BrokenRubikWhat is agentic commerce?
Agentic commerce is online buying and selling carried out by AI agents acting on a person's (or a company's) behalf — discovering products, comparing options, and completing purchases with little or no human interaction at the moment of sale. Instead of a shopper browsing a storefront, an AI assistant like ChatGPT queries a merchant's catalog, checks live pricing and availability, and places the order — which still has to land in the merchant's commerce and ERP systems to be paid, fulfilled, and accounted for.
That last clause is the part most definitions skip, and it is the part this guide is about. Agentic commerce gets covered as a consumer trend and a payments story. For the businesses actually selling — especially mid-market companies running NetSuite as their system of record — it is an operations and architecture question: what does it take for an AI agent to buy from you, correctly, at scale?
It is also a topic that just went through its first boom-and-bust cycle. Search interest in "agentic commerce" grew more than 4x year over year, an alphabet of protocols appeared (ACP, AP2, and friends), the first platform-owned checkout launched inside ChatGPT — and then the early data came in and forced a correction. Understanding what failed and what survived is the fastest way to understand where this is actually going.
The five levels of agentic commerce
A useful maturity ladder for this space — loosely modeled on the levels of autonomous driving — circulates in the payments industry; fintech analyst Simon Taylor credits a version of it to Stripe:
| Level | What happens | Who decides |
|---|---|---|
| 1 | Human chooses the items; the agent completes the purchase | Human |
| 2 | Agent researches and shortlists; human picks; agent buys | Human, assisted |
| 3 | Human delegates a bounded task ("buy coffee beans under $20"); agent completes it | Shared, with limits |
| 4 | Agent manages ongoing buying — replenishment, subscriptions — under policies | Agent, governed |
| 5 | Agent anticipates needs and purchases proactively | Agent |
As of mid-2026, real-world activity sits between levels 1 and 2 on the consumer side. The more interesting movement — covered below — is happening on the business-to-business side, where level 3 and 4 behavior (policy-governed, repetitive purchasing) maps directly onto what procurement teams already do.
The first reality check: what the Instant Checkout failure taught everyone
The first commercial implementation of agentic commerce was OpenAI's Instant Checkout, launched in September 2025 alongside the Agentic Commerce Protocol it co-developed with Stripe. The model: complete the purchase entirely inside ChatGPT, with OpenAI sitting between merchant and customer and reportedly taking a fee on top of payment processing.
The data arrived quickly, and it was not kind:
- Walmart put roughly 200,000 products into Instant Checkout starting in November 2025. By March 2026, Walmart reported that purchases made inside ChatGPT converted at one-third the rate of users who clicked through to walmart.com — a drop of about 66%. Walmart's product chief called the in-chat experience "unsatisfying."
- Merchant adoption stayed minimal. Analyst Marc Massar notes that only a handful of Shopify's millions of merchants ever integrated, while Amazon blocked external shopping agents outright and Shopify kept checkout gated behind its own Shop Pay.
- OpenAI confirmed it is phasing out Instant Checkout, shifting to checkout experiences that merchants control — either handing off to the merchant's site or running merchant-controlled apps inside ChatGPT. Etsy's native ChatGPT app, launched in May 2026, is the template: the catalog is discoverable in conversation, but the merchant runs the experience.
The lesson is not that agents don't matter. It is that the agent doesn't own the transaction — the merchant does. A checkout that was ruthlessly optimized for humans converted three times worse when buried inside a chatbot, and the platform-rent model gave merchants no reason to participate. What survived the correction is the plumbing: protocols for exposing catalogs to AI assistants, with orders flowing into systems the merchant already runs.
What survived: protocols and the merchant-owned order flow
Strip away the failed checkout experiment and agentic commerce in 2026 looks like this:
- The Agentic Commerce Protocol (ACP) survives as the open standard for making a catalog discoverable to AI assistants — products, prices, availability — with purchase handoff to the merchant's own flow.
- Google's Agent Payments Protocol (AP2) and a growing family of machine-payment standards address how agents authenticate and pay, with the card networks and PayPal all running their own initiatives.
- The merchant's stack does the real work. Discovery may start in a chat window, but inventory truth, pricing, tax, payment, fulfillment, and revenue recognition all happen in the merchant's commerce platform and ERP.
That re-centering is why agentic commerce stopped being a payments-industry story and became an architecture story. The question shifted from "which AI platform do I list on?" to "is my stack capable of serving a machine customer?"
What agentic commerce means for an ERP like NetSuite
NetSuite does not have an "agentic commerce" product, and this guide won't pretend otherwise. What it has is the position that matters: the system of record. Whatever agent or protocol ends up in front — an ACP catalog feed, a marketplace app, a B2B customer's procurement bot — the transaction has to land somewhere real: inventory committed, the order processed, payment reconciled, revenue recognized. That somewhere is the ERP, and that is true no matter which AI platform wins the front end.
Being the system of record in an agentic channel translates into three operational responsibilities:
- Catalog truth. Products, pricing, and availability served accurately and fast enough for a machine that checks before every recommendation. NetSuite already holds this data; the work is exposing it cleanly.
- Order ingestion that can't be fooled. Orders arriving from an agent-driven channel need validation, deduplication, and a clear path through fulfillment — the same sales order discipline you already run, hardened for a client that retries.
- Reconciliation. Whatever channel the order came from, finance closes the books in one place. A new channel that bypasses the ERP creates exactly the split-brain problem ERPs exist to prevent.
On the AI-access side, the piece that exists and is documented today is NetSuite's AI Connector Service — Oracle's implementation of the Model Context Protocol, which gives external AI models governed, permission-aware access to NetSuite data and actions. It is plumbing for agents working with your ERP rather than buying from it, but it is the same architectural direction: structured, validated machine access instead of screen-scraping and CSV exports.
Commerce-specific integrations between AI assistants and ERP platforms will keep being announced across the industry. Treat announcements as direction, not roadmap: verify what is actually available for your account before planning around any of it, and invest in the parts that pay off regardless — which is what the next section is about.
Wondering what agent-ready means for your stack?
We build and integrate NetSuite commerce architectures — SuiteCommerce, Shopify, headless, and the API layer underneath. Tell us your setup and we'll give you an honest read on what agentic commerce does and doesn't change for you.
Get an honest assessmentWhy B2B gets agentic commerce first
The consumer data says shoppers still prefer real storefronts at the moment of purchase. The business-to-business data points the other way, for a structural reason Marc Massar articulates well: the buying that automates first is the buying nobody enjoys. Supplier qualification, reordering, quote comparison, compliance checks — repetitive, policy-governed, high-volume purchasing that companies already staff procurement teams to do.
An AI procurement agent arrives with what the industry calls structured intent: exact SKUs or specs, quantities, price ceilings, delivery requirements. It does not need a persuasive landing page. It needs accurate, machine-readable answers — and a way to place an order that does not fail silently.
For NetSuite merchants this matters because so many of them are exactly the businesses on the receiving end: wholesale distributors, manufacturers, B2B brands. If that is you, the first agent that hits your storefront API will more likely be a customer's reordering bot than a ChatGPT shopper. The practical preparation — structured catalogs, reliable order APIs, B2B pricing logic exposed correctly — overlaps almost entirely with what a good B2B portal or headless build already requires.
How to get your stack agent-ready: five concrete moves
Agent-readiness is not a product you buy; it is a set of properties your commerce architecture either has or does not. The good news: every one of them is also just good engineering for your human storefront.
- Make the catalog machine-readable. Structured product data, clean APIs or feeds, accurate attributes. If your catalog only exists as rendered HTML, agents are scraping you at best and skipping you at worst. The read-path patterns in our NetSuite headless commerce architecture guide apply directly.
- Treat inventory and pricing accuracy as a contract. Real-time availability from the system of record, with caching that is fast but honest. Overselling to an agent generates orders you cannot fulfill — and agents remember.
- Keep the checkout and order flow merchant-owned. The Instant Checkout experiment settled this: do not outsource your transaction to a platform that taxes it. Expose discovery; own the order.
- Make order writes idempotent. Agents retry aggressively. External-ID deduplication on sales order creation is the difference between one order and three duplicates from one impatient bot.
- Pick your agent-facing surface deliberately. An ACP feed for consumer assistants, MCP tools for controlled agent access to NetSuite data, or structured endpoints for B2B customers' systems — each fits a different motion, and they share the same underlying API layer. That layer is what our ERP ecommerce integration service builds.
The horizon: invisible commerce
There is a more radical thesis worth knowing about, even if you should not architect for it yet. Fintech analyst Simon Taylor argues that the endgame is not a better checkout for agents but no checkout at all — "invisible commerce," where payments become event-driven and machine-native. An agent pays for parking the moment you park. A procurement agent pays a supplier the moment goods ship. Discovery collapses into structured registry lookups; pricing pages become machine-readable headers; new machine-payment protocols give agents their own way to pay.
Taylor's sharpest observation is that every level of the standard agentic-commerce ladder still assumes a visible purchase event — and that assumption may be the thing that expires. It is early, and the protocol landscape is unsettled. But notice what the invisible-commerce future and the merchant-owned-checkout present both assume: machine-readable catalogs, trustworthy real-time data, and a reliable way to create transactions in a system of record. Whichever version arrives, the preparation is the same.
Where this leaves NetSuite merchants
A channel, not a replatform. That is the honest summary of agentic commerce in 2026. The platform-owned checkout died on contact with real conversion data. What remains is a slower, more durable shift: AI assistants becoming a discovery surface for consumers, procurement agents becoming real API customers for B2B sellers, and the protocols underneath consolidating.
The businesses positioned for it are the ones whose commerce architecture is already clean — accurate system of record, machine-readable catalog, merchant-owned order flow, resilient APIs. If you have been investing in that direction for headless or composable reasons, agentic commerce costs you incremental work, not a rebuild. If your storefront is a tangle of manual syncs and stale caches, agents will not be your biggest problem — they will just be the newest customer your stack disappoints.
We will keep this guide updated as the protocols and the NetSuite integration evolve — including whatever Oracle announces at SuiteWorld 2026, where agentic commerce is a safe bet for keynote material.
Sources and references
- Walmart conversion data — Search Engine Land
- OpenAI's commerce strategy shift — TechCrunch and Adweek
- Agentic Commerce Protocol — OpenAI
- Five levels framework — via Simon Taylor; Stripe's agentic commerce guide is a useful general primer on the space
- Etsy's native ChatGPT app — TechCrunch
- "Agentic commerce is dead, long live agentic commerce" — Marc Massar
- The invisible commerce thesis — Simon Taylor
Frequently asked questions
What clients ask before signing
BrokenRubik
NetSuite Development Agency
Expert team specializing in NetSuite ERP, SuiteCommerce development, and enterprise integrations. Oracle NetSuite partner with 8+ years of experience delivering scalable solutions for mid-market and enterprise clients worldwide.
Related Articles
Composable Commerce: MACH Architecture & NetSuite
What composable commerce is, how MACH architecture works, and when best-of-breed beats monolithic. Plus how NetSuite fits as the ERP backbone.
NetSuite AI 2026: Native Features, MCP, and Custom Agents
NetSuite AI in 2026: Oracle's native tools (NetSuite Next, Ask Oracle, EPM agents), what you build with MCP, and where third-party AI fits. Use cases included.
NetSuite B2B Portal: SuiteCommerce vs Alternatives
Compare every NetSuite B2B portal option — SuiteCommerce, MyAccount, and third-party alternatives. Real costs, features, and which approach fits your use case.