ChatGPT What Is Agentic Commerce? Definition & Examples (2026)
Summary
Key takeaways
- Agentic commerce is a transaction model in which AI agents can discover products, compare options, request quotes, build carts, place orders, and coordinate post-purchase tasks on behalf of buyers or merchants.
- The defining difference between conversational and agentic commerce is delegated authority: the agent is permitted to take commercial actions rather than only provide recommendations.
- In B2B commerce, agents must operate within contract pricing, approved suppliers, account-specific catalogs, spending limits, payment terms, and internal approval rules.
- Buyer agents and merchant agents serve different roles: buyer agents represent procurement teams, while merchant agents help sellers qualify demand, prepare quotes, enforce pricing, and coordinate orders.
- The ecommerce platform cannot be treated as the only source of truth when pricing, inventory, credit, fulfillment, and account rules live in ERP, PIM, CRM, OMS, WMS, or CPQ systems.
- Machine-readable product, pricing, availability, shipping, and policy data are foundational because agents interact primarily through structured data and APIs rather than visual storefronts.
- Agentic commerce protocols such as ACP and UCP provide standardized ways for agents and merchants to exchange commerce data and coordinate transactions.
- Payment authorization is a separate layer from product discovery and checkout orchestration, requiring identity, permissions, tokenized credentials, spending controls, and auditable mandates.
- In B2B, the near-term opportunity is more likely to be agent-assisted discovery, reordering, quoting, and availability checking than unrestricted autonomous procurement.
- Most merchants should prioritize data quality, API readiness, integrations, identity, permissions, and governance before investing heavily in specific agentic commerce protocols.
When this applies
This applies when a retailer, manufacturer, distributor, wholesaler, marketplace, or B2B seller wants AI to move beyond product recommendations into real commercial workflows. It is particularly relevant for businesses preparing for AI shopping assistants, procurement agents, automated reordering, quote generation, or machine-led product discovery. The model becomes especially valuable when buyers repeatedly perform predictable tasks that can be expressed through structured data and governed APIs while still respecting existing pricing, inventory, approval, and payment controls.
When this does not apply
This does not apply when a business simply wants a chatbot that answers questions or recommends products while the customer still performs every transactional step manually. It is also premature when product data is unreliable, account pricing cannot be accessed programmatically, integrations are unstable, or important commercial rules exist only in employee knowledge. Agentic commerce should not be treated as permission to let an LLM bypass established financial, procurement, security, or operational controls.
Checklist
- Define which commercial actions an AI agent should be allowed to perform.
- Separate discovery, recommendation, quoting, ordering, payment, and post-purchase workflows.
- Identify the authoritative system for product, price, inventory, customer, and order data.
- Make product information structured and machine-readable.
- Expose reliable APIs for catalog, availability, pricing, cart, and order operations.
- Map customer-specific pricing and contract rules to authenticated accounts.
- Define buyer roles, permissions, spending limits, and approval thresholds.
- Keep high-value or exceptional transactions behind human approval where required.
- Connect ERP, PIM, CRM, OMS, WMS, and CPQ systems where they hold authoritative data.
- Evaluate relevant protocols such as ACP, UCP, and supporting agent standards.
- Separate commerce orchestration from payment authorization and credential management.
- Add logging, auditability, retry handling, and escalation for agent actions.
- Start with a narrow workflow such as reorder, quote preparation, or availability lookup.
- Measure accuracy, cycle time, manual intervention, and commercial outcomes.
- Expand agent autonomy only after the underlying workflow proves reliable and governable.
Common pitfalls
- Confusing a conversational chatbot with an agentic commerce system.
- Allowing an agent to act without clearly defined delegated authority.
- Assuming the ecommerce platform contains all data required to complete a B2B transaction.
- Exposing catalog prices when actual customers buy under negotiated contract pricing.
- Building agent workflows on inaccurate inventory or product data.
- Treating protocols as a substitute for reliable backend architecture.
- Connecting agents directly to transactional systems without permissions and approval controls.
- Automating high-value procurement before proving simpler workflows such as reorders or availability checks.
- Optimizing only for one AI ecosystem while standards and channels continue to evolve.
- Implementing agentic checkout before fixing data quality, integrations, identity, and governance.
What is agentic commerce? Agentic commerce is buying and selling done by AI agents on behalf of people or companies. The agent finds products, compares options, builds a cart, and, with permission, completes the purchase. The person or the company sets the goal, the budget, and the rules. The agent does the work. In 2026, most agentic commerce is “discover with an agent, buy on the merchant site”. Fully delegated purchases are live in narrow use cases and growing.
Agentic commerce in numbers, September 2026
- AI-referred traffic to US retail sites was up 62% year over year in July 2026 and up 1,219% since October 2024 (Adobe Analytics).
- AI-referred shoppers generate 53% more revenue per visit than other shoppers (Adobe Analytics, July 2026).
- In March 2025, AI-referred visits converted 38% worse than other traffic. In March 2026, they converted 42% better (Adobe Analytics).
- Gartner expects AI agents to intermediate 90% of B2B buying by 2028, worth more than 15 trillion US dollars.
- 25% of the content on US retailer home pages and 34% of product-page content cannot be read by AI models (Adobe, April 2026).
Agentic commerce definition
Elogic Commerce definition. Agentic commerce is a form of commerce in which an AI agent, acting under a mandate from a buyer or a seller, performs one or more steps of a transaction, from product discovery to payment and post-purchase service, without a human doing each step by hand.
Three conditions must be true. If one is missing, it is not agentic commerce. It is AI-assisted ecommerce.
- The agent acts. It does not only answer. It searches a catalog, filters options, builds a cart, requests a quote, or completes a checkout.
- The agent works under a mandate. A person or a company gives it a goal, a budget, limits, and permission. The mandate can be small (“find three options under 200 euros”) or large (“reorder packaging film when stock falls below two weeks of use”).
- The agent uses live commerce data and systems. It reads real prices, real stock, and real delivery options. It connects to real checkout, payment, and order systems through APIs and protocols.
A simple test shows the difference. A chatbot can explain a returns policy. An agent can apply that policy to the order, start the return, update the commerce system, trigger the refund, and confirm that the work is done.
The word “agentic” comes from agentic AI: AI systems that plan, use tools, and take actions to reach a goal. Agentic commerce applies agentic AI to the transaction itself.
Related terms
You will see several names for the same idea. Agentic AI commerce, agentic ecommerce, and agent-based commerce mean the same thing as agentic commerce. Agent-assisted commerce is the lower part of the scale, where the human approves the final action. Programmatic commerce means rule-based transactions between systems, often agent to agent. Zero-click commerce is a marketing term for a purchase that completes without a visit to the merchant site. AI commerce is any use of AI in digital commerce, including personalization and search.
Two types of agents exist:
- Shopper agents (buyer side). They work for the customer, a consumer or a business buyer. Examples: an assistant in ChatGPT, Gemini, Claude, or Perplexity; a procurement agent inside an ERP; a brand’s own agent on its website.
- Merchant agents (seller side). They work for the company that sells: catalog search, pricing, inventory, quotes, order operations, and customer service.
What agentic commerce is not
Agentic commerce is often confused with four older ideas. The test is the same each time: does the AI take the action, or does a human still do it?
| Term | What it is | Why it is different |
|---|---|---|
| Chatbot | Answers questions in a chat window | It does not take actions in commerce systems |
| Conversational commerce | Selling through chat, voice, or messaging apps, often with a human agent or a scripted bot | The interface is conversational, but a human still does the steps |
| AI personalization | Recommendation engines, ranking, and dynamic content | The AI shapes what a human sees. It does not act for the human |
| Headless or composable commerce | An architecture that separates the front end from commerce services through APIs | It is a foundation for agentic commerce, not the same thing |
| Marketplace automation | Rule-based repricing, feed syncing, and auto-reorder | Rules do not plan or adapt to a goal. Agents do |
| Agent-assisted commerce | The agent researches, compares, and prepares a cart or a shortlist. The human approves the final action | It is the first two levels of agentic commerce, not a separate model. See the five levels below |
How does agentic commerce work?
An agentic transaction runs in six steps, from intent to post-purchase. Each step needs specific data, permissions, and systems from the merchant. If one step is missing, the agent stops or drops the merchant from the shortlist.
Take one instruction: “Find a business-class flight to New York under 3,000 euros that lands before noon, show me the best three, and book my choice.” The agent turns the sentence into criteria, searches live inventory, ranks the options by price and fit, asks for approval, and completes the booking. The interface moves from pages and forms to intent and delegation.
| Step | What the agent does | What the merchant must provide |
|---|---|---|
| 1. Intent | The person states a goal and limits. The agent turns them into a structured task. | Nothing yet. Clear product naming and categories help the agent map the intent to your catalog. |
| 2. Discovery | The agent searches catalogs, feeds, and the web. It reads product data, prices, stock, shipping, and reviews. | A machine-readable catalog: schema.org Product and Offer data, product feeds, complete specifications, an agent-accessible API or MCP server, and no blocks on the AI crawlers you want to serve. |
| 3. Evaluation | The agent compares options against the mandate: price, fit, delivery, returns, and trust. | Complete and accurate attributes. A missing attribute means your product is filtered out, not shown as “unknown”. |
| 4. Cart or configuration | The agent builds a cart, a bundle, or a quote request. In B2B it applies customer-specific pricing and contract terms. | Cart and pricing APIs that return the same numbers a human would see, in real time. |
| 5. Authorization | The person approves the purchase, or the agent proves it holds a valid mandate. Payment credentials are tokenized. The agent never sees raw card data. | Support for agent-aware payment tokens from your PSP or card network, and fraud rules that can tell a verified agent from a bot. |
| 6. Checkout and post-purchase | The agent completes checkout on the merchant surface or through a protocol. It then tracks the order, returns, and reorders. | Reliable order APIs, order status webhooks, and a returns process that an agent can call. |
The agentic commerce stack
The six steps run on four layers. Almost every protocol announced in 2025 and 2026 fits into one of them.
| Layer | What it does | Main standards and products in 2026 |
|---|---|---|
| Communication | Lets an agent read data and call tools on a site, a service, or another agent | MCP, the Model Context Protocol (started by Anthropic, now under the Linux Foundation); A2A, Agent2Agent (Google) |
| Commerce journey | Standardizes discovery, cart, checkout, and order management between an agent and a merchant | UCP, the Universal Commerce Protocol (Google with Shopify, Etsy, Wayfair, Target, and Walmart): discovery, cart, checkout, orders, and post-purchase, designed to work with A2A, AP2, and MCP. ACP, the Agentic Commerce Protocol (created by Stripe, OpenAI, and Meta; current specification April 2026): checkout sessions, cart and product feed, delegated payment, delegated authentication, orders, and webhooks |
| Authorization and trust | Proves that a human authorized the spend and that the agent is legitimate | AP2, the Agent Payments Protocol (started at Google, now governed by the FIDO Alliance); Visa Trusted Agent Protocol and Visa Intelligent Commerce; Mastercard Agent Pay |
| Settlement | Moves the money | Card networks, Stripe, Adyen Agentic, PayPal; machine-to-machine rails such as x402 (Coinbase) and MPP (Stripe) |
The layers compose. One purchase can use MCP to read a catalog, UCP to build the cart, AP2 to prove the mandate, and a Visa or Mastercard agent token to pay. Merchants will need to support more than one standard. For a full breakdown of the payment and authorization layers, read our guide to agentic payments.
The five levels of agentic commerce
Agentic commerce has five levels of autonomy, from agent-assisted discovery to agent-to-agent trade. Elogic Commerce uses this model, similar to the levels used for self-driving cars, to describe how much of a transaction an agent performs. The levels help a team set the right scope for a pilot and the right controls for each step.
| Level | Name | What the agent does | Who approves | 2026 example | Status, September 2026 |
|---|---|---|---|---|---|
| 0 | Manual commerce | AI ranks or recommends only. The human searches, compares, and buys. | Human | A standard ecommerce site with a recommendation engine | Still the majority of transactions |
| 1 | Agent-assisted discovery | The agent finds and compares options and explains them. The human clicks through and buys on the merchant site. | Human | ChatGPT shopping answers with retailer apps; Google AI Mode; Perplexity; Amazon Rufus | Mainstream. The fastest-growing traffic channel to retail sites |
| 2 | Agent-built cart with handoff | The agent builds the cart or the configuration and hands it to the merchant’s checkout. | Human, at checkout | Google Universal Cart; Shopify Agentic Storefronts; the Claude Commerce Agents shopper agent (Shopify, Priceline) | Rolling out. The main pattern for the 2026 holiday season |
| 3 | Delegated purchase within a mandate | The person sets intent, budget, and rules once. The agent completes the checkout and reports back. | Agent, inside the mandate. Human reviews exceptions | Visa tokenized credentials for agent-initiated checkout with spend caps; Mastercard Agent Pay, live in Singapore and Malaysia; AP2 mandates | Live in narrow use cases and pilots |
| 4 | Autonomous replenishment and procurement | The agent monitors usage, stock, or contracts and reorders inside policy. | Agent, inside policy. Human audits | B2B replenishment agents inside ERP and procurement systems; tail-spend purchasing | Early production in B2B |
| 5 | Agent-to-agent commerce | Buyer agents and merchant agents negotiate price, terms, and delivery, then settle. | Agent, inside policy | Mastercard Agent Pay for Machines; x402; A2A negotiation pilots | Experimental |
Where the market is now: Level 1 is mainstream. Level 2 is rolling out on the largest platforms. Level 3 is live in narrow, controlled cases. Level 4 exists in B2B replenishment. Level 5 is experimental. The most common mistake is to plan for Level 3 while the catalog is not ready for Level 1.
Agentic commerce vs traditional ecommerce
The difference between agentic commerce and traditional ecommerce is who does the work and where the decision is made. In traditional ecommerce a human navigates your site. In agentic commerce an agent reads your data and may never show your site at all.
| Dimension | Traditional ecommerce | Agentic commerce |
|---|---|---|
| Who does the work | The human searches, compares, and buys | The agent does the steps. The human sets the mandate |
| Main interface | Website, app, marketplace listing | Assistant, agent, or API. Your site may never be seen |
| Discovery driver | Search rankings, ads, merchandising | Machine-readable data, protocol support, trust signals, and how AI systems describe you |
| Unit of competition | The page and the funnel | The product attribute, the offer, and the answer |
| Conversion event | Add to cart and checkout on your site | Agent selection, then checkout on your surface or through a protocol |
| Cost of bad data | Some lost conversions | Removal from the shortlist |
| Brand role | Persuasion on the page | Trust, completeness, and consistency in every place the agent reads |
| Fraud and trust | Bot detection and 3-D Secure | Agent verification, mandates, tokens, and consent proof |
| Key metrics | Traffic, conversion rate, average order value | AI-referred sessions, agent inclusion rate, agent conversion rate, revenue per agent visit, catalog readability |
Agentic commerce examples in 2026
The clearest examples of agentic commerce in 2026 are AI shopping assistants that discover products and hand off carts, merchant-owned agents, and B2B replenishment agents. Each example below is marked with its level.
B2C examples
- Product discovery in ChatGPT (Level 1 to Level 3). Shoppers ask for options and get product results. Purchases move to the retailer’s app or website. OpenAI wound down its in-chat Instant Checkout in March 2026 after low merchant adoption and weak conversion. Since June 2026, tokenized Visa credentials support agent-initiated checkout inside OpenAI products, with spend limits set by the user.
- Google Universal Cart (Level 2). Google’s Universal Commerce Protocol powers a cart that works across Search AI Mode, Gemini, YouTube, and Gmail, with checkout at the merchant.
- Shopify Agentic Storefronts (Level 2). Shopify exposes catalog and cart to agents so that the merchant stays the point of sale.
- Claude Commerce Agents (Level 2 shopper agent; merchant agent for operations). Anthropic published an open blueprint on 2 September 2026 for shopper agents and merchant agents that run on a merchant’s own channels. Visa, Mastercard, Accenture, and Shopify are launch partners. Named early users include Shopify and Priceline. The blueprint does not include checkout or payments. Those stay in the merchant’s stack. Read our analysis of Claude Commerce Agents.
- Amazon (Level 1 to Level 3). Rufus answers product questions across Amazon. The Buy for Me feature, tested from 2025, lets Amazon’s agent buy from third-party brand sites for the customer.
B2B examples
- Replenishment agents (Level 4). An agent watches consumption or stock levels and reorders from approved suppliers inside a budget and an approval rule.
- Procurement research agents (Level 1). An agent shortlists suppliers, collects specifications and compliance documents, and prepares a comparison for the buyer.
- Quote-to-order agents (Level 2, seller side). A merchant agent turns a natural-language request into a configured quote with customer-specific pricing and lead times.
- Merchant operations agents (seller-side automation). Anthropic’s merchant agent design answers questions about sales, flags stock that will run out before a promotion, and proposes prices and campaigns.
A lesson from 2025 to 2026
In-chat checkout came before the data was ready. Walmart reported that checkout inside ChatGPT converted at about one third of the rate of its own site, and product data accuracy was a main cause. Forrester counted roughly 30 Shopify merchants live on Instant Checkout by February 2026. The pattern that works today is “discover with the agent, buy on a surface with accurate live data”.
Which categories feel agentic commerce first?
Categories with repeat purchases, structured attributes, and clear decision rules move first. Categories where taste, emotion, or negotiation decide the purchase move last. This holds in B2C and in B2B.
| Segment | Category | Exposure | Why | First move for the seller |
|---|---|---|---|---|
| B2C | Grocery and household replenishment | High | Frequent, rule-based, low risk | Substitution rules, recurring baskets, delivery windows as data |
| B2C | Beauty and personal care | High | Routines, ingredients, and reviews are comparable | Structured ingredients, concerns, and replenishment triggers |
| B2C | Consumer electronics | High | Specifications, compatibility, and price drive the choice | Complete specifications, warranty and compatibility data |
| B2C | Fashion | Medium-high | Discovery is easy, fit and taste are hard | Size data, fit guidance, clear returns |
| B2C | Furniture and home | Medium | High consideration, delivery constraints | Dimensions, materials, delivery rules, assembly data |
| B2C | Luxury | Low to medium | Brand, scarcity, and service matter more than comparison | Protect the experience; use agents for research and after-sales |
| B2B | MRO, consumables, and packaging | High | Known SKUs, contract prices, reorder cycles | Replenishment agents on customer-specific price lists |
| B2B | Chemicals and industrial components | High | Specification-driven, compliance documents required | Machine-readable specifications, safety data, and certificates |
| B2B | Pharma and healthcare supply | Medium-high | Repeat orders, strict compliance and licensing | Licence checks and audit trails as API calls |
| B2B | Capital equipment and configured products | Low to medium | Negotiated, multi-stakeholder, long cycles | Quote-to-order agents that prepare, not decide |
The state of agentic commerce in September 2026
Agentic commerce is small in absolute terms and the fastest-growing commerce channel in relative terms. The table collects the most-cited figures with their sources and dates. Elogic Commerce updates it each quarter.
| Metric | Value | Source and date |
|---|---|---|
| AI-referred traffic to US retail sites, year over year | +62% (July 2026); +393% (Q1 2026); +693% (holiday 2025) | Adobe Analytics, via Digital Commerce 360, August 2026 and April 2026 |
| Growth in AI-referred retail traffic since October 2024 | +1,219% | Adobe Analytics, July 2026 |
| Revenue per visit, AI-referred vs other traffic | +53% | Adobe Analytics, July 2026 |
| Conversion rate, AI-referred vs other traffic | 38% worse in March 2025; 42% better in March 2026 | Adobe Analytics, April 2026 |
| Engagement of AI-referred visitors | +28% add to cart; +59% time on site; 33% lower bounce | Adobe Analytics, July 2026 |
| Shopify merchants, AI-referred traffic and orders, year over year | Traffic about 8 times; orders about 13 times | Shopify, Q1 2026 |
| AI-influenced online sales, Cyber Week 2025 | 67 billion US dollars globally; AI touched 20% of orders | Salesforce, December 2025 |
| Consumers who have used AI for online shopping | 39%; 85% of them say it improved the experience | Adobe survey, April 2026 |
| Consumers open to an AI agent in the shopping journey | 85% | Accenture, cited by Anthropic, September 2026 |
| Retailer content not readable by AI | Home pages: 25%; product pages: 34% | Adobe, April 2026 |
| B2B buying intermediated by AI agents by 2028 | 90%; more than 15 trillion US dollars | Gartner, November 2025 |
| Agentic AI projects cancelled by the end of 2027 | More than 40% | Gartner, 2025 |
| Global agentic commerce opportunity by 2030 | 3 to 5 trillion US dollars | McKinsey, October 2025 |
| Consumers who use AI for part of the buying journey | 45%; 83% share worries about privacy, data misuse, or unsolicited marketing | IBM Institute for Business Value, January 2026 |
| Ecommerce sales enabled by AI agents by 2030 | 25%; 55% of digital consumers will start research in LLM platforms | Deloitte, cited 2026 |
| Shoppers who expected to use AI in holiday purchases | 64%; 84% for ages 18 to 24 | Shopify Global Holiday Report, 2025 |
What the data says
- The traffic is real and it is high intent. Growth slowed from 393% to 62% year over year because the base grew, not because interest fell. The quality metrics keep improving.
- Checkout inside assistants underperformed. The merchants that win in 2026 keep checkout on their own surface and make the handoff easy for agents.
- The protocol war became a protocol stack. UCP, ACP, AP2, and MCP moved toward neutral governance and composition, not replacement.
- The card networks built the trust layer. Visa and Mastercard now supply agent identity, tokens, spend controls, and fraud tools. That removes the main blocker for Level 3.
- Readiness is the bottleneck. One quarter of retailer homepage content and 34% of product-page content is invisible to AI. In B2B, Gartner reports that 74% of procurement leaders say their data is not AI-ready.
Agentic commerce timeline, 2024 to 2026
Agentic commerce moved from a protocol in November 2024 to packaged merchant agents in September 2026. The milestones below explain why the market looks the way it does today.
| Date | Event | Why it matters |
|---|---|---|
| November 2024 | Anthropic releases the Model Context Protocol (MCP) | Gives agents a standard way to read data and call tools |
| April 2025 | Amazon tests Buy for Me | A marketplace agent buys from third-party sites for the customer |
| April to May 2025 | Mastercard Agent Pay and Visa Intelligent Commerce are announced | The card networks commit to agent-initiated payments |
| September 2025 | Google publishes the Agent Payments Protocol (AP2). OpenAI and Stripe publish the Agentic Commerce Protocol (ACP) and launch ChatGPT Instant Checkout | The first open standards for mandates and in-chat checkout |
| October 2025 | Visa Trusted Agent Protocol with Cloudflare | Verified agent identity at the HTTP layer |
| November to December 2025 | Holiday AI traffic up 693% (Adobe); 67 billion US dollars of AI-influenced Cyber Week sales (Salesforce); Gartner publishes its 15 trillion US dollar B2B forecast | Agentic demand becomes measurable |
| December 2025 | MCP is donated to the Linux Foundation’s Agentic AI Foundation | Neutral governance for the communication layer |
| January 2026 | Google launches the Universal Commerce Protocol (UCP) at NRF with Shopify, Etsy, Wayfair, Target, and Walmart | A full-lifecycle commerce standard for the open web |
| March 2026 | OpenAI winds down Instant Checkout and moves ChatGPT commerce to discovery plus retailer apps. UCP adds cart and catalog access | The market settles on “discover in AI, buy on site” |
| April 2026 | AP2 is donated to the FIDO Alliance. Adobe reports AI traffic up 393% and the conversion reversal | The trust layer gets neutral governance. The quality of AI traffic is proven |
| May 2026 | Google Universal Cart across Search AI Mode, Gemini, YouTube, and Gmail | Level 2 at Google scale |
| June 2026 | Visa and OpenAI enable tokenized agent checkout with spend caps. Visa launches Agent Score and the Agentic Directory. Mastercard launches Agent Pay for Machines. Adyen launches Adyen Agentic | Level 3 infrastructure becomes available |
| September 2026 | Anthropic launches Claude Commerce Agents with Shopify, Visa, Mastercard, and Accenture | Merchant-owned agents become a packaged option |
Why agentic commerce matters for your business
Agentic commerce matters because it brings higher-intent demand at a lower cost of sale, and because a business that agents cannot read will lose that demand to one they can. The benefits and the risks below apply to both B2C and B2B.
Benefits of agentic commerce for businesses
- Higher-intent demand. AI-referred visitors convert better and spend more per visit than other traffic. See the data table above.
- A new channel with low marginal cost. Once your data is agent-readable, every assistant that reads it becomes a distribution point.
- Lower cost of sale in B2B. Agents can take over repeat orders, quotes, and order status requests. When Elogic Commerce connected Adobe Commerce to SAP S/4HANA for Armacell, order approvals became five times faster and manual orders fell by 40%. Those are the workflows a Level 4 agent needs before it can operate. See the Armacell case study.
- Always-on selling. Agents work outside office hours and across languages.
- Better data. Agent traffic shows which attributes and offers win, at the attribute level.
Risks and open problems
- Loss of control over presentation. The agent, not your page, describes your product. Wrong or missing data becomes the agent’s answer.
- Agent verification and fraud. Merchants must tell a verified agent from a scraper or a fraud bot. Visa Trusted Agent Protocol, Mastercard agentic tokens, and AP2 mandates address this, but adoption is early.
- Margin pressure. When agents compare every offer, price and delivery terms become transparent. Differentiation moves to service, availability, and trust.
- Liability and consent. Who is responsible when an agent buys the wrong item? Mandates and consent proofs are the answer, but the rules differ by market. In the EU, strong customer authentication under PSD2, the coming PSD3 and PSR framework, and the EU AI Act all apply.
- Pilot failure. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027 because of cost, unclear value, or weak risk controls. Scope each pilot to one level and one use case.
The three trust questions
Every agent-initiated payment must answer three questions that a human checkout answered by default. Who pressed “buy”? The card belongs to a person, but software made the click, so the merchant and the issuer need proof that the agent is a legitimate representative. Who detects fraud? Typing speed, mouse movement, and device fingerprints disappear when an agent buys, so risk models must be rebuilt around agent identity and mandates. Who is liable when the agent orders the wrong size or the wrong date? Mandates, consent records, and audit logs decide that. The protocols in the stack table exist to answer these three questions.
Agentic commerce in B2B
B2B is where agentic commerce reaches Level 4 first. The reasons are structural. B2B orders repeat. Prices are in contracts. Approval rules already exist. Buyers prefer self-service: Gartner reports that 67% of B2B buyers prefer a rep-free purchase, and 45% used AI in their most recent purchase. Gartner also expects 90% of B2B buying to be intermediated by AI agents by 2028.
There is a counterweight. Gartner also expects 75% of B2B buyers to prefer human-led sales for complex purchases by 2030. The realistic split: agents take tail spend, replenishment, and research; humans keep negotiated, high-risk, and new-category purchases.
What a B2B agent needs from your commerce platform
- Customer-specific pricing and contract terms through an API, with the same numbers a human sees.
- Real-time stock and lead times by warehouse.
- Product data with complete technical attributes, units of measure, and compatibility rules, usually from a PIM such as Akeneo, Pimcore, or inRiver.
- Approval workflows and spend limits that the agent can call, not only a web form.
- ERP integration for orders, invoices, and credit limits: SAP S/4HANA, Microsoft Dynamics 365, Oracle NetSuite, Visma, or Odoo.
- A successor to punchout: an MCP server or a UCP-style endpoint that a buyer’s agent can use instead of a browser session.
- An audit log of every agent action for finance and compliance.
Where to start in B2B: replenishment of known SKUs for existing customers. The mandate is clear, the data exists, and the risk is contained. Elogic Commerce builds this on Adobe Commerce, Shopify Plus, commercetools, SAP Commerce Cloud, and Shopware, connected to the ERP and PIM systems above. See our agentic commerce services.
How to prepare: an agentic commerce readiness checklist
To prepare for agentic commerce, make your products readable, your prices consistent, your checkout handoff clean, and your payments agent-aware. Marketplaces and merchants use the same ten checks.
- Structured product data on every product page: schema.org Product and Offer with price, availability, GTIN, and shipping details.
- A complete, consistent product feed in Google Merchant Center and in the formats that ACP and UCP read.
- Full technical attributes in the PIM. Fill every field an agent can filter on.
- Crawl access for the AI systems you want to serve (for example GPTBot, ClaudeBot, PerplexityBot, and Google-Extended), with rate limits instead of blocks.
- Prices, stock, and delivery promises that are identical across the site, the feed, and the API.
- An agent endpoint: an MCP server or protocol support (UCP, ACP) on the platforms where you sell.
- A checkout handoff that keeps the cart, the customer, and the discount when an agent sends a shopper to you.
- Agent-aware payments: tokenized credentials, agent identification, and fraud rules from your PSP and the card networks.
- Consent, mandate, and audit records for every agent-initiated action.
- Measurement that separates three things: AI-assisted discovery, agent-built carts, and agent-executed purchases. One “AI channel” line in analytics is not enough.
Metrics for agentic commerce
| Metric | What it measures | Where to measure it |
|---|---|---|
| AI-referred sessions | Visits that arrive from ChatGPT, Gemini, Claude, Perplexity, Copilot, and other assistants | Web analytics referrer and UTM rules; Adobe and GA4 AI source groupings |
| Agent inclusion rate | Share of relevant assistant answers or shortlists that include your brand or product | Prompt monitoring tools such as Ahrefs Brand Radar or Semrush AI Visibility, by category prompt |
| AI answer accuracy | Whether assistants describe your products, prices, and policies correctly | Monthly prompt audit against the catalog and policy pages |
| Catalog readability score | Share of products with complete structured data, feed fields, and policy fields | PIM completeness reports; the Elogic Commerce Agentic Commerce Readiness Index checks |
| Agent conversion rate | Conversion of AI-referred sessions and agent-built carts against organic search | Analytics segments; Shopify channel attribution for Agentic Storefront orders |
| Revenue per agent visit | Revenue divided by AI-referred sessions | Analytics; compare with the Adobe benchmark of 53% above other traffic |
| Agent-executed orders | Orders completed by an agent inside a mandate (Level 3 and above) | Order flags from the PSP or card network agent indicators; ERP order source codes |
Measure your position with the Elogic Commerce Agentic Commerce Readiness Index, our crawl-based study of how ready commerce sites are for AI agents.
The future of agentic commerce, 2027 and beyond
The future of agentic commerce is agent-built carts as the default B2C pattern, delegated purchases limited by spend caps, and autonomous replenishment in B2B. This is the Elogic Commerce view, based on the data above.
- Level 2 becomes the default B2C pattern by the 2027 holiday season. Agent-built carts with merchant checkout will be a standard feature on the major platforms.
- Level 3 grows through spend caps, not full delegation. Consumers will delegate small, repeat purchases with hard limits first.
- B2B reaches Level 4 for replenishment before B2C reaches Level 3 at scale.
- Protocol support becomes a platform feature. Adobe Commerce, Shopify, commercetools, and SAP Commerce Cloud will ship UCP, ACP, and MCP endpoints as core capabilities, not custom builds.
- Agent inclusion rate becomes a board-level metric, next to organic visibility and paid return on ad spend.
- Regulation catches up. Expect rules on agent consent, liability, and disclosure in the EU and the US by 2028.
Frequently asked questions about agentic commerce
What is agentic commerce in simple terms?
Agentic commerce is when you tell an AI agent what you need and it finds, compares, and buys it for you within the limits you set.
Can you give an example of agentic commerce?
A shopper asks an assistant for a road bike under 1,500 euros with delivery in three days. The agent reads catalogs from several stores, compares options, builds a cart at the best-fit store, and hands it to that store’s checkout for the shopper to approve. A B2B example: a procurement agent reorders packaging film when stock falls below two weeks of use, inside an approved budget.
What is the difference between agentic AI and agentic commerce?
Agentic AI is the general technology: AI systems that plan, use tools, and take actions. Agentic commerce is the use of agentic AI to do the steps of a purchase or a sale.
What is the difference between agentic commerce and agent-assisted commerce?
In agent-assisted commerce the agent prepares the work and the human approves the final action. That is Level 1 and Level 2 of the Elogic Commerce model. In full agentic commerce the agent completes the action inside a mandate, which is Level 3 and above. Most 2026 journeys are agent-assisted.
What are the risks of agentic commerce?
Loss of control over how products are presented, agent fraud and verification, margin pressure from transparent comparison, unclear liability when an agent makes a mistake, and failed pilots. Accurate data, agent verification, mandates, and small scoped pilots reduce each risk.
Is ChatGPT shopping agentic commerce?
Yes, at Level 1 and Level 2 of the Elogic Commerce model. ChatGPT discovers and compares products and sends shoppers to retailer apps or sites to buy. OpenAI wound down in-chat checkout in March 2026. Agent-initiated checkout with tokenized Visa credentials and user-set spend limits was introduced in June 2026.
What is agentic checkout?
Agentic checkout is a checkout that an agent completes for a person, either on the merchant’s surface or through a protocol such as ACP or UCP, with payment through a tokenized credential and a proof of the person’s mandate.
Does my website still matter in agentic commerce?
Yes. Your site is still the source of truth that agents read and the place where most checkouts complete in 2026. It must now serve two readers: people who browse and agents that parse structured data, policies, and APIs.
Which platforms support agentic commerce?
Shopify through Agentic Storefronts, UCP, and ACP; Adobe Commerce, commercetools, SAP Commerce Cloud, and Shopware through their APIs, MCP servers, and protocol integrations. Payment support comes from Stripe, Adyen, PayPal, Visa, and Mastercard.
Does agentic commerce work for B2B?
Yes. B2B is the first place agents reach autonomous replenishment because orders repeat, prices are contracted, and approval rules exist. Gartner expects agents to intermediate 90% of B2B buying by 2028.
How is agentic commerce different from conversational commerce?
Conversational commerce changes the interface to chat or voice. Agentic commerce changes who does the work. An agent can act through any interface, including no visible interface at all.
What is an agentic commerce protocol?
A shared standard that lets an agent and a merchant exchange catalog, cart, checkout, order, and payment information. The main protocols in 2026 are UCP, ACP, AP2, and MCP.
Agentic commerce glossary
| Term | Definition |
|---|---|
| AI agent | Software that uses an AI model to plan, call tools, and take actions toward a goal. |
| Shopper agent | An agent that acts for a buyer, consumer or business. |
| Merchant agent | An agent that acts for a seller: catalog, pricing, inventory, quotes, order operations, and service. |
| Mandate | The goal, limits, and permission a person or an organization gives an agent. AP2 formalizes it as signed Intent, Cart, and Payment mandates. |
| Agentic checkout | A checkout completed by an agent, on a merchant surface or through a protocol, with a tokenized credential and a proof of the mandate. |
| Agent-readable catalog | Product data that an agent can read and trust: structured data, feeds, and APIs with complete attributes. |
| ACP (Agentic Commerce Protocol) | The OpenAI and Stripe standard for product feeds and checkout between an agent and a merchant. |
| UCP (Universal Commerce Protocol) | The Google standard, built with Shopify and major retailers, for discovery, cart, checkout, and orders. |
| AP2 (Agent Payments Protocol) | A standard for cryptographic proof that a person authorized an agent’s payment. Started at Google, now governed by the FIDO Alliance. |
| MCP (Model Context Protocol) | A standard for connecting AI models to tools and data. Started by Anthropic, now under the Linux Foundation. |
| A2A (Agent2Agent) | The Google protocol for communication between agents. |
| Trusted Agent Protocol (TAP) | The Visa standard, built with Cloudflare, for verifying agent identity and intent at the HTTP level. |
| Agent Pay | The Mastercard program for agent-initiated payments with agentic tokens. Agent Pay for Machines extends it to agent-to-agent transactions. |
| Tokenized credential | A payment token that stands in for card data, so the agent never holds the card number. |
| Human in the loop | A control point where a person must approve an agent’s action before it proceeds. |
| Agent inclusion rate | The share of relevant agent answers or shortlists that include your product or brand. |
Cite this definition
Elogic Commerce (2026). “What Is Agentic Commerce? Definition, Examples, and How It Works.” https://elogic.co/blog/what-is-agentic-commerce/. Free to quote and to reuse the five-level model with a link to this page.
About the author
Paul Okhrem is the CEO and Founder of Elogic Commerce, a B2B and B2B2C commerce engineering agency founded in 2009 with 200+ specialists and offices in Tallinn, New York, London, Stockholm, Dresden, and Prague. He is a member of the Forbes Technology Council and received the Magento Community Engineering Award at Adobe Imagine 2019. Elogic Commerce is an Adobe Commerce Silver Solution Partner, a Hyvä Bronze Partner, and holds ISO 27001, SOC 2 Type II, and ISO 9001 certifications.
Make your commerce platform agent-ready
Elogic Commerce builds agent-readable catalogs, MCP and protocol endpoints, agent-aware checkout, and B2B replenishment agents on Adobe Commerce, Shopify Plus, commercetools, SAP Commerce Cloud, and Shopware. Start with an agentic commerce readiness assessment, or see AI Chat Core, our working commerce agent. Contact office@elogic.co.
Sources
- Adobe Digital Insights, “AI traffic surges to retail sites; much of their content is not machine-readable”, April 2026.
- Digital Commerce 360, “Adobe: AI-referral traffic spending, converting more than counterparts”, 19 August 2026.
- Digital Commerce 360, “Adobe: AI-referred traffic to retail sites doubles in a year”, 17 June 2026.
- Digital Commerce 360, “Gartner: AI agents will command $15 trillion in B2B purchases by 2028”, 28 November 2025.
- Digital Commerce 360, “Anthropic debuts Claude features focused on agentic commerce”, 2 September 2026.
- Anthropic, “Building commerce agents with Claude”, 2 September 2026.
- Forbes, “Why OpenAI’s Checkout Retreat Spells Trouble For Its Commerce Strategy”, 10 March 2026.
- Google, “New tech and tools for retailers to succeed in an agentic shopping era” and the UCP developer guide.
- Stripe, Agentic Commerce Protocol documentation and the ACP specification.
- IBM Think, “What is agentic commerce?”, 23 January 2026.
- McKinsey, “The agentic commerce opportunity”, October 2025.
- Deloitte, agentic commerce and AI shopping agents guide.
- Adobe newsroom, holiday shopping season 2025 results, January 2026.
- Shopify, “Agentic Commerce: Benefits and How To Get Started”, 2 April 2026, and the 2025 Global Holiday Retail Report.
- FIDO Alliance, trusted AI agent interactions and AP2 governance, 28 April 2026.
- Visa Intelligent Commerce and the Trusted Agent Protocol announcement.
- Mastercard Agent Pay and the launch announcement.
- Salesforce, Cyber Week 2025 results, December 2025.
- Shopify, “Agentic Commerce on Shopify: How It Works”, 2026.
- Accenture, Consumer Pulse Research 2026.
Get in Touch
Looking for a partner to grow your business? We are the right company to bring your webstore to success.