Your Next Customer Might Be an AI Agent. Can It Buy From You?

There is a door on my website that no human is meant to walk through.

An AI agent arrives at a plain web address and asks one of my questions. My site replies with a price instead of an answer: ten US cents, payable in a digital dollar called USDC. The agent signs the payment, asks again, and gets what it came for. No cart. No login. Nobody clicks anything. The first time I watched that whole handshake run end to end in testing, it was over before I had finished reading the log.

I built it because I was curious what selling to a machine felt like from the seller’s side. What stayed with me was how ordinary it was. A transaction happened, and there was nobody in it.

It left me with a question I now put to owners. If a piece of software turned up at your business tomorrow carrying a budget and a shortlist, could it actually buy from you?

Faster Than Your Website Upgrades

Gartner made this its headline prediction for 2026. By 2028, it expects 90 per cent of business-to-business buying to be handled through AI agents, with more than US$15 trillion of spend flowing through agent exchanges. Procurement, restocking, supplier comparisons, quotes. The unglamorous buying that keeps every business running.

The consumer side is moving too, and Australians are not hanging back. Accenture’s Consumer Pulse research this year found 62 per cent of Australian consumers would rely on AI for at least half of their purchasing decisions. Separate research by Logica for Commerce and PayPal found 63 per cent of Australian online shoppers are interested in trying an AI shopping tool, and one in five already use AI when they shop online.

I would take any 2028 forecast with a pinch of salt. Analysts have called the timing of plenty of technology shifts too early. The direction is harder to argue with. More and more of your buyers are going to send something to do the reading before a person turns up.

Back in July I wrote about whether AI names your business when someone asks it for a recommendation. That is the first half of this. The second half is what happens after it finds you.

What An Agent Does When It Gets To You

A buying agent has three jobs. Find the options, evaluate them, and transact with the best one. Most of the conversation so far has been about the first job. I think a lot of small businesses are going to fall out at the second.

Picture an agent asked to compare three local builders for a commercial fit-out, or three bookkeepers for a trades business. It goes looking for what a careful buyer would want. Price, or at least a range. What is included. Lead times. Service area. Minimums. Payment terms. How to book.

Then it reads what is actually there. On plenty of SME websites, that is a paragraph about passion, a stock photo of a handshake, and a button that says “Call for a quote”. A human might ring. An agent treats it as a locked door and moves on to the supplier whose information it could read.

There is nothing personal in it. The agent is simply literal. It will not work out that your Facebook page has newer prices than your website, or that the PDF price list from 2023 buried in an old newsletter is out of date. It takes what you published at face value, and if you published nothing, you get nothing.

The Loss You Will Never See

The same Logica research found only 4 per cent of Australian shoppers want AI help at the checkout itself. People want the research and the comparison done for them, and they want to make the final click.

That sounds reassuring. A human still decides. But look at where the decision sits. The agent builds the shortlist, and the human chooses from it.

If you are not on that shortlist, nothing happens. No enquiry comes in. There is no abandoned cart, no bounce in your analytics, no polite email saying they went with someone else. The deal is lost before anyone knew there was a deal, and it leaves no trace in any report you will ever run.

That is the part that worries me. Businesses will get quieter over the next few years and put it down to a soft market, when what actually happened is they became unreadable to the thing doing the first cut.

Keeping a human in the loop at the final decision is a good thing, and it works in your favour, because people still choose on trust and relationship. They just only get to choose between the names they were shown.

A Data Problem Wearing A Sales Hat

Being readable to an agent mostly comes down to the same unglamorous work I wrote about in getting your business data AI-ready. One current version of the truth about what you sell, what it costs and how to buy it. Most businesses have three versions, all slightly different, and a fourth that lives in the owner’s head.

The tidy-up you have been putting off for years used to be an internal problem. It is now customer-facing.

What Agent-Ready Looks Like For A Normal Business

You do not need to rebuild your website or learn a payments protocol. For most SMEs the work is closer to housekeeping.

• Publish prices, or honest ranges. “Packages from $2,500” gives an agent something to compare. “Contact us” gives it a reason to skip you. If your work is genuinely bespoke, say what drives the price and give a typical range.
• One page per service, with the facts on it. What is included, who it is for, turnaround, service area, what the client needs to provide. Written plainly, kept current, in one place.
• Let people book without a phone call. A calendar link, a short form, an online quote request. Anything an agent can complete, or hand back to its person with a clear next step.
• Add structured data to your site. Schema markup labels your pages for machines: this is the business, these are the services, these are the prices, these are the common questions. A decent web developer can set it up in an afternoon.
• Test it the way a buyer would. Ask an AI tool to find and compare three businesses like yours in your area and report back on price, inclusions and how to book. Read what it says about you, and notice what it could not find.
• Leave the fancy stuff until later. Most SMEs do not need a machine-payable API this year. I built one for my own site because I wanted to understand the plumbing first-hand. For most businesses, the housekeeping above will do far more in the next twelve months.

If you want a second pair of eyes on how your business reads to a machine, it is the kind of thing I work through in AI consulting.

You Are Also The Buyer

Everything above is about selling. The same shift is going to land in your own purchasing, and that is where I would start experimenting.

Think about the buying your business does every month. Restocking consumables. Comparing freight quotes. Renewing software you forgot you were paying for. Chasing three quotes for anything over a certain amount. That is exactly the kind of repetitive comparison agents handle well, with a person approving the final spend.

So build a small one. An agent that watches your three main suppliers and tells you when prices move, or pulls a month of quotes into one comparison. You do not need to be a developer, and I’ve written about why building with AI beats just using it. Once you have watched an agent struggle with a badly organised supplier website, you will never look at your own site the same way again.

Where It Goes Sideways

None of this is risk-free. Payments and fraud people are openly worried about telling a legitimate shopping agent apart from a malicious bot, and the standards for doing that are still being written.

My rule is simple. Do not give an agent write access to anything you cannot afford to have go wrong. Anything machine-facing should be read-only where possible, priced, logged and capped. The endpoint on my site has a hard daily limit and records every request, so if something misbehaves, the damage has a ceiling.

And treat it like any other change to how the business runs. An agent-facing page that nobody owns and nobody updates is just another AI experiment that looked good once. Put a name against keeping it current.

It Will Not Announce Itself

I keep coming back to how unremarkable that first handshake felt. One machine asked, another answered, a few cents moved, done.

I suspect this whole shift is going to look like that from the inside. No launch day, no headline telling you it has arrived. Just a slow thinning of enquiries for the businesses agents could not read, and a story about a tough year that everyone in the building believes.

People will still choose who they trust. That part of business is going nowhere. What is changing is who decides whether that trust ever gets a chance to start, and I would much rather be readable to that gatekeeper now, while it is cheap and still a little strange, than work it out in 2029 from a sales graph that stopped going up.

Frequently Asked Questions

What is agentic commerce?

Agentic commerce is buying and selling where an AI agent does some or all of the work on behalf of a person or a business. The agent searches for options, compares them against what its owner asked for, and in some cases completes the purchase. Gartner expects 90 per cent of business-to-business buying to be handled through AI agents by 2028. On the consumer side, most people currently want agents to research and compare while they keep the final decision.

Will AI agents really buy from small businesses?

They are already starting to research on behalf of buyers, and that is the part that matters most right now. Logica research for Commerce and PayPal found 63 per cent of Australian online shoppers are interested in trying AI shopping tools, while only 4 per cent want AI help at the checkout. So for a small business, the near-term effect is mostly about which businesses a person gets shown. Machines placing orders comes later. If an agent cannot read your prices, services and booking options, you are less likely to make the shortlist.

How do I make my business readable to AI agents?

Publish the facts a careful buyer would want, in plain language, in one current place. That means prices or honest ranges, what each service includes, turnaround times, service area, terms, and a way to book or request a quote without a phone call. Add structured data (schema markup) to your website so machines can identify your business, services and prices, then test the result by asking an AI tool to compare you with competitors in your area.

Do I need an API or crypto payments for AI agents to buy from me?

Not yet, for most small and medium enterprises. Machine-payable endpoints exist, and protocols like x402 let an agent pay for something over the web in a single request, but the bigger win this year is making your existing information clear, current and complete. Build the foundations first. If your business sells something that is naturally bought in small, repeatable units, such as data, reports or bookings, a direct agent interface may be worth exploring after that.

Is it safe to let AI agents transact with my business?

It can be, with sensible limits. Standards for verifying legitimate shopping agents are still developing, and fraud teams are watching closely. Keep anything machine-facing read-only where possible, set spending and usage caps, log every request, and keep a human approving anything that changes an order, a price or a customer record.