Agentic commerce is the practice of letting an AI agent research, compare, decide and complete a purchase on a person's behalf, with the human setting the parameters and approving the spend rather than clicking through the steps. It replaces the familiar browse, filter and checkout flow with a single instruction in plain language. The inflection point arrived on 29 September 2025, when OpenAI added Instant Checkout to ChatGPT, letting US users buy without leaving the chat. Since then Google, Visa, Mastercard, PayPal and Stripe have all shipped infrastructure to make agent-driven buying work at scale. For brands the question stops being whether you rank or even whether you are recommended, and becomes whether you are the product the agent actually picks.
What is agentic commerce?
Agentic commerce describes shopping where an autonomous AI agent does the work a person used to do by hand: interpreting the need, finding candidate products, comparing options, checking availability and price, applying discounts and, with permission, paying. The classic example is a single instruction like "find me trail running shoes under $150, in my size, delivered by Friday." The agent turns that into searches, narrows a shortlist, and presents a choice or completes the order outright. According to IBM, the defining trait is intent-driven, conversational transactions rather than menu navigation. Salesforce and Deloitte frame the same shift as moving the buyer's effort from the person to the software. The agentic commerce meaning that matters here is practical: a layer of software now sits between the shopper and the storefront, and that layer makes the decisions. It is a direct application of what is agentic ai to the specific job of buying things.
How agentic commerce works
The flow runs through five rough stages: intent, research, comparison, cart and checkout, with a fulfilment and post-purchase tail. The person states a goal and any constraints, the agent retrieves candidate products from search, catalogs and merchant feeds, then it compares them against the stated criteria and its own read of quality signals. AI shopping agents do not weigh hero imagery or brand mood the way a human browsing a site might; they parse structured attributes, specifications, availability and price. Crucially, most current implementations keep a human in the loop, asking for explicit confirmation before money moves, which is both a trust feature and, for now, a regulatory comfort. The mechanics of that selection step are worth understanding in detail, because they decide which products even reach the shortlist, a topic covered in how AI shopping agents choose products. The agent is, in effect, a buyer that reads machine-readable data far better than it reads marketing.
The 2026 landscape: where you can already buy through an agent
The most visible entry point is OpenAI's Instant Checkout inside ChatGPT, launched on 29 September 2025 for US Plus, Pro and Free users. It began with US Etsy sellers and is rolling out to more than a million Shopify merchants, including Glossier, SKIMS, Spanx and Vuori, as reported by CNBC and Digital Commerce 360. The first version handled single items, with multi-item carts and more regions on the roadmap, per PYMNTS. Google has moved on a parallel track: its Universal Commerce Protocol launched at NRF 2026, and a Universal Cart shown at Google I/O 2026 is designed to follow a shopper across Search, Gemini, YouTube and Gmail, as covered by TechCrunch. The practical upshot is that the storefront is no longer the only place a sale closes. We unpacked the merchant side of the ChatGPT shift in ChatGPT Shopping is here, and the broader trajectory in the future of AI search.
The payment rails behind it
Agent-driven buying needs plumbing that lets software pay safely without handing it a raw card number, and 2025 and 2026 saw the major players ship competing but increasingly interoperable standards. The Agentic Commerce Protocol, co-developed by OpenAI and Stripe and open-sourced, lets Stripe merchants enable agentic payments in roughly one line of code and uses scoped, encrypted payment tokens authorised for a specific amount and merchant, with user confirmation at each step, per the Stripe newsroom. Google's Agent Payments Protocol (AP2) launched on 16 September 2025 with more than 60 partners including Mastercard, PayPal and Amex, and uses three signed mandates, Intent, Cart and Payment, to create a verifiable audit trail. Card networks layered their own controls on top: Visa's Intelligent Commerce Connect offers one integration spanning ACP, UCP and other protocols, with Visa predicting millions of consumers buying through agents by the 2026 holidays per its investor release, while Mastercard's Agent Pay uses Agentic Tokens scoped to a specific agent, merchant and consent policy. The standards are converging fast, and the detail is worth its own read in agentic commerce protocols.
Want to see this in action?
See how every major AI model talks about your brand. Free to start.
| Protocol | Backed by | Launched | What it covers |
|---|---|---|---|
| ACP | OpenAI + Stripe | Sept 2025 | Discovery and checkout in ChatGPT; open-sourced |
| AP2 | Google + 60+ partners | Sept 2025 | Agent payments via signed Intent, Cart and Payment mandates |
| UCP | NRF 2026 | Full journey, discovery to post-purchase; interoperable with AP2, A2A, MCP | |
| Visa Intelligent Commerce Connect | Visa | 2026 | One integration across ACP, UCP, Trusted Agent and Machine Payments |
| Mastercard Agent Pay | Mastercard | Apr 2025 | Agentic Tokens scoped to agent, merchant and consent |
How big is it? The adoption data
The growth figures are striking, though most originate from vendor and analyst research and are best read as "according to X" rather than settled fact. Adobe Analytics reported that AI-driven traffic to US retail sites rose 805% year over year on Black Friday 2025 and 693% across the full holiday season. Salesforce, which tracks roughly 1.5 billion shoppers, attributed $14.2 billion in Black Friday online sales to generative AI and agents, within a wider AI-influenced Cyber Week of around $67 billion. Looking forward, Salesforce and Deloitte project that roughly a quarter of online retail could be agent-enabled by 2030, with US agentic commerce estimated in the $300 billion to $500 billion range. The caveat is trust: a checkout.com survey found only 14% of consumers currently trust AI to place orders for them, rising to 29% in Gen Z and 30% in millennials, and 83% reported privacy concerns. The traffic is real and accelerating; the autonomous purchasing is earlier than the headlines suggest.
What agentic commerce means for brands: becoming the product the agent picks
The shift that matters for brands is that the agent compresses a category down to a shortlist of three or four products, and you are either inside that consideration set or invisible. Agents scan for explicit, structured, machine-readable data, the attributes, specs, materials, use cases, availability and price, not the vibe of a hero image. As ML6 and Retail Dive put it, specificity wins: "single-origin Colombian, USDA organic, ideal for pour-over" gives an agent far more to match against than "medium roast, caramel notes." That reframes visibility from a marketing outcome into a product and integration capability, where clean catalogs, enriched metadata and structured data decide whether an agent can understand a SKU at all, a point commercetools makes plainly. The same recommendation dynamics that already govern AI search apply here, as we covered in how AI models choose which brands to recommend and, for retail specifically, in AI visibility for e-commerce brands. AI search was about being recommended; agentic commerce extends that to being bought, and both are downstream of whether the model knows, trusts and names your brand. You cannot improve what you do not measure, which is why tracking your AI recommendation and share of voice now, before agent purchasing scales, is the unglamorous prerequisite that Honeyb exists to handle.
What to do now
Start by making your product data legible to a machine. Audit your catalog for complete, accurate, structured attributes, and prioritise the fields an agent uses to match intent: size, material, use case, availability, price and clear specifications. Check whether your platform supports the emerging rails, since Shopify and Stripe merchants already have a path into ChatGPT's Instant Checkout, and the protocols are converging rather than fragmenting. Strengthen the third-party signals agents lean on, including reviews and authoritative comparisons, because those feed the shortlist as much as your own pages do. Most importantly, begin measuring whether AI engines name and recommend your brand today, so you have a baseline before agent-driven buying moves from a holiday-season novelty to a default. The brands that treat this as a data and visibility problem now, rather than a marketing campaign later, are the ones that will keep landing in the agent's shortlist.
