Agentic commerce is the fastest-moving shift in online retail since the launch of mobile shopping. In just over a year, AI agents have moved from answering product questions to actually completing purchases — comparing prices across merchants, filling carts, authenticating payments, and confirming delivery, all without a human clicking "Buy Now."
Agentic Commerce Definition
Agentic commerce is the practice of delegating shopping and purchasing tasks to autonomous AI agents that can browse, evaluate, transact, and follow up on behalf of a human user or business.
The keyword is autonomous. A traditional chatbot that recommends a product still requires you to click through, log in, fill in card details, and check out. An agentic system completes the entire flow itself. You say, "Find me running shoes under $120 with a wide toe box, and reorder my usual coffee," and the agent does it — discovery, decision, payment, and post-purchase tracking included.
Three properties separate agentic commerce from earlier AI shopping experiences:
- Goal-directed reasoning. Agents pursue an objective ("buy a birthday gift for my sister") rather than respond to a single query.
- Tool use. Agents call APIs, navigate websites, query inventory, and authorize payments using infrastructure built specifically for them.
- Authorized autonomy. Agents act within spending limits, merchant allowlists, and policy guardrails set by the user or the platform — they don't need a human in the loop for every step.
How Agentic Commerce Actually Works: The Stack in 2026
Most agentic checkout flows in 2026 share four layers:
The agent layer. This is the LLM-powered system the user interacts with — ChatGPT, Claude, Perplexity, Google's Gemini, or a vertical agent embedded in a retailer's app. The agent interprets intent, plans steps, and decides what to buy.
The discovery layer. Agents need machine-readable product data, not screenshots of marketing pages. Standards like Google's Merchant Center feed, Shopify's Storefront API, and emerging agent-friendly product feeds (often built on schema.org/Product extensions) let agents enumerate SKUs, prices, availability, and shipping rules.
The payments layer. This is where the biggest 2025–2026 changes happened. Stripe launched agent-aware payment APIs and shared credentials, Visa rolled out Visa Intelligent Commerce, Mastercard launched Agent Pay, and PayPal opened its Agent Toolkit. These products give agents a way to pay without storing or seeing raw card numbers — using tokenized credentials, scoped authorizations, and cryptographic proof of user consent.
The trust layer. Merchants need to know whether the entity hitting their checkout is a real customer's agent or a scraper. Standards like the Agentic Commerce Protocol (ACP), jointly published by OpenAI and Stripe, and Google's Agent Payments Protocol (AP2) give merchants verifiable signals about the agent, the user behind it, and the limits of the authorization.
When all four layers work together, the user experience collapses from "open ten tabs, compare, decide, check out" to a single sentence — and the merchant gets a verified, paying customer without ever rendering a checkout page.
Why Agentic Commerce Is Happening Now
Three things had to be true for agentic commerce to take off, and all three landed inside an 18-month window.
LLMs got reliable enough to act, not just describe. Frontier models from late 2024 onward — GPT-4o and successors, Claude 3.5/4 Sonnet, Gemini 2.0 — were the first generation that could plan multi-step shopping tasks and recover from errors well enough that consumers would trust them with money.
Distribution arrived. OpenAI launched Operator in early 2025 and rolled checkout directly into ChatGPT later that year. Perplexity shipped Buy with Pro. Google embedded agentic shopping into Gemini and Search. Suddenly hundreds of millions of users had an agent in front of them, asking what they wanted to buy.
Payments unlocked. Without agent-specific payment rails, every transaction either had to fall back to copying a card number into a checkout form (clumsy, insecure) or to a wallet that only worked on one merchant. The 2025 wave of agent wallets and tokenized agent credentials — covered in depth in our agent payments deep dive — was the missing piece.
Who's Building the Agentic Commerce Economy
The ecosystem is broader than most operators realize. A simplified map:
Agent platforms — OpenAI (ChatGPT, Operator), Anthropic (Claude with computer use and agent skills), Perplexity (Comet, Pro Buy), Google (Gemini, AI Mode in Search), Microsoft (Copilot), and a long tail of vertical agents.
Payment & wallet infrastructure — Stripe (Agent Toolkit, shared credentials, ACP), Visa (Intelligent Commerce, Visa Trusted Agent Protocol), Mastercard (Agent Pay), American Express (agent commerce pilots), PayPal (Agent Toolkit), and emerging agent-native wallets like Stripe Link for agents and Coinbase's x402 for crypto-native machine payments.
Commerce platforms — Shopify (Universal Cart, agent-friendly storefronts), Salesforce (Agentforce Commerce), Adobe (AEM agent integrations), and BigCommerce.
Protocols and trust — Anthropic's Model Context Protocol (MCP) for tool access, OpenAI/Stripe's Agentic Commerce Protocol, Google's Agent Payments Protocol, and Cloudflare's agent verification layer.
Vertical agentic-commerce startups — companies building agent-first storefronts, agent-readable product graphs, attribution and analytics for agent traffic, and merchant-side tools to convert agent visitors. (StartShop is one of them — purpose-built to make Shopify and headless storefronts win in the agentic channel.)
What Agentic Commerce Means for Merchants
The temptation is to treat agentic commerce as "another channel like mobile or social." That framing misses the point. Agentic traffic behaves differently in three ways that matter for revenue:
Discovery is invisible. When a shopper asks ChatGPT for "the best noise-cancelling headphones under $300," your product either makes the agent's shortlist or it doesn't. There's no scroll, no second-page result. Agent SEO — sometimes called AEO (Answer Engine Optimization) or GEO (Generative Engine Optimization) — is a real discipline now.
Conversion is binary. A traditional shopper bounces 70% of the time. An agent that has decided to buy converts at near 100%, but only if your checkout, inventory, and shipping APIs respond cleanly to its calls. Broken structured data, unstable APIs, or a checkout that demands a CAPTCHA will silently knock you out of the consideration set.
Loyalty is portable. When the agent is the relationship, the user's loyalty is to the agent, not the merchant. This is good news (the agent will reorder from you if you delivered) and bad news (the agent will switch you out if a competitor is cheaper or faster). Merchants that win in agentic commerce treat agents as a key customer segment with their own SLAs, pricing logic, and post-purchase handling.
The merchants taking agentic commerce seriously in 2026 are doing four things:
- Publishing structured, agent-readable product feeds with real-time inventory and shipping data
- Accepting agent payment credentials (Visa Intelligent Commerce, Mastercard Agent Pay, Stripe shared credentials) at checkout
- Tracking and attributing agent traffic separately from human traffic in their analytics
- Tuning content, descriptions, and reviews so they rank well inside agent answers, not just Google search
The Risks: Fraud, Bias, and the Merchant-of-Record Problem
Agentic commerce is not a free lunch. Three open problems are shaping the regulatory and engineering conversation in 2026.
Fraud and impersonation. If an agent can pay, an agent can be tricked into paying. Prompt injection attacks — where a malicious product page convinces an agent to buy something the user didn't ask for, or to leak the user's credentials — are an active research area. The trust protocols (ACP, AP2, Visa's Trusted Agent Protocol) exist specifically to mitigate this.
Bias and steering. When an agent picks one merchant over another, that's a recommendation with real economic consequences. Regulators in the EU and US are beginning to ask how agents rank merchants, whether platforms can pay for placement, and how to disclose those relationships.
Merchant-of-record and chargebacks. Who is liable when an agent buys the wrong thing? The user authorized the agent, but didn't authorize that specific purchase. Card networks updated their dispute frameworks in late 2025 to handle agent-initiated transactions, but the case law is still being written.
Where Agentic Commerce Goes Next
By the end of 2026, three things are likely:
- Agent-initiated transactions will cross 5–10% of US e-commerce GMV in select categories (electronics, household reorder, travel), per estimates from analysts at Forrester and Gartner.
- Every major card network and PSP will have a production agent product, with the early movers (Visa, Mastercard, Stripe, PayPal) holding a real distribution lead.
- A standardization moment will arrive — likely a convergence between ACP, AP2, and the W3C Web Payments group — that lets a single agent transact across networks without bespoke integration.
If you're a merchant, a developer, or a payments operator, the work to do today is concrete: get your product data agent-readable, accept at least one agent payment credential, and start measuring agent traffic as a first-class channel.
Agentic commerce isn't coming. It's here, and the merchants that treated it as a 2027 problem are already losing share to the ones that treated it as a 2025 priority.
StartShop is the agentic commerce platform that helps Shopify and headless merchants get discovered, transacted with, and reordered by AI agents.