/ agent readiness

The buying is moving to machines. Be readable when it arrives.

Buying software now asks for price, stock and lead time by itself, and it acts on whatever comes back. We make an industrial catalogue readable, answerable and auditable by those agents, inside your own infrastructure and without moving your data.

Check your own domain on the right. Twelve checks, three seconds, no account, nothing stored, no private address requested. The quote underneath is generated from your own scan.

agent-readiness /api/check?url=
mscdirect.com fastenal.com grainger.com dxpe.com this site
waiting for a domain
enter your address, or scan an example above
ready
twelve checks, reported separately, with response codesalso as an API
15 of 22
do not answer a machine: 8 refuse, 4 serve a placeholder, 3 never reply
0 of 22
name an AI agent in robots.txt, or offer a way to query a catalogue
3.2 / 10
readiness of Hubbell (hubbell.com), an electrical manufacturer, audited 16 October 2026

/ in plain words

Software is starting to do the shopping

A buying assistant, not a person, is increasingly the one asking what you have, how many, how fast and how much. It asks your website the way one program asks another. If your site only answers people, the assistant moves on and you never hear the request you lost.

A PERSON ARRIVESyoubrowser, ordinary visitor200 OKSOFTWARE ARRIVES{ }the same request, no browser403Same catalogue, same part. The only difference is who is asking. 15 of the 22 sites we measured answer this way.
the same catalogue and the same part number, seen by a person and by a program

Nothing here needs a new system. Your catalogue and your prices stay where they are. What changes is that a machine can ask about them, under rules you set, and you receive the log of what was asked.


/ the whole thing in one picture

Four layers. Three belong to Google, OpenAI and Stripe.

Autonomous purchasing was standardised in four layers, and three of them are owned by the platforms. The first one, the layer that lets anyone ask your catalogue a question, is the only one that has to run inside your own systems. That is the layer this product builds.

LAYER 1 / ACCESSHow an assistant asks your catalogueMCP, Anthropic, 2024Linux Foundation, Dec 2025YOUR LAYERLAYER 2 / COMMERCEFrom finding it to ordering itUCP, Google with Shopify, Jan 2026ACP, OpenAI with Stripe, 2025not yoursLAYER 3 / MANDATEProof the buyer allowed the spendAP2, Google, 2025FIDO Alliance, Apr 2026not yoursLAYER 4 / SETTLEMENTPaying cents per questionx402, Coinbase, 2026MPP, Stripe with Temponot yours
the stack as it stands in October 2026, with the owner and the date each layer was standardised
/ in plain words

Layer one is the door

The door an assistant knocks on when it wants to know something about your catalogue. It is yours, and today it is closed.

/ in plain words

Layer two is the counter

How the order gets placed once the assistant has decided to buy. Google and OpenAI own this one.

/ in plain words

Layer three is the permission

Proof that a real buyer allowed the spend. Owned by the same platforms, handed to the standards bodies.

/ in plain words

Layer four is the change

The payment, counted in cents per question. Stripe and Coinbase own it and charge almost nothing.

Read the detail layer by layer, with owners and dates: the agent economy.


/ how it works

Four steps, and your data never leaves

The gateway is a reader, not a warehouse. It stands in front of the systems you already run and answers questions on their behalf, under rules you set, and every call is logged for you.

1Assistant asksfor a part number, a price, a leadtime->2Your rule applieswhat this caller is allowed to see->3Your system answersfrom the ERP or price list you runalready->4Answer is loggedthe log is delivered to you, notkept by us
one question, from the assistant to your own system and back

What it is, in one sentence

A service window on the catalogue you already run, answering six questions that any clerk would understand, and refusing everything else until you say otherwise.

What it is not

Not a new ERP, not a data warehouse, not a marketplace, and not a place where your contract prices are copied. In the first month it does not even accept orders.


/ before and after

The same question, asked twice

Three questions a buying assistant asks, and what it gets from a typical distributor site today against what it gets once the gateway is in place.

TODAYWITH THE GATEWAYDo you have it, and how manya page with no stock figurequantity and lead time->What does it cost usnothing a machine can readyour price for that account->Send the certificatethe agent guessesthe document, with its fingerprint->
left: what an agent receives from a typical distributor site today. right: the same question with the gateway in place

/ where the industry stands

15 of 22 do not answer a machine

We requested twenty-two US industrial distributors as a program, not as a browser. Each bar is one check across all of them: green passes, amber partly answers, red refuses. The right column is the same thing in plain words.

passwarningfailA machine is allowed in at all7 of 22There is a map of what you sell9 of 22A file describing you to machines exists1 of 22Machines can be authorised to act1 of 22Price and stock are readable by softwarenone quote fullyRules for AI agents are publishednone answer fullyProducts are stated in machine termsall 22 refuseYour catalogue can be asked a questionnone answer fullyA description of what may be requestednone answer fullyYou describe yourself to other agentsnone answer fullySomething callable exists on your sitenone answer fullyPayment per call is possiblenone answer fully
field run 2026-10-16, twelve checks, twenty-two US industrial distributors, one location, one afternoon

The pattern is not random. The checks almost everyone passes are the free ones: a sitemap, and being allowed in at all. Every check that requires a decision about machines fails across the whole industry at once, which is why this is a decision problem rather than a technology problem. Every row, every response code: the evidence, raw data in field-run.json.


/ why now

Eighteen months, and a deadline already set

Every marker below is a standard that shipped, or a compliance date a regulator has already published. This is not a forecast.

2024MCPa tool for agents2025ACPagent checkoutJan 2026UCPsearch to orderApr 2026AP2signed mandatesQ3 2026x402138M paymentsFeb 2027EU passportbattery deadlinethe stack was standardised in eighteen monthsand the first compliance date is already set
standards that shipped between 2024 and 2026, and the first hard compliance date for an industrial product

/ what an agent asks

Three questions, and what most sites return

We sent the requests a procurement agent would send to seven distributor sites on 16 October 2026. This is what came back.

What the agent asksWhat most sites returnCost to you
"Do you have 6205-2RS in stock"A 404 from a script, or a page with no price and no stockThe agent recommends another supplier
"What does it cost for my account"Nothing machine-readable: the price is rendered for a human onlyYou are compared on list price, not on your agreement
"Send the datasheet and the certificate"No document endpoint exists, so the agent guesses from a descriptionCompliance risk lands on the buyer, and then on you

Four of the seven refused the search path outright, one served a captcha, one returned a large page with no price anywhere in it, and one did not answer at all. None returned a price, a stock figure and a document together.


/ what we do

Measure it, serve it, sign it

Three products in sequence. The diagnostic exists so you never buy the gateway on faith, and the records exist so what an agent relays about you can be checked years later.

Step one

Diagnose

$2,500, credited

A live agent run against your public surface and, under NDA, your catalogue. You get the raw transcript: every request, every response code, every search that returned nothing.

  • One domain, five working days
  • Gaps listed by product line
  • Credited against month one
How it runs ->
Core product
Step two

Serve

$1,500/mo per catalogue

Six tools over the catalogue you already run, under your access rules, read-only for the first month, with every call logged and delivered to you.

  • MCP and REST, in your VPC or ours
  • Your prices, your rules, per buyer
  • Every call logged
The platform ->
Step three

Prove

$500/mo per catalogue

Signed records for the facts that have to survive an audit, each with the basis of the claim stated separately and the evidence bound by its fingerprint.

  • Ed25519 signatures, sha256 evidence
  • Published keys, offline verification
  • Revocable, never silently
Verifiable records ->

/ what you are choosing between

Four routes, and only one makes the catalogue answer

The work is already being bought in three other forms. Comparing them honestly is more useful than another list of features, and it tells you what our floor price is competing with.

RouteWhat it costsWhat it actually changesDoes your catalogue answer?
Agency retainer, AI visibilitymedian $6,850 a month (First Page Sage, 1,146 quotes, Oct 2026)They write content and try to get you mentioned in AI answers. Your catalogue still cannot be asked a question.no
A visibility dashboarda subscription, usually a few hundred a monthIt shows where you appear. It changes nothing in your own systems.no
Build it yourselfyour developers, several months, then maintenanceFull control. You also own the mapping to conventions that keep moving.partly
Connect.ARLI gateway$1,500 to $4,500 a month, metered, from $8,000 to connectReads the catalogue where it lives and answers the six questions, under your rules, with every call logged.yes

The first two routes sell attention. The third sells control and costs you a year. The fourth is the only one that changes what a machine can get out of your catalogue, and it is priced below the first one. The longer comparison, in plain words: plain words.


/ what a year costs

Priced below the market, and shown in one picture

Market rates for this work are published, so we price against them rather than against our own ambitions. The median agency retainer buys attention; ours buys a catalogue that answers.

One catalogue, Standard$29,300Two catalogues, full stack$65,500Median agency retainer$82,200
first year, including integration and the diagnostic, from our published price list

Verified by buyers, brokers and auditors: free, forever, without an account. Full price list and worked examples: pricing.


/ the questions a buyer asks

What happens to your data and your prices

Six answers that a procurement review asks for, in the order the questions usually arrive.

/ 01

Your data does not move

Read-only in the first month. The gateway answers from your systems and stores nothing about your catalogue contents.

/ 02

Your prices stay yours

Your rules decide who sees which price. No contract price is published. We never see the contents of a buyer agreement.

/ 03

The log is delivered to you

Every call, every refusal, every question that returned nothing, in your hands rather than in our dashboard.

/ 04

Revocation is one command

A catalogue, a record or an entire deployment can be withdrawn without a support ticket.

/ 05

It runs in your cloud or ours

Enterprise deployments run inside your environment, with SSO and your own audit trail.

/ 06

No SOC 2, and we say so

We are small and honest about it. Ask for the questionnaire and we fill it in before any commercial conversation.

Data flow, obligations and the cryptography behind the records: security.

/ what we will not tell you

Agent traffic is small today

You are not buying traffic from us and we do not promise any. What you are buying is the position, and the work that makes a catalogue correct for machines also makes it correct for people.

0.2%

of sessions

came from agents on enterprise sites, in the figures we could verify.

11%

of consumers

say they would let an AI complete a purchase on their behalf today.

2027

Gartner expects

more than 40% of agentic projects to be cancelled before the end of it. We do not sell agent traffic.


$ connect.arliwork.com --diagnose

Start with one domain

Send a domain and a product line. We run the diagnostic and send the transcript before any invoice.

hq@arliwork.com · Shenzhen Capernaum Technology Co., Ltd · PIAKHOUSKI DZMITRY