Skip to content
← Blog
Agentic Payments

Agentic Payments in Australia: From a Photo of Your Living Room to a Kmart Delivery

· 9 JULY 2026 ·8 min

Send an agent one photo of your living room and one sentence: 'Recommend some furniture from Kmart, show me how it'd look, keep it under $300, and order it.' Every piece of that loop exists in Australia today — and we built MilyPay to be part of the plumbing.

Listen: Agentic Payments in Australia: From a Photo of Your Living Room to a Kmart Delivery~10 min read

Here's the message. One photo of your living room, and:

"Hey — can you recommend some furniture and ornaments and things Kmart sells that would suit this room? Show me how they'd look, check what's in stock near me, keep the whole thing under $300, and order it."

Every piece of that request is either live in Australia today or one integration away. That's what makes it worth walking through properly — not as a demo fantasy, but as a systems diagram of which rail does which job. And because we've spent the last year building one of those rails, we'll be upfront: the last section is about our own product.


The Half That Already Exists: Kmart's Joy

Kmart Australia shipped an AI shopping assistant called Joy, built on Google's Gemini Enterprise agentic platform. It already does the perception half of our scenario: natural-language product search, photo uploads for recommendations, virtual try-on for clothing, and a "see it in my space" feature that previews furniture and homewares in your actual room. It launched on kmart.com.au with the app rollout following mid-2026.

So "look at my living room and suggest a side table, a throw, and some ornaments that match" is not speculative — that's the product. What Joy can't yet do is the second half of the sentence: hold a budget and buy it. Today the human still drives checkout. The agent recommends; you tap through the cart like it's 2024.


The Half That's Arriving: The Payment Leg

Closing the loop means an agent that can transact. In Australia, three rails are converging on that, and they slot into different parts of the job.

1. Mastercard Agent Pay — the card rail learns about agents

Mastercard ran Australia's first authenticated agentic transactions using Agent Pay: AI agents registered with the network, transacting with agent-specific tokenized credentials rather than a raw card number. The merchant can verify a real, registered agent is paying within a mandate the cardholder approved. The short version: it brings agents inside the existing card network, chargebacks and all. For a retailer like Kmart, that's the path of least resistance — the checkout already takes Mastercard.

2. PayTo — the budget your bank actually enforces

Australia has a quietly perfect primitive for agent budgets: PayTo, the mandate system on the New Payments Platform. A PayTo agreement is a standing authorisation — payee, amount limits, frequency — that lives in your banking app, where you can see it, pause it, or cancel it. Swap "Netflix" for "my shopping agent, $300/month cap" and you have exactly the guardrail the living-room scenario needs: the agent can spend to the cap and cannot spend past it, enforced by your bank rather than by a system prompt. Agent-initiated PayTo mandates are the most natural bridge between Australian bank accounts and agentic commerce, and the banks already run the infrastructure. (We covered where PayTo sits in Australia's broader payments overhaul earlier this year.)

3. x402 and stablecoins — the machine-to-machine layer

The third rail is the one we build on: x402, HTTP-native payments settling on Solana in ~400ms for fractions of a cent. In the Kmart scenario it's less about the $300 checkout and more about the plumbing around it: your agent paying per call for the data the errand needs — validating the delivery address, checking the weather for the delivery window, tracking the parcel.

This layer is already live in Australia, because we shipped it. MilyPay, a Milysec company, sells agents pay-per-call access to Australian data at fractions of a cent per request, settled in AUDD, the AFSL-regulated AUD stablecoin — the settlement leg we made the case for in May. No API keys, no accounts: the agent hits the endpoint, gets a 402 challenge priced in AUDD, pays, and the data comes back in the same round-trip. Machines paying machines, per call, in Australian dollars end to end.


The Full Loop, Played Out

Put the rails together and the photo message becomes a pipeline:

  • Perceive. The agent reads the room from your photo — style, colours, gaps. (Joy's photo search does this today.)
  • Recommend & visualise. It pulls candidates — a side table, a lamp, ornaments, a throw — and renders them into your room with "see it in my space."
  • Check stock. Per-store availability near your postcode; swap out the lamp that's only in stock 400km away.
  • Budget. You said $300. The agent optimises the basket to $287 including delivery — and the PayTo mandate means $300 is a hard ceiling, not a suggestion.
  • Order & wait. The agent checks out via an authenticated Agent Pay credential, gets the order confirmation, and watches the delivery tracking so you don't have to. You get one message: "Arriving Thursday."

The Last Mile Is Human: Assembly, Seniors, and Gifts

The most underrated part of the scenario isn't the payment — it's what happens after the boxes arrive. Flat-pack furniture assumes an able-bodied adult with an Allen key and patience. A lot of real customers aren't that:

  • "I'm a senior and I don't know how to build this — I need assembly." The agent books a local assembler (the Airtasker-style gig layer Australia already has) for the delivery afternoon, paid from the same mandate.
  • "It's a gift for my mum in another city." The agent ships to her address, schedules assembly for after the delivery window, and sends you the photo when it's done. Your mum never touches an Allen key; you never leave the chat.
  • Office fit-outs: "Six desks, four shelving units, under $2,000, assembled by Friday, invoice the company card" — the same pipeline, with the mandate pointing at a corporate account instead.

This is where agentic commerce stops being a payments story and becomes an accessibility story. The customers who benefit most from "talk to it like a person, and it handles everything including the screwdriver" are precisely the customers self-service e-commerce has been quietly failing for twenty years.


The Piece We Ship Today

We're not waiting for the whole loop to exist before building our part of it.

Milysec is a Melbourne venture studio focused on AI and Solana development, and we've been shipping x402 integrations and writing the playbook for AUD-denominated agent payments — from why Australia needs its own stablecoin to how x402 turns HTTP into a payment rail. MilyPay is that playbook shipped: the x402 service provider for the Australian market, live today.

What it does, concretely:

  • Australian data, per call. Property address validation for the delivery, business identity checks, weather for the delivery window — with parcel tracking on the roadmap.
  • Priced in cents and fractions of cents. No subscriptions, no minimums. The agent pays for exactly the calls it makes.
  • Settled in AUDD via the PayAI facilitator — Australian dollars end to end, on Solana, in under a second.
  • Discoverable by agents. MilyPay services are listed in the Pay.sh catalog and MCP, so an agent can find, price, and pay for them at runtime without a human provisioning anything.

It's a working preview of the Kmart scenario's plumbing — the exact shape retail stock, delivery, and assembly APIs will take when they join the loop.


The Honest Read

None of this ships without answering the boring questions. Who eats the loss when the agent orders the wrong ottoman — the consumer, the agent platform, or the merchant? (Agent Pay's answer: the card network's existing dispute machinery. x402's answer: largely nobody, yet.) Australian Consumer Law guarantees don't evaporate because a bot clicked buy, but the ACCC hasn't said much about mandates and autonomous checkout. Agent authentication is early — registered-agent credentials exist on the card rail, while the open-protocol side is still standardising identity. And the practical risk isn't sci-fi: it's an agent confidently ordering six of something because a product page listed a per-carton quantity. Budgets enforced by banks (PayTo) and mandates visible in your banking app are the right defaults precisely because agents will sometimes be wrong.

But the direction is set. The perception layer is live at Kmart, the payment rails did their first authenticated agentic transactions this year, and the budget primitive already sits in every Australian banking app. The photo-to-delivery loop isn't a whitepaper — it's an integration backlog.


Further Reading


Milysec builds AI applications and Solana blockchain applications in Australia. MilyPay is our x402 service layer for the Australian market. If you're building agentic commerce — or want your API discoverable and payable by agents — get in touch.

Agentic PaymentsAustraliaRetailx402PaymentsAI AgentsAUDDSolana

Milysec · 9 July 2026

About Milysec

Milysec is an Australian venture studio building at the intersection of AI and Solana. We ship products and infrastructure that make programmable finance and agentic systems usable in Australia.

Learn more about Milysec →