The job

One list, four shops, one job to be done.

Mel is setting up a study nook for her Year 7 son before term: shelves and paint from Buildsworth, a desk and chair and printer from OfficePoint, tubs and a lamp from Daymart, and a monitor none of them stock.

One ordinary household job that happens to cross three Westmark Group brands, with the LoyaltyCo loyalty program as the thread running through them. The consumer need never changes. What changes is the machinery underneath, and the merchant gets to decide what its loyalty asset becomes.

One job, told three times

Study nook$900
  • shelves + paintBuildsworth
  • desk, chair, printerOfficePoint
  • tubs + lampDaymart
  • monitornone of them stock it
Illustrativethe scraper

Invisible loyalty, and a checkout it can't complete.

Mel asks ChatGPT to plan and price the nook. It reads each brand's logged-out website price; her loyalty benefits live behind her login, so to the agent they do not exist. Comparing on viewable shelf price plus a delivery fee it should not count, it decides Amazon is cheaper for the desk and the tubs and splits the order.

Then the cart asks for a sign-in the agent cannot provide, and the job lands back with Mel; she finishes it herself that night, re-keying everything, if at all.

ChatGPTReading the page…
buildsworth.com.au
browses like a person
Illustrativethe merchant's agent

She opens the app, and hands over the job.

This time Mel does not start in ChatGPT. She opens the LoyaltyCo app, its own front door, and gives the agent a job to do. It already knows the house: recent paint purchase in Colorex Natural White; OfficePoint's own order history says the household printer is six years old.

The merchant assembles the cross-brand cart with benefits applied natively, because the buyer's agent is the loyalty layer. The monitor is not stocked, so it sources TechFirst where her consented card view suggests she's shopped before. One approval, four orders, under budget.

LoyaltyCo
Mel · signed in
Setting up a study nook for my son before term — desk and chair, shelves, storage, and a monitor. Around $900.
Illustrativethe platform's agent

Same job, same rails. She never opens the merchant app.

Now we run it again, but Mel starts where she often starts, in ChatGPT, and the mandate goes with her. It discovers the same products through each brand's feed, builds the same cross-brand cart, and because she linked the LoyaltyCo GPT app, her benefits resolve at checkout.

Her card data, consented to the AI platform now, powers the personalisation: merchant-level rather than item-level, so it knows the Thursday pattern and where the card already shops, but not what they bought. Four orders confirm, including TechFirst.

Mel is perfectly happy. The merchant is still present, but it no longer decides; it has been demoted to a choice on the consumer's platform.

ChatGPT
Setting up a study nook for my son before term — desk and chair, some shelves, storage, and a monitor. Budget's about $900, use my LoyaltyCo GPT app.

The decision

Own the agent, or be a pass-through.

That is the choice underneath agentic commerce for a merchant loyalty business: build a buyer's agent good enough, and honest enough about its own gaps, that the customer chooses to open it — or watch the relationship move to whoever's agent they already use.

The agent has to be the easiest front door for the jobs customers run across the brands anyway. The merchant agent wins those moments by being better at them — not by being the only option.

The agent is a design proposition, and the customer has to want and trust it while the gravity pulls toward the AI platforms.

The jobs it has to win

Recurring household shopBirthday or Christmas giftBack-to-school setupHome-office refreshWeekend home-improvementGarden and outdoor reset

The comparison

The infrastructure does not pick a winner.

The payment rails and the authorisation trail are identical: the same credential, the same settlement, the same signed Intent → Cart → Payment trail, each brand still the merchant of record.

What differs: the front door, the discovery protocol, who personalises, what the merchant keeps and loses. All follows from one thing: who holds the mandate and the data consent. The merchant's agent, or the platform's.

Enabling stack

Data TokenspersonalisesAP2authorisesTAPvouchesClick to Pay / SRCresolves identityACP / UCPcheck outVICis the credentialICCaccepts itVisaNetsettles
TodayThe merchant's agentThe platform's agent
where the journey startsChatGPTthe LoyaltyCo app, the coalition's own front doorChatGPT
catalogue discoverabilityDOM scraping vs bot wallsMCP tools per brandACP feeds per brand
consumer legibilitynone, a generic plancoalition data (deep) + Data Token (broad), anticipates the needData Token consented to the platform
loyalty value at the decisioninvisible, behind the loginnative, the agent is the loyalty layerresolved at checkout via Click to Pay / SRC
authorisation record· unchangednoneAP2 Intent → Cart → PaymentAP2 Intent → Cart → Payment
who holds the mandate & consentnobodythe coalitionthe platform
merchant of record· unchangedeach brandeach brandeach brand
what the merchant keepsthe sale, if it survives the gauntletrelationship, standing mandate, data consent, cross-brand basket, dispute evidencemargin, merchant-of-record, dispute evidence
what the merchant losescomparisons it never saw; the legitimate orders its bot defences killedthe occasional out-of-coalition sale, deliberatelythe relationship, upstream of every cart
dominant failurebot blocks, invisible benefits, 3DS stallsmandate-scope disputes; out-of-coalition economics; earning the app-openplatform ranking decides; loyalty demoted to pass-through

Agent anatomy

What's inside the buyer and seller agent, each capability mapped to the protocol that implements it.

How to read each element

Protocol-backed

A machine standard handles it. Deterministic.

Cryptographic

Consent, mandate, payment. Signed and verifiable.

Improvised

No protocol — bridged by interpretation / browser automation. Fragile.

Absent

The element doesn't exist on this side in this state.

Pick a state and watch each part shift, or simply fall away

The two agents — the parts change with the state above

Buyer agentshops for the consumer
Identity & consent
Passkey · AP2 intent mandate
program-scoped consent
cryptographic
Discover the seller
pre-resolved endpoint
protocol-backed
Read the catalogue
MCP · agreed feed
protocol-backed
Choose & build the cart
ACP cart
protocol-backed
Spend authority
AP2 cart mandate · trusted-agent proof · e.g. Visa TAP
program-scoped
cryptographic
Payment instrument
network token · e.g. Visa VTS
cryptographic
Settlement & loyalty
in-program rails · LoyaltyCo points
points accrue
protocol-backed
Where they meet
agent ↔ agent · agreed protocols (MCP · ACP · AP2)
Seller agentthe store exposes this
Be discoverable
directly addressed
protocol-backed
Expose the catalogue
agreed MCP · feed
protocol-backed
Accept an agent order
native ACP checkout
protocol-backed
Verify agent & mandate
AP2 mandate · trusted-agent proof · e.g. Visa TAP · 3DS
cryptographic
Accept payment
network token acceptance
cryptographic
Settle & accrue loyalty
in-program settle · LoyaltyCo accrual
protocol-backed
The other question →Live demo
The Digital Shelf

Who decides what you see?

Built on Shopify's live Global Catalog over MCP: real retrieval, a look inside the selection, and the brands competing for a slot they can't see.

When you ask an AI to shop for you

Two things get decided before you see anything: what surfaces, and who’s doing the shopping.

Live demo
The Digital Shelf

Who decides what you see?

Built on Shopify's live Global Catalog over MCP: real retrieval, a look inside the selection, and the brands competing for a slot they can't see.

Enter →Illustrative
The Loyalty Agent

Who shops for you?

One household job, three horizons: today's fragile browser automation, the merchant's own agent, the platform's agent. What the merchant keeps, and what it loses, in each.

Enter →

The shelf decides what gets seen. The mandate decides who gets to buy.
Merchants need an answer to both.