Both sides of the exchange

Your next buyer may never load your page

Two automations are happening at once, and almost everyone is building for only one of them. The work is becoming agents — and so is the audience. When both ends of the exchange are machines, marketing stops being persuasion and becomes an interface problem.

What changed

Four shifts, all of them already measurable

01

Discovery stopped sending traffic

Search answers instead of linking. The optimization target moved from holding a rank to being the thing that gets cited — and a citation cannot be bought the way a keyword could.

69%

of searches end without a click

02

The buyer is becoming an agent

Assistants shortlist, compare and increasingly transact. Agent-ready has a definition: your product data, your APIs and your flows legible to a machine with no human mediating.

90%

of B2B buying agent-intermediated by 2028 (Gartner)

03

The funnel collapses into two jobs

An agent reads your site long before a person appears. What is left is being discoverable and being convertible — everything in between was scaffolding for a human reader.

89%

of B2B buyers already start with genAI

04

Attribution dies with the click

In an agent-mediated funnel there is no click to attribute. The only way left to know what worked is to have run the action and measured the outcome.

0

clicks to attribute in an agent funnel

Being represented

A product needs an agent-readable presence like it needs a website

An assistant that summarises your category and recommends three products is making an editorial decision from whatever it can read, parse and trust about you. That surface is new, mostly unowned, and — unlike ad inventory — not for sale.

Bladesmith writes it and ships it the same way it ships anything else in your product: as a pull request you review.

  • llms.txt

    A plain-language map of what your product is, who it is for and what it costs, written for the model that will summarise you.

  • JSON-LD

    Structured product facts — pricing, features, category, comparisons — so an assistant states them instead of inferring them.

  • MCP endpoint

    Your product, callable. The fastest-growing AI-era companies let other agents operate them; this is that surface.

  • .well-known

    The machine-readable identity an agent looks for before it trusts what your marketing says.

Being measured

Share of model is the new ranking

How often does an assistant name you when someone asks it a buying question in your category — and who does it name instead? It is the one number that tells you whether any of this is working, and no incumbent reports it.

Measuring it is only half the job. The other half is doing something about it, proving the change moved the number, and remembering what worked. That is the loop, pointed at the channel that replaces search.

How the loop works

Share of model

“What's the best tool for automated refunds?”

Illustrative

We go first

This site publishes its own machine-readable presence

Structured data in the page, and a plain-text brief written for the assistants that will be asked about us. If we would not do it to ourselves, we have no business proposing it to you.

$15T

routed through agent-to-agent exchanges by 2028

Gartner's estimate of agent-intermediated B2B commerce.

25%

of enterprise software purchases AI-mediated by end-2026

The rails — Stripe ACP, Google AP2 — are being laid now.

+35%

organic clicks for brands cited in AI answers

Citation is the currency that replaced ranking.

Find out what the machines currently think you are

Paste your URL. The plan includes what an assistant can read about you today — and the first fixes worth shipping.

Connect GitHub instead

Free tier · no credit card · nothing is sent, posted or shipped without your approval