Buyer-in-Chat Stumbled. Merchant-Operated Is the Real Agentic Commerce.

Sam Bakker · Founder, gocushy · July 22, 2026

When people say "agentic commerce," they usually mean one of two things, and they almost never say which. The confusion matters, because the two bets can succeed or fail independently — and betting your company on the wrong one is an expensive mistake to make quietly. So let's separate them cleanly, tell the honest story of how each is going, and explain why we placed gocushy where we did.

Two bets wearing one name

Bet one — buyer-in-chat. A consumer is talking to an assistant, asks for a pair of running shoes or a course, and completes the purchase without ever leaving the chat. The assistant is the storefront. The behaviour change being wagered here is enormous: people have to start buying inside a chatbot.

Bet two — merchant-operated. The buyer's experience doesn't change at all — they still land on an ordinary checkout page and pay. What changes is who runs the seller's back office: pricing, offers, order bumps, upsells, subscriptions, affiliates, invoices, refunds. Instead of a human clicking through a dashboard, the seller's AI agent does the repetitive operator work. The behaviour change here is small and already underway: a solo founder who'd rather say "set this up" than click through fourteen screens.

These are not the same thing. It's entirely possible for one to arrive years before the other.

The honest scorecard on buyer-in-chat

Here's where candour is more useful than hype. OpenAI launched an in-ChatGPT "Instant Checkout" and then shut it down in March 2026 after lackluster uptake — by then roughly 30 merchants had integrated. That's not a rumour we're spinning; it's the most visible attempt at consumer buyer-in-chat, and it stumbled.

But read the wreckage carefully. The underlying Agentic Commerce Protocol (ACP), co-developed by Stripe and OpenAI, survived as infrastructure. And across 2026 the payment networks kept building parallel rails: Visa Intelligent Commerce, Mastercard Agent Pay, Google's AP2, PayPal. The plumbing for machines to transact is real and getting realer — it just turned out the first consumer-facing product was early, not that the idea was wrong. Buyer-in-chat isn't dead. It looks like it showed up before its audience did.

If you're building a company, "early" is a genuinely hard word. Early can mean five years or fifteen, and you don't get to know which while you're waiting.

Why we bet on the operator, not the shopper

Merchant-operated is the sturdier bet for one blunt reason: it doesn't need consumers to change. Nobody has to learn to trust a chatbot with their card. The buyer still lands on a fast, mobile-friendly page and checks out the way they always have. The only thing that moves is the operator's seat — from a human in a dashboard to the seller's own agent.

And that change is already happening. The founder who drafts emails with AI, builds pages with AI, and reads their numbers with AI, then breaks out of the conversation to hand-operate the cart — that person wants the last manual thing automated too. We don't have to manufacture the demand or wait for a cultural shift. We just have to build the tool the shift is already reaching for. If the phrase is new to you, we wrote a primer on what an MCP checkout actually is.

Where gocushy sits

gocushy is a checkout and funnel platform that lives entirely on the merchant-operated side. It runs on your own Stripe account through Stripe Connect direct charges — money lands with you on every sale, and gocushy never holds funds and never becomes the merchant of record. On top of that, 28 MCP tools let your agent operate order bumps, one-click upsells, subscriptions, coupons, affiliates, Stripe Tax, invoices, dunning, refunds, and an activity log it can read back.

Because we're letting a language model near money, the guardrails are the product as much as the tools are: prices are enforced server-side so a checkout can't be mispriced, page copy is limited to whitelisted checkout blocks, countdowns hide at zero and can never fake-reset, buyer-sensitive pages sit behind signed URLs, and delivery warnings catch a product you forgot to attach. Meanwhile the buyer sees none of it — just an ordinary page that loads fast on a phone.

The part where we might be wrong

Now the honest hedge, because a build-in-public post that only lists reasons we're right isn't worth much. We could be wrong about timing. Buyer-in-chat could arrive faster than we think — the rails are being laid quickly, and a single good consumer experience could pull the future forward by years.

Here's why that scenario doesn't sink us. gocushy speaks MCP and settles on Stripe — the same two standards the buyer-in-chat world is being built on. ACP came out of Stripe and OpenAI; the card networks' agent rails settle through the same payment plumbing we already sit on. So if in-chat buying does take off, a gocushy store is positioned to be reachable through those rails rather than stranded on the wrong side of the shift. We're not betting the whole company on merchant-operated instead of buyer-in-chat; we're building on the shared foundation, and starting where the behaviour change has already happened.

That's the difference between conviction and stubbornness: pick the bet you believe in, then make sure being early on the other one doesn't kill you.

Frequently asked questions

Sam Bakker is the founder of gocushy. He has a long track record building launch and funnel software for creators, and builds from New Zealand.