ChatGPT
Perplexity
Claude
Gemini
Grok
Agentic commerce is the use of autonomous AI agents to search for products, compare options, and complete purchases on behalf of a human, often with minimal or no manual clicking through a checkout flow. Instead of a shopper browsing a storefront, an AI agent interprets a request such as "find me a waterproof jacket under $150 and order the best-reviewed one," then executes the research and transaction across merchant systems. The agent acts as an intermediary between intent and purchase, using structured product data, payment credentials, and merchant APIs rather than a graphical interface built for human eyes.
What Is Agentic Commerce?
Agentic commerce describes a shift from humans operating shopping interfaces to AI agents operating on a human's behalf, using natural-language instructions instead of clicks. Mastercard, Visa, OpenAI, and Google have each shipped protocols or products in 2025 and 2026 that let a chatbot or personal assistant search inventory, hold a shopper's preferences in memory, and settle payment through a tokenized credential. Salesforce reported in its 2026 Shopping Index that AI-agent-referred traffic to retail sites grew over 1,200% year over year, while IBM's Institute for Business Value has framed the category as "the next interface for commerce" since 2025.
The term sits one layer above "conversational commerce," which is chatting with a bot that answers questions. Agentic commerce requires the agent to take action: adding an item to a cart, choosing a shipping tier, and authorizing payment, typically without the merchant's normal web checkout page ever loading in a browser a human can see. Perplexity, ChatGPT, Gemini, and Amazon's Rufus have each added some version of buy-for-me functionality since 2024, with OpenAI's Instant Checkout and Perplexity's agentic shopping tool cited most often as the two most mature consumer-facing implementations.
| Term | What it means | Example |
|---|---|---|
| Conversational commerce | Chat interface answers questions, human still clicks to buy | A chatbot recommends three laptops, user clicks through to the store |
| Agentic commerce | AI agent completes the transaction with delegated authority | Agent compares laptops, selects one, and pays using a stored credential |
| Autonomous commerce | Agent initiates and completes purchases without a per-transaction prompt | A replenishment agent reorders detergent when stock runs low |
Why It Matters Now: A Short Timeline
Agentic commerce moved from concept to shipped infrastructure in under two years, driven by agents becoming reliable enough to hold multi-step context and by payment networks building the needed trust layer.
- 2023 to early 2024: Early "shopping copilot" chatbots (Amazon Rufus, Shopify Sidekick) answer product questions but still hand the user off to a normal checkout page.
- Mid 2024 to 2025: Model providers add tool-use and computer-use capabilities. Anthropic's Claude and OpenAI's GPT models gain the ability to click through real web UIs, letting early prototypes complete purchases end to end, though without a standardized trust layer.
- September 2025: Google publishes the Agent Payments Protocol (AP2), an open standard for how an agent proves it has a user's authorization before a payment executes, with early support from more than 60 partners.
- September to October 2025: OpenAI ships Instant Checkout inside ChatGPT with Etsy and Shopify merchants, alongside the Agentic Commerce Protocol built with Stripe. Visa and Mastercard separately announce Trusted Agent Protocol and Agent Pay.
- 2026: Salesforce, Adobe, and IBM publish research quantifying agent-driven retail traffic and forecasting double-digit shares of ecommerce revenue moving through agent-mediated transactions. Deloitte's 2026 holiday retail outlook gave agentic commerce its own line item in consumer spending forecasts for the first time.
How Agentic Commerce Works
An agentic commerce transaction runs through a sequence of steps that replace the human clicks of a normal checkout with machine-readable exchanges between the agent, the merchant, and a payment authorizer.
- Intent capture. The user gives the agent a goal in natural language, such as a product category, budget, and preferences.
- Discovery and comparison. The agent queries merchant catalogs, product feeds, or APIs, often via structured data formats merchants expose for agent consumption, and ranks candidates against the stated criteria.
- Decision. The agent selects a specific product and variant, applying any rules the user set in advance (price ceiling, preferred retailer).
- Authorization. The agent presents a cryptographically signed credential or mandate proving the user approved this purchase, the mechanism at the center of protocols like AP2 and Visa's Trusted Agent Protocol.
- Transaction execution. Payment is tokenized and settled through the card network, and the merchant's order system confirms fulfillment back to the agent.
- Confirmation and follow-up. The agent reports the completed order to the user and may handle downstream tasks like tracking or return initiation.
| Step | Human checkout | Agentic checkout |
|---|---|---|
| Product discovery | Manual browsing, search bar, filters | Agent queries structured product feeds |
| Comparison | User reads reviews, opens tabs | Agent parses reviews and specs programmatically |
| Cart and payment | User clicks add-to-cart, enters card | Agent submits signed payment mandate |
| Identity check | CAPTCHA, 2FA, login | Agent-specific authentication token |
| Confirmation | User sees order confirmation page | Agent relays summary to user via chat |
Agentic Commerce vs Traditional E-commerce
The two models differ most in who reads the product page and who makes the final call. Traditional ecommerce is built for a human eye. Agentic commerce is built for a machine reader that needs structured, verifiable facts before committing a payment on someone else's behalf.
| Dimension | Traditional E-commerce | Agentic Commerce |
|---|---|---|
| Primary interface | Website or app UI | Chat or API call, no browser session for the human |
| Who decides | Human shopper | AI agent, within user-set rules |
| What convinces the buyer | Visual merchandising, brand storytelling | Structured data, verified specs, review signals |
| Checkout friction | Forms, CAPTCHA, manual card entry | Signed mandate, tokenized credential |
| Speed | Minutes to hours across sessions | Seconds to a few minutes |
| Data format merchants must expose | Human-readable web pages | Machine-readable feeds (schema.org Product, APIs) |
| Trust mechanism | Site reputation, payment page padlock | Agent authentication protocol (AP2, ACP, Trusted Agent) |
Key Technologies Powering Agentic Commerce
Four categories of technology had to mature together before agentic commerce became viable at consumer scale: reasoning-capable LLM agents, standardized payment authorization protocols, machine-readable product data, and agent identity verification.
| Technology layer | What it solves | Examples |
|---|---|---|
| LLM agent reasoning | Holding a multi-step shopping goal across dozens of intermediate actions | Anthropic Claude, OpenAI GPT, Google Gemini with tool use |
| Payment authorization protocols | Proving an agent has real human consent for a specific purchase | Google AP2, OpenAI/Stripe Agentic Commerce Protocol, Visa Trusted Agent, Mastercard Agent Pay |
| Structured product data | Letting an agent read specs and pricing without a human-facing page | schema.org Product markup, merchant APIs, product feeds |
| Agent identity and reputation | Letting a merchant tell a legitimate agent apart from a scraper or fraud script | Signed tokens issued by the model provider or payment network |
Merchants without clean structured data are effectively invisible to a shopping agent, no matter how polished their storefront looks to a human visitor.
Use Cases Emerging in 2026
Agentic commerce is showing up first in categories with clear specifications and lower emotional stakes, where an agent can compare options confidently without needing to "feel" the product.
- Replenishment purchases. Household staples and consumables (detergent, pet food, printer ink) are the leading category, since the agent can rely on past purchase history instead of fresh discovery.
- Comparison-heavy categories. Electronics, appliances, and travel bookings suit agentic commerce because the decision hinges on specs a machine can rank (battery life, price per liter, direct flight vs. layover).
- In-chat checkout inside AI assistants. ChatGPT's Instant Checkout and Perplexity's shopping agent let a user complete a purchase inside the same conversation where they asked for a recommendation.
- B2B procurement agents. Repeat business purchasing (office supplies, materials reordering) is adopting agentic workflows faster than consumer retail in some sectors, since the purchase rules are already codified in vendor contracts.
- Merchant-side agent readiness. Brands are auditing product pages so an agent can extract accurate specs, pricing, and availability, a task that increasingly includes making video and image assets machine-legible.
How Brands Can Start Preparing
Getting ready for agentic commerce is less about building a new storefront and more about making existing product information legible to a non-human buyer.
- Publish clean structured data. Ensure every product has accurate schema.org Product markup and current pricing, since an agent that receives stale data will simply route the sale elsewhere.
- Register with emerging agent protocols. Merchants on Shopify, Etsy, or similar platforms should confirm integration with OpenAI's Agentic Commerce Protocol, Google's AP2, or the card-network equivalents as they expand.
- Keep product video and imagery accurate to spec. Agents cross-check visual claims against structured claims, so product video showing features not actually in the spec sheet creates a mismatch an agent can flag or distrust.
- Produce fast, spec-accurate product video at scale. Brands selling across many SKUs need product and demo videos that match structured data without a slow, manual cycle per listing. This is one honest example of where an AI video agent like Pexo fits in: brands feed Pexo a product description, image set, or URL, and it returns a finished video built around models such as Seedance 2.0 and Kling AI, keeping the visual asset in line with the same spec facts an agent reads in parallel. It supports agent-readiness, not a replacement for the payment infrastructure above.
- Monitor agent-referred traffic separately. Analytics that lump agent visits in with regular bot traffic will undercount a channel Salesforce and Adobe both flagged as growing fast in 2026.
For teams researching the broader AI-agent landscape, see what an agent-as-a-service model looks like and how autonomous AI video agents work. Brands evaluating video production for high-SKU catalogs can compare the best AI video agents, the best AI video generators for ecommerce, and options built for online stores.
Conclusion
Agentic commerce reframes checkout as a conversation between an AI agent and a merchant's systems rather than a human clicking through a webpage. The protocols enabling it, Google's AP2, OpenAI and Stripe's Agentic Commerce Protocol, Visa's Trusted Agent Protocol, and Mastercard's Agent Pay, all shipped or expanded within roughly a twelve-month window across 2025 and 2026, which is why the category moved from theory to measurable retail traffic so quickly. For brands, the practical task is making product data, pricing, and video content accurate and machine-readable enough that an agent can confidently transact on a shopper's behalf. Tools like pexo.ai fit into that second half, keeping visual content aligned with the same facts an agent is verifying, without pretending video production solves the payment infrastructure that agentic commerce actually depends on.
Frequently Asked Questions
What is agentic commerce in simple terms? Agentic commerce is when an AI agent searches, compares, and buys products on a person's behalf, using natural-language instructions instead of the person clicking through a website checkout themselves.
Is agentic commerce the same as conversational commerce? No. Conversational commerce is chatting with a bot that answers questions, but the human still clicks through to complete the purchase. Agentic commerce means the agent itself executes the transaction, including payment authorization.
What companies are behind agentic commerce protocols? Google built the Agent Payments Protocol (AP2), OpenAI and Stripe built the Agentic Commerce Protocol, and Visa and Mastercard built Trusted Agent Protocol and Agent Pay respectively, each addressing how a payment network verifies an AI agent's authority to transact.
How does an AI agent actually pay for something? The agent presents a signed authorization, often called a mandate or token, proving the human user approved a specific category, amount, or transaction, which the payment network verifies before settling the charge.
Is agentic commerce safe from fraud? The protocols are designed specifically to reduce fraud risk by requiring cryptographically verifiable consent for each transaction, but the category is new enough that dispute-resolution and chargeback norms are still being established across card networks.
Which retailers currently support agentic checkout? Etsy and Shopify merchants were among the first supported inside ChatGPT's Instant Checkout in late 2025, and Perplexity's shopping agent has expanded merchant partnerships through 2026, with more retailers integrating as the protocols standardize.
Does agentic commerce replace human shopping entirely? No. It is best suited to categories with clear specifications, like replenishment items and comparison-heavy electronics, while considered or emotionally driven purchases still tend to involve a human browsing directly.
What is a shopping agent's "mandate"? A mandate is a signed digital authorization that defines the scope of what an agent is allowed to purchase on a user's behalf, such as a spending limit, product category, or time window, used by protocols like Google's AP2.
How should a brand prepare its product pages for AI agents? Brands should publish accurate schema.org Product markup, keep pricing and inventory data current, and ensure visual assets like images and video accurately reflect the specs an agent will be reading, since mismatches can cause an agent to distrust the listing.
Does agentic commerce affect B2B purchasing too? Yes, and in some sectors it is moving faster than consumer retail, since repeat B2B purchases like office supplies or materials reordering already follow codified rules an agent can execute against.
Where does AI video fit into agentic commerce? Video itself is not part of the payment or authorization layer, but brands still need product video that stays accurate to the same specs an agent cross-checks, which is why AI video agents such as pexo.ai are increasingly used to keep visual assets in sync with structured product data across large catalogs.
