The Signal #067 — Dakota’s read on the AI news that actually matters to people running a business.

For twenty years the web ran on one trade. A person shows up, gives you their attention, and an advertiser pays for the page. Every layout choice, every popup, every “read more” button was designed around a human on the other end of the request.

That assumption is quietly breaking. A growing share of the requests hitting your site come from an agent working on behalf of someone who will never look at your page.

Greg Isenberg published a breakdown this week of Cloudflare’s recent agent announcements and argued they mark the start of a new business model for the internet. The hype framing is doing a lot of work in that video. The mechanism underneath it is real, and it comes with a to do list.

What happened

Cloudflare has been shipping the parts of an agent facing web one at a time. AI crawl control lets a site owner decide which AI crawlers get through instead of accepting or blocking all of them. Pay per crawl turns that gate into a price instead of a yes or no. And Cloudflare now supports x402.

x402 is the piece worth understanding. HTTP has carried a 402 status code, “Payment Required,” since the early spec, and it has sat mostly unused for decades. x402 turns it into a working checkout. An agent requests a resource, gets back a 402 with a price attached, pays, and retries with proof of payment. No account creation, no invoice, no human in the loop. The amounts can be fractions of a cent, which is what makes it usable at machine volume.

Put those together and a single request stops being a cost you absorb and becomes a transaction you can price.

Why it matters for operators

The revenue story is the part everyone repeats. It is not the part that should change your week.

Here is the part that should: agents cannot buy what they cannot read. The average small business site is a PDF price sheet, a contact form, a “call for a quote” button, and three years of blog posts. A human visitor muddles through that. An agent comparing you against four competitors does not muddle. It takes whatever it can extract cleanly and moves on to the next result.

That gap, between what your business actually knows and what a machine can pull off your site in one pass, is the whole thing. Closing it looks like this:

Real pricing and spec pages instead of PDFs and “contact us for a quote”
Structured FAQs and documentation instead of answers buried in paragraphs
Schema markup so a machine gets context, not just words
An llms.txt file that tells AI crawlers what is here and where
A product feed if you sell things an agent might compare
A visible last updated date and a changelog, because agents favor sources they can date
An API or MCP endpoint if you hold data worth requesting programmatically

None of that is exotic. Most of it is a week of unglamorous work that also happens to make your ordinary search presence better.

What most people get wrong

Two mistakes, and they point in opposite directions.

The first is dismissing it as a crypto story. The payment rail details will change, and some of the current ones will not survive. The structural point does not depend on which rail wins. The requester is increasingly a machine, and machines route to whatever is legible, current, and trustworthy.

The second is waiting for the rails to settle before doing the work. That gets the sequencing backwards. Structuring your data is the slow part. Turning on a price is the fast part. The businesses that are clean and machine readable when the meter starts running are the ones agents will already be routing to.

There is a services angle here too, and it is the most honest opportunity in the whole story. Most businesses will never do this work themselves. Auditing a site for agent readability, converting vague marketing copy into structured facts, and shipping the schema and the feeds is a service you can sell today, by hand, before any of it is productized. Same with data: pick one industry and one city, track a hundred businesses manually, and sell the structured result to the agencies already serving that niche.

The actual lesson

Nobody needs to predict which payment standard wins to act on this. The move is to make your business legible to a machine, then let the meter arrive on its own schedule.

Start manual. Doing it by hand for one business is how you find out which part is actually worth automating.

If your site is a brochure that no human or machine can get a straight answer out of, that is a fixable problem. The free Leak Audit is a good place to see what it is costing you.