I Made My Website Agent-Ready. Here's What That Actually Means.

What agent-ready actually means

Think about how you use your own website. A person lands on it, reads the words, looks at the pictures, and figures out what you do and what it costs. That works because people are good at reading a page.

An agent is bad at that, and it hates doing it. Making an AI read a whole marketing page to find your price is slow, expensive, and easy to get wrong — it might grab an old number, or a price from a different service, or just guess. What an agent actually wants is to ask a flat question — what does this shop charge, do they cover my town — and get back a clean, exact answer with no marketing wrapped around it.

Agent-ready just means I built that second path. The website still looks and reads normally for people. But underneath it, there's now a machine-readable version of the important facts — the services, the prices, the towns I cover, how a project works — sitting at fixed web addresses an agent can pull straight from. The same facts the humans see, just served in a way a machine can trust.

The honest part: almost nobody does this, and almost nothing reads it yet

I'm not going to sell you a revolution. Here's the real state of it, because the honesty is the whole point.

The newest piece of this — a browser standard called WebMCP that lets a site hand tools directly to an AI — is brand new. One developer recently scanned over 111,000 of the top websites on the internet looking for it and found it on exactly zero of them. The two known sites that had wired it up logged zero actual agent calls. It's a real thing, shipping in Chrome as a trial, with essentially no adoption yet.

The older, simpler piece — a plain-text file that hands AI a clean summary of your site — is doing better: roughly 28% of the biggest sites now publish one. But be careful what that means. When people study who's actually reading those files, it's mostly coding assistants like the ones software developers use, not the consumer AI search engines everyone's picturing. Google has said outright that it doesn't use them.

So why build any of it? Two reasons, both boring and both true. First, it barely cost me anything. Every one of these machine-readable answers is generated automatically from data my site already maintains for its normal pages — I'm not keeping a second copy of my prices, I'm publishing the same one in a second format. When a price changes on the site, it changes for the agents in the same build, automatically. Second, this is only going one direction. When the agents do show up in numbers — and the browser makers and AI companies are clearly betting they will — the businesses an agent can actually query and transact with are the ones that get picked. I'd rather be early and quiet than late and scrambling.

What I actually shipped

Three layers, plainest to fanciest.

Plain-text docs for AI. Every page on the site now has a stripped-down, plain-text twin an AI can read without wading through design and code, plus a single index file that lists them all. This is the part coding assistants already pull today.

Structured answers. A handful of fixed web addresses that return clean data instead of a web page — my full service list with real prices, the 197 towns I've built local pages for (with their ZIP codes), and how a project actually works from quote to launch to ownership. Plus a small table-of-contents file that tells an agent these tools exist and how to call them. No login, no scraping, no guessing.

Callable tools in the browser. Using that new Chrome trial standard, an AI agent working inside a browser can now call my site directly — ask it for services and pricing, check whether it covers a location, or pull the process — and get structured answers, not a screenshot it has to interpret.

Here's the whole thing in one concrete example. An agent wants to know if I cover Olathe. It calls the coverage tool with ZIP code 66061 and gets back, every single time, the same answer: yes, with the exact link to that town's page. No reading, no scraping, no maybe. And if I don't have a page for a town, it gets an honest not on the published list — but the wider metro and remote work are covered, here's how to ask. The machine gets a straight answer either way, because a straight answer is the entire point.

Why this matters if you run a business

Set my site aside for a second and think about yours. Right now your customers find you by searching and reading. Increasingly, some of them are going to send an assistant to do that for them — find me a plumber in Lee's Summit who can come Tuesday, book the one with good reviews under this price. When that assistant goes shopping, the businesses it can actually read and query cleanly are the ones that make the shortlist. The ones it can't make sense of just quietly don't get considered.

That's not real money yet — see the honest section above, it's early. But it's the same shape as every shift before it. The businesses that were findable when search moved to Google won. The ones that were fast when everyone moved to phones won. This is the next version of can they find you, and it's going to reward the people who were early.

The reason I'm telling you all this instead of keeping it a trade secret: this is now a thing I know how to build, because I built the whole thing on my own site first. It sits right next to the local SEO work I already do — same core idea, making you findable and legible to the systems people actually use to choose a business, just extended to the machines. If getting ahead of that sounds worth it, it's a conversation worth having. No pressure and no jargon, same as always.

What I deliberately didn't build

Restraint is part of the craft, so here's what I left on the table on purpose.

I didn't build an AI chatbot that answers questions about my business in its own words. There's a well-known system for that, and I looked hard at it. The problem is the economics: to answer a single visitor's question, its standard setup makes several calls to a paid AI model — not one, several, every time — and it needs an always-on server my simple, fast static site doesn't have. Structured, deterministic answers cost me nothing per question and can't invent a wrong price. For a business where the price being exactly right matters, that's the correct trade.

And I didn't build tools that let an agent do things — submit forms, make changes — without a human in the loop. The security people looking at this are blunt about it: the standard for handing tools to agents doesn't define a safety layer yet, so you start read-only. My quote form is set up so an agent can help someone fill it out, but it still lands in my inbox for me to answer personally, exactly like it always has. When the do-things side is safe to open, I'll open it. Not before.

The temptation with anything new is to build all of it because you can. The better move is to build the parts that are real, honest, and safe today — and wait on the rest.

So — are the agents here yet?

No. Not really, not yet. And rather than guess, I'm counting.

Every time an AI crawler or agent fetches a page on my site, it gets logged — which engine, which page, and specifically whether it's pulling the machine-readable files I just described. So I'm not going to hand you a made-up number about traffic that isn't there. What I can tell you is that the plumbing is in place, it costs almost nothing to keep running, and the day the agents actually start showing up, I'll have the receipts to prove it — page by page, engine by engine.

That's the whole thing, honestly. Not a revolution. A small, cheap, early bet that the web is about to get a second kind of reader — built on the one rule that runs through everything I do: give a straight answer, in plain terms, to whoever's asking. Even when who's asking is a machine.