Reviews, provenance, authentication, and named sources. Agents rank on trust they can verify, not brand equity they cannot see.

Presence in the sources models cite

Models answer from third-party sources, not only your own site. Absent from what they cite, you are absent from the answer.

A feed an agent can buy from

One transactable product feed the agent completes the purchase against, whichever rail it arrives on.

Not one of those four is a protocol. All four are data.

One rule keeps the work from aging out. Decouple the AI and data layer from the commerce platform, so a single clean pipeline feeds site search, third-party agents, and conversational interfaces at once. Bind the AI layer to the platform and every replatform forces you to re-engineer it. Separate them and your catalog stays portable across whichever rail wins. It is the same call we made delivering Empower Global's platform, 13 integrated systems across 4 Salesforce clouds in one calendar year. The lesson that program taught, and every one since, is that the data layer has to outlive the platform decision.

The order is the strategy.

Here is the whole conviction in one line. AI is sequenced, not deployed.

Baseline your AI share of voice first. Run your real category buying questions through the assistants your customers already use, and measure where you appear. Then fix the feed and the schema, because discovery without a readable catalog converts nothing. Then seed the third-party sources models cite. Then instrument it and hold it, because share of voice moves week to week. Ninety days, in that order.

This is not a year-long transformation you have to defend to a steering committee. The schema, the feed, and a real presence in AI answers are a this-quarter move. The brands taking agent traffic today started ninety days ago. Reverse the order, integrate a protocol before you have a catalog worth integrating, and you get the pilot that demos well and never converts to production. None of it waits on a standards body. Every step is inside your control right now, which is exactly what makes the delay so expensive.

The protocol war will resolve itself. Your readiness will not. The gap between the brands agents recommend and the brands agents skip is opening now, quietly, before most boards have named it. If you want the method behind it, the AI and Agentic Commerce practice is where we lay out why AI is sequenced, not deployed.