Your new customer is a neural net. How not to sleep through AI agents

In June 2026 Harvard Business Review ran a headline that made marketers’ eyes twitch a little: “How Do You Market to an AI Customer?” (HBR). The idea is simple and uncomfortable: brands have a new kind of buyer - not a human, but a neural net that searches, compares and chooses on their behalf.

It sounds like a scene from the future. In fact the infrastructure is already being built, and the giants are building it. Let’s break it down without the panic: what’s real here, what’s hype, and above all what touches you if you run services or a small business rather than a billion-dollar marketplace.

What’s actually happening

In six months the big platforms have bet hard on agents that buy for people.

Amazon shipped the AWS Agentic Shopping Assistant - a tool that lets any retailer embed an AI shopping helper. Under the hood is the same technology as Alexa for Shopping, which drove nearly $12 billion in incremental sales in a year. Rollout is promised in about 60 days, and the retailer keeps its data and catalog (GeekWire).

At I/O 2026 Google showed the Universal Cart - a smart cart on Gemini models that gathers products from Search, YouTube and Gmail and watches prices and compatibility. And for agents that check out on their own, Google is expanding two protocols: the Universal Commerce Protocol for checkout and the Agent Payments Protocol for payment. The U.S. rollout is set for summer 2026 (Google).

Microsoft, OpenAI and Walmart are moving the same way. The projection everyone now cites: by 2030 more than 30% of online purchases will run through AI agents one way or another (GeekWire).

Where the hype ends and reality begins

Now the cold water. Forrester, in its mid-2026 “State of Agentic Commerce”, says it plainly: real autonomy is still rare. People happily hand the AI discovery and comparison, but they press the “pay” button themselves. The gap between the hype and how people actually behave is wide, and the analysts advise not to rush but to prepare - “strategy over urgency” (Forrester).

How raw it still is shows in the giants fighting. In one early test it went to court: Perplexity’s shopping bot was blocked on Amazon. The platforms are still arguing whether to let other people’s agents buy on their turf at all.

The calm conclusion: an agent that buys everything itself is not today, and not for everyone. But the step that has already happened matters more than the purchase itself - a machine now stands between you and the customer, filtering the options.

What this has to do with services and small business

The loud part is all retail: carts, catalogs, payment. If you run B2B services or a small local business, it’s easy to wave it off - “agents with carts aren’t my story.” That’s a mistake.

For services, the “AI customer” looks different. It’s not an agent that bought cleaning or marketing from you out of a cart. It’s a neural net answering a person’s question “find me a contractor for X” with a short list of three to five names. The person then chooses by hand - but only from the list the machine built.

And that’s the whole point. The customer used to google it themselves, open a dozen tabs and compare. Now the neural net compares, and the person gets a ready selection. Not in it - you’re simply not considered. Not “you rank lower”, but you’re not in the conversation at all.

I see this on my own projects . Whoever the AI sees clearly in the sources is who it pulls into answers: a cleaning project with history and systemic SEO gets cited by Copilot dozens of times; a fresh site with no footprint in the index - zero. The machine recommends not the best but the legible. The one it has clear data on.

What to do

Don’t buy “agentic” solutions on hype. For services and small business something else works now - four steps.

1. Become the one AI finds and recommends. This is GEO: factual pages, reviews, mentions, markup. The goal is to land in the short list when a person asks the neural net who to hire. Forrester calls it a “machine advantage”, and it’s built through content the AI can read and cite.

2. Give the machine structured facts. What you do, for whom, in what geo, your price range, your niche - in a form that’s easy to compare: Schema markup, a clear services page, an llms.txt. Agents and neural nets choose by comparable data. No data, no comparison, no you in the list.

3. Don’t confuse hype with the task. You don’t need an “AI agent with a cart” right now. You need to be recommended in answers. That’s cheaper, simpler, and works today - not by 2030.

4. Watch how AI describes you. A neural net can not only fail to recommend you but also lie - I covered that in the piece on Google’s liability for AI Overviews. Check regularly what the machine says about you, and fix the sources.

Bottom line

The “AI customer” isn’t a marketing scare story, and it isn’t a reason to rush out and buy the trendy word “agentic”. The giants are building carts and payment protocols, but for services and small business the shift is simpler and closer: a machine now stands between you and the person and decides who even gets shown.

So the task for now is down to earth. Not to learn how to sell to a robot, but to make the robot find you, understand you and recommend you. It’s the same hygiene as with regular AI search: clear facts, a clean footprint in the sources, regular checks.

Want to know whether the neural net sees you and whether you make its short lists - message me on Telegram @dimik90 or order an audit. I’ll look at what AI returns for your topic and where you are in that picture.

How it works on a live project is in the cleaning case with SEO from zero in 50 days. And how a neural net can lie about a business and who answers for it is in the breakdown of the AI Overviews ruling.