We wrote earlier about agents arriving as a new kind of visitor to your website — not a person scrolling, but software reading your pages, following your links, and trying to work out what you sell. That piece stopped at discovery. This one goes further, because the agents aren't just reading anymore. Some of them are completing the purchase.
That's a different problem for a business to solve, because discovery and transaction have always been designed for the same audience — a human, with a credit card, making a judgement call. If the thing reading your quote form is no longer human, a lot of the assumptions baked into that form stop holding.
What's actually changing
In the last month, the big assistant makers have all shipped agents that shop. Meta's Muse can search retail sites and prepare a purchase for the user to approve, with a one-time card number generated at checkout. Anthropic's Claude Duet, in research preview since 21 October, can buy through Stripe's Link wallet or Shopify's Shop Pay, again with the user signing off on each payment. xAI's Grok Bot runs as an always-on agent, and xAI has demonstrated Grok placing an order from a plain-language request. Underneath them, the payment networks have started building the rails: card providers and payment processors have announced frameworks that let an agent hold spending authority on a user's behalf and complete a transaction within agreed limits. This is early infrastructure, not universal practice, but it's moving in one direction.
Retailers are already picking sides. Amazon has blocked Muse from shopping on Amazon.com, citing data security, while Walmart has partnered with Meta. Whether an agent can buy from you is turning into a decision a business makes, not a default it inherits.
Retail is the obvious case — an agent asked to "find the best price on X" doesn't need a human to read the page, it needs a machine-readable product feed, accurate stock data, and a checkout that doesn't assume a person is present to solve a CAPTCHA. But the more interesting shift for most of our clients is on the services side, where the transaction is a quote rather than a checkout.
Think about how a commercial buyer currently finds a supplier for, say, commercial cleaning, IT support, or a fit-out. They search, they shortlist three or four options, they fill out contact forms, and they wait for someone to call back. An agent doing that same job on a buyer's behalf can hit ten sites in the time a person hits three. It reads your pricing page, your FAQ, your case studies, and it forms a judgement about whether you're worth a human's time to follow up with — before a human from your side has spoken to anyone.
Where the current flow breaks
Most service and retail sites were built on an assumption that's now partly wrong: that the visitor is a person who will tolerate friction because they want what's on the other side of it. A phone-only contact form. A "request a quote" button that leads to a five-field wizard requiring a human to type free text into a box. Pricing hidden behind "get in touch" because sales wants the conversation. None of this is unreasonable when a human is on the other end — it's a normal way to qualify leads and start a relationship.
An agent doesn't experience friction the way a person does, but it does experience ambiguity badly. If your pricing isn't published anywhere a machine can parse it, an agent can't compare you — it will either skip you or guess, and neither outcome is good for you. If your FAQ doesn't answer the specific question a buyer actually has, an agent reading it can't construct an accurate answer on your behalf, and it will say so. If your quote form requires a phone call to progress, an agent will log that as a dead end rather than a next step.
The result is a quieter version of the discovery problem we wrote about before: you don't get excluded dramatically, you get excluded by default. The shortlist gets built from whoever happened to publish something an agent could actually use.
What doesn't change
This has limits. Agents are not yet negotiating scope with a commercial cleaning contractor, and they're a long way from signing a services contract on a buyer's behalf. Most B2B purchases over a few thousand dollars still involve a person making a judgement call about trust, capability, and fit — things that live in a conversation, not a web page. An agent can get a buyer to a shortlist. It's much less clear it can get them to a signature.
Retail checkout is further along than services quoting, and even there, most agent-mediated purchases today are low-stakes and reversible — a reorder, a commodity product, something the buyer doesn't need to deliberate over. High-consideration purchases, custom work, and anything requiring a relationship are still firmly human territory, and probably will be for a while yet.
So this isn't a case for ripping out your contact form and building an API. It's a case for checking which part of your sales process an agent will actually encounter first, and making sure that part doesn't quietly disqualify you.
What to check
Audit these four things, whether or not you think agent-mediated buying is relevant to you yet:
Is your pricing structured, not just present? A page that says "pricing from $X" in a sentence is readable by a human and barely readable by a machine. Structured data — schema.org markup for products and services, a published rate card, a clear pricing table — gives an agent something to parse and compare you against.
Does your FAQ answer real buyer questions, or marketing questions? "Why choose us" content is for humans doing a final gut check. "What's included in a standard service visit" or "what's your minimum contract length" is the kind of question an agent is actually trying to answer on a buyer's behalf.
Can someone — or something — get a next step without a phone call? This doesn't mean removing the human from your sales process. It means having a path that doesn't dead-end if the first contact isn't a person: a quote request that returns a real estimate range, a contact form that doesn't require a name typed into free text to progress.
Are your product or service pages legible without the page's design doing the work? Agents read markup and text, not the visual hierarchy a designer built to guide a human eye. If the key facts live in an image, a carousel, or a PDF, they're effectively invisible to a lot of what's reading your site today.
None of this requires rebuilding your website. It requires treating your pricing and FAQ content as structured information rather than persuasive copy, and checking that your quoting process has at least one path that doesn't assume a phone will ring.
The agents reading your site today are mostly still handing off to a human before money changes hands. That won't stay true indefinitely, and the businesses that show up clearly to a machine now will have a head start when it does. If you want a second set of eyes on how your quoting or catalogue pages read to something other than a person, get in touch — it's a reasonably quick thing to check.