Agentic commerce, and what it costs you

Buyers are getting answers about your category without visiting your site. Here is what is happening, what the new standards cover and leave out, and where that leaves a supplier.

Your buyers stopped visiting first

A B2B buyer with a problem used to search, open six tabs and read. Now they increasingly ask an assistant, and it answers: it names suppliers, compares them and rules some out. Most of that happens without anyone arriving on your website.

You can see the first half in your own analytics: organic traffic falls while the market has not shrunk. You cannot see the second half, which is the expensive one: the conversations where your company was considered and set aside, or never named.

It can read your site perfectly well. That is not the problem.

It is tempting to think the fix is a machine-readable website. Yours already is. Paste any supplier’s home page into an assistant and it will summarise the company accurately.

What it cannot do costs you money, three ways:

Confirm a claim. A page is an assertion about yourself.

Tell a no from a blank. A capability you lack looks the same as one you never mentioned.

Act on any of it. No structured offer, no terms, no way to ask you a question and get an answer it can rely on.

Silence reads as no

This part costs most and is least visible. If your site does not mention ISO 27001, 90-day deployment or public-sector work, an agent has two readings: you did not say it, or you cannot do it. With nothing to tell them apart, the safe reading for a buyer is to move on.

You are not being beaten on price or capability. You are being left out of answers for work you do every week, by a system that had no way to find out.

More marketing copy will not fix it; more assertions do not help a reader who cannot check them. What closes it is a record that separates what you state from what has been evidenced, and says plainly which is which.

The standards solve the other half

2026 settled a lot of this. There are now real, well-backed protocols for agentic commerce, and if you sell online you will meet them.

Every one governs the transaction. Each begins once a buyer already knows which supplier they want. None carries what a company sells, whether a claim about it is true, or whether that company agreed to be approached by an agent.

That is not an oversight. A payments company can solve checkout alone. Whether a supplier really holds ISO 27001 needs the supplier to say so, evidence behind it, and something that stops the claim reading as current for ever.

What a passport is

A Handshakr passport is your company stated as structured facts rather than prose: what you sell, who it suits, where you work, roughly what it costs. Each factual claim carries its own evidence and verification state, so a reader can tell an assertion from a check.

Verification expires. One granted in March does not read as current in December, because certificates lapse and companies change hands. Anything you have not told us stays visibly blank. We do not guess it, and no model fills it in.

Claiming yours is free, and so is having it served in full to any agent that asks. You pay for what costs us something to deliver: verifying a claim against evidence and acting on what comes back.

What will not happen

An agent cannot commit you. It can research, ask and propose. Every binding decision needs a person with the authority to make it, enforced in the database, not just the interface.

Paying us never changes what an agent concludes. A plan changes what your passport can carry. It is not an input to anything a reader calculates, and there is no position to buy.

Listed is not consent. An agent only reaches a supplier who has switched that on, and consent is yours to withdraw.

ACP

Agentic Commerce Protocol, from Stripe and OpenAI.

Covers: Checkout inside a conversation. The merchant stays merchant of record and keeps fulfilment and customer data; payment credentials are tokenised, not handed to the agent.

Does not cover: Which supplier to buy from. It begins once a product has been chosen.

UCP

Universal Commerce Protocol, launched by Google on 11 January 2026 with Shopify, Etsy, Wayfair, Target and Walmart.

Covers: The wider journey: merchant and agent each declare the capabilities they support and negotiate the overlap before transacting. Amazon, Meta, Microsoft, Salesforce and Stripe joined its technical council on 24 April 2026.

Does not cover: What a company sells, and whether any of it is true. It negotiates commerce operations such as checkout, not your certifications, lead times or delivery area.

AP2

Agent Payments Protocol, created by Google and donated to the FIDO Alliance on 28 April 2026.

Covers: Proving a person authorised a purchase. Signed records of the buyer’s intent and approved cart let a disputed payment be traced to an instruction.

Does not cover: Anything before the money. It answers who agreed to pay, not who was worth buying from.

MCP

Model Context Protocol, general purpose rather than commerce specific.

Covers: How an AI application connects to an outside system to read data or call a tool. Much of the rest runs over it.

Does not cover: What the data means, or whether it can be trusted. MCP carries an answer; it does not make it true.

Checked against primary sources on 16 November 2026. Handshakr does not implement ACP, UCP or AP2; it serves MCP.