Complete Guide

What is Agentic Commerce? How AI Shopping Agents Buy for You

The complete guide to understanding agentic checkout: how AI agents autonomously complete purchases, transforming ecommerce from manual checkout to machine customers buying on your behalf.

12 min read
For Merchants & Engineers

Quick Definition

Agentic commerce is ecommerce where AI agents autonomously complete purchases on behalf of users. Instead of humans browsing and clicking checkout, users delegate shopping tasks to AI assistants like ChatGPT or Claude, which discover products, evaluate options, and execute transactions programmatically through merchant APIs.

What's New Right Now

  • ChatGPT Shopping: Integrated with select merchants including Etsy. Walmart partnership announced October 2025.

  • Perplexity Payments: Checkout via PayPal and Venmo now available in Perplexity AI assistant. (PayPal)

  • ChatGPT Instant Checkout: OpenAI launched Instant Checkout, allowing users to buy directly from merchants like Etsy in chat, powered by the Agentic Commerce Protocol. (OpenAI)

  • Google Agent-to-Agent Protocol (A2A): Google launched A2A to enable AI agents to communicate and coordinate actions across platforms, with 50+ partners including PayPal, Salesforce, and ServiceNow. (Google)

  • Agent Payment Protocol (AP2): Google announced AP2, an open protocol for secure agent-led payments using verifiable digital credentials, with 60+ partners including Mastercard, PayPal, and American Express. (AP2)

  • Visa Trusted Agent Protocol: Initiative for identifying and authenticating legitimate shopping agents.

Philipp Rufinatscha

Philipp Rufinatscha

CEO, AgenticCart

Last updated: October 17, 2025

Understanding the Concept

What is Agentic Commerce?

Agentic commerce represents a fundamental shift in how online transactions occur. Rather than users manually navigating websites, comparing products, filling out forms, and completing checkout, they delegate the entire purchasing process to an AI agent. The user provides intent ("Reorder my coffee beans," "Buy a birthday gift for my mom under $50"), and the agent autonomously handles discovery, evaluation, and transaction completion.

This isn't an incremental improvement to existing checkout flows. It's a new transaction paradigm. The "shopper" is software, not a human. The merchant's job shifts from optimizing button placement and form fields to exposing structured APIs that agents can reliably query, interpret, and use to transact.

For merchants, this means enabling programmatic access to product catalogs, inventory status, pricing, and checkout. For users, shopping becomes a delegation task rather than a manual chore. For agents, completing purchases requires authentication, consent management, delegated payments, and robust error recovery, with minimal human intervention after initial approval.

Agentic commerce vs. adjacent concepts

Conversational commerce

You chat with a bot for recommendations, then you complete checkout. In agentic commerce, the agent, under your consent, can place the order autonomously.

Chatbots

Answer questions and guide shoppers. Agentic agents have delegated authority to complete the transaction end-to-end (with approvals as needed).

Recommendation engines

Predict and suggest products. Agents research, select, and purchase, not just suggest.

What agentic commerce is not

Not just recommendations: Agents complete purchases, not merely suggest items.

Not autofill: Agents make decisions, handle errors, and adapt to constraints, beyond form filling.

Not RPA: Agents use structured, documented APIs (with idempotency and webhooks), not brittle screen scraping.

Not a single protocol: It's a category/architecture. Multiple approaches (e.g., ACP, MCP, A2A, AP2) can work together.

Note: High-value, age-restricted, or first-time purchases can require explicit human approval based on policy.

Transaction Flow

How Agentic Commerce Works

Agentic commerce follows a multi-step flow from user intent to fulfillment. Each step involves specific inputs, actions, potential failures, and success signals.

Step 1

Intent & Discovery

Understanding what you need and finding the best options

"Reorder my usual coffee beans, but check if there's a deal on the 2lb bag. Need it by Friday."

Natural language understanding extracts product attributes, quantity, preferences (size, flavor, brand), delivery constraints, and price signals

Agent queries merchant APIs (WooCommerce, Shopify, etc.) or aggregator platforms using structured search parameters and filters

Results are normalized across sources, enriched with real-time inventory and pricing data, and validated against user purchase history

All candidates are logged with provenance metadata (source, timestamp, query context) for audit trails and transparency

Shortlist generation follows configurable policies (typically 3–5 ranked options with comparative pricing and availability)

Options found & ready
Ambiguous or missing data
Step 2

Evaluate & Purchase

Comparing options and completing the transaction

Found 2lb coffee at $34.99 (15% off)

Order #AC-8472991 · Completed in seconds (network/processor dependent)

Evaluates options on price, delivery window, quality signals (with provenance)

Authenticates using OAuth 2.0 / OIDC where available (or signed API access)

Re-validates price, tax, shipping, and stock at checkout; uses idempotency keys on cart/payment/order writes

Executes tokenized payment with SCA/EMV 3-DS when required; enforces spending limits and approval gates for high-risk purchases

Returns an order confirmation and consent/audit references

Payment confirmed & order placed
Payment or stock issues
Step 3

Fulfill & Support

Tracking delivery and handling any issues

Thursday

Order ships from warehouse

Friday 3:42pm

Delivered successfully

After delivery

Merchant logged as reliable

Subscribes to signed order/shipment webhooks with retry; falls back to scheduled polling when necessary

Sends milestone updates; handles returns/RMAs via API; updates merchant trust/reliability scores

Maintains tamper-evident audit logs and trace IDs for explainability and disputes

Delivered & merchant trusted
Delays or delivery issues
Real-World Applications

Use Cases & Examples

Agentic commerce delivers measurable value across diverse shopping scenarios. Here are the most common patterns and their real-world impact.

Agentic Commerce Examples in Production

ChatGPT Shopping & OpenAI Instant Checkout

ChatGPT can complete purchases (Etsy live), with Shopify integration announced; Walmart has said it's exploring a partnership.

Perplexity Buy with Pro

Perplexity's 'Buy with Pro' lets users purchase via PayPal and Venmo from AI search results. (PayPal)

Vertical AI Shopping Agents

Industry-specific agents for electronics, fashion, and groceries handling repeat purchases and price monitoring.

B2B Procurement Agents

Enterprises are adopting procurement agents that automate sourcing, supplier comparisons, and draft POs via API-driven workflows (e.g., SAP's Sourcing Agent).

01

Subscription Replenishment

Your coffee consumption is predictable. The agent knows you brew 2 cups daily and your current bag has maybe 5 days left. Rather than letting you run out or forcing you to remember to reorder, it checks your preferred merchant, finds your usual beans are in stock, notices there's a 15% discount on the 2lb bag, and places the order automatically.

This isn't speculative, it's based on actual purchase history, confirmed SKU, and delivery timing that ensures arrival before you run out. Merchants often see improved retention and customer lifetime value because the friction of remembering to reorder simply disappears. The agent handles variant confirmation by cross-referencing previous purchases, eliminating the "wrong product" problem that plagued earlier automation attempts.

02

Price-Watch Automation

You've been eyeing a standing desk for months. It's $399 but you'd buy it at $320. Instead of manually checking prices or setting up unreliable alerts, you tell your agent: "Buy this if it drops below $320." The agent monitors continuously, and when a flash sale hits at $315, it validates stock availability, confirms shipping costs won't push the total over budget, and completes checkout in seconds.

The speed matters because good deals vanish quickly. Users report significantly faster purchase execution and the elimination of "saw the deal but it sold out while I was checking out" scenarios. The agent re-validates price and stock at checkout and aborts if the total exceeds the threshold, preventing the classic flash-sale trick where advertised prices revert during checkout. This is opportunistic commerce, buying only when conditions align perfectly with your constraints.

03

Enterprise Procurement Automation

A manufacturing facility's inventory system detects that aluminum stock will hit reorder thresholds in 72 hours. The procurement agent queries approved supplier APIs, compares bulk pricing across three vendors, verifies lead times, checks for recent quality issues in the ERP system, and places a 500-unit order with the optimal supplier, within pre-approved vendor lists and budget thresholds.

Companies using these systems report automating routine procurement with high order accuracy. The key difference from consumer use cases is approval workflows and constraint enforcement: the agent never exceeds budget caps, always uses approved vendors, and escalates anomalies (like sudden 40% price spikes) for human review. This is high-stakes automation where mistakes are expensive, which is why audit trails, rollback capabilities, and clear decision logs are non-negotiable.

Value Analysis

Benefits and Trade-offs

Agentic commerce delivers measurable improvements in conversion rates, support costs, and customer lifetime value, and can deliver improvements within a quarter depending on baseline, catalog complexity, and payment/SCA mix. Early adopters report meaningful conversion improvements, driven by the elimination of checkout friction and decision paralysis.

However, this shift introduces new challenges around trust, explainability, and compliance that merchants must address deliberately. Users need transparency into agent decisions, clear consent mechanisms, and confidence that automation won't result in unwanted purchases or compliance violations. The trade-offs are real but manageable with proper infrastructure.

What You Gain

Higher conversion rates by eliminating checkout friction. Agents don't exhibit human drop-off; failures are mainly technical (auth, stock, SCA). Retries and idempotency reduce abandonment-like failures. The path from intent to completed order typically takes seconds to a minute.

Reduced support costs as agents autonomously handle product questions, inventory checks, order tracking, and basic returns. Can reduce Tier-1 volume; expect a temporary spike during rollout and more complex exceptions routed to humans.

Improved customer lifetime value through subscription automation and proactive reordering. Customers who would have churned due to inconvenience stay because agents handle the tedious parts.

True 24/7 revenue generation without requiring night staff or international support teams. Agents operate across all time zones simultaneously, capturing purchase intent whenever it emerges.

What You Must Manage

Trust barriers prevent adoption if users fear unauthorized purchases or opaque decision-making. Transparent consent flows, spending limits, and clear approval thresholds address this, but building trust takes time and iteration.

Explainability requirements mean you need comprehensive audit trails. Users asking "why did you buy this?" expect clear answers showing what options were considered, how decisions were weighted, and why this choice won.

Product data quality becomes critical because sparse metadata breaks agent decision-making. Missing specifications, vague descriptions, or incorrect inventory status lead to poor choices and higher return rates until data improves.

Compliance complexity increases with automated purchasing. Log consent receipts and decision rationale per order; gate age-restricted SKUs; enforce region/country rules at API time. Age verification, regional restrictions, prescription requirements, and export controls all need programmatic enforcement with proper error handling and escalation paths.

How to Measure Success

Track these metrics to validate agent performance and identify optimization opportunities:

Agent conversion rate

Auth approval rate

Median time-to-order

Refund/return rate vs human baseline

Chargeback rate

% orders requiring human approval

Agent success rate (no manual fallback)

CSAT for agent-placed orders

Technical Approach

Implementation Patterns

Four core capabilities enable agentic commerce. These patterns are protocol-agnostic and work with ACP for merchant APIs, A2A for agent coordination, MCP for tool access, and AP2 for payments, alongside existing standards (OAuth 2.0, OpenID Connect) and network-level verification emerging (e.g., Trusted Agent Protocol). Choose the integration approach that fits your stack.

Standards You Will Touch

Identity & Consent

OAuth 2.0 with granular scopes. Users grant explicit permissions for specific capabilities: view products, place orders under $100, access history. DPoP or mTLS token binding where supported. Every transaction includes consent receipts, mandate IDs, and audit trails.

Watch out: Vague scope descriptions lead to blind approvals

Product Catalog APIs

Expose structured APIs with search, filtering, variants, and live inventory. Agents convert natural language queries ("organic coffee under $20 with free shipping") into API parameters and get machine-readable responses.

Quality metadata is non-negotiable: complete descriptions, specifications, images, and accurate stock status enable intelligent agent decisions.

Watch out: Sparse metadata breaks decision-making

Checkout Automation

Stateless cart APIs with idempotency keys. Agents create carts, apply discounts, calculate shipping, and complete checkout. Minimize round-trips; make every write idempotent; support async confirmation via webhooks. Stock reservation/hold windows and concurrency controls prevent oversells. SCA exemptions where allowed; fall back to challenge when required.

Watch out: Missing idempotency causes double-charging

Payment Delegation

Processors like Stripe support delegated charging via Payment Intents and Setup Intents (off_session) with stored payment methods. Use Connect only in platform/marketplace setups. PCI-DSS maintained: agents never touch raw card data.

Watch out: Never pass raw card data through agents

Real-Time Updates

Webhooks notify agents of status changes: shipped, delivered, delayed. Verify signatures, retry with exponential backoff, ensure idempotency in webhook handlers to avoid double-processing. Agents monitor autonomously and alert users only when needed. Include tracking numbers and ETAs.

Watch out: Webhook failures break agent awareness

Reference Architecture

1. User Intent
Natural language request to agent
2. Agent Processing
Parse intent, extract constraints
3. Product API
Query catalog, inventory, pricing
4. Cart API
Idempotent cart creation
5. Payment Processor
Tokenized payment, SCA where required
6. Webhooks
Order status updates to agent
7. Post-Purchase
Returns, learning, trust signals

Critical patterns: Retries with exponential backoff, rate limiting (per-agent), circuit breakers on API failures, idempotency keys on all write operations. See idempotency and webhook docs.

Edge Cases to Handle

Age restrictions
Regional limits
Rate limiting
Stock changes
Price fluctuations
Payment failures
Address validation
Return windows
Multi-merchant cart
Price at checkout drift
Address normalization
Pre-order/backorder

Protocols Powering Agentic Commerce

Agentic commerce isn't a single API; it's a stack. In practice you'll combine merchant commerce surfaces (ACP), agent–tool interfaces (MCP), agent interoperability (A2A), and secure payments (AP2):

ACP (Agentic Commerce Protocol)

A merchant-facing API schema so agents can search products, price offers, build carts, and place orders with idempotency and webhook callbacks. ACP defines resources and events that map cleanly to autonomous checkout.

MCP (Model Context Protocol)

Connect agents to merchant tools/data (catalog, inventory, shipping) via a standardized, secure interface.

A2A (Agent-to-Agent)

Interoperability between agents: discovery, messaging, and coordination for multi-party purchase flows.

AP2 (Agent Payments Protocol)

Standardized agent payments with signed mandates, risk roles, and multi-rail support (cards, A2A/open-banking, stablecoins).

Note: "A2A payments" (account-to-account bank transfers) are a payment rail that AP2 can use; not the same as A2A agent interoperability.

Protocol Comparison: ACP vs MCP vs A2A vs AP2

Protocol Purpose Transport Auth Maturity Maintained By
ACP Merchant commerce APIs (catalog, cart, checkout) REST/JSON OAuth 2.0 Emerging OpenAI + Stripe
MCP Agent↔tool/data access JSON-RPC 2.0 (stdio/WebSocket/HTTP) Host-provided Early production Anthropic
A2A Agent interoperability/coordination HTTP/gRPC (varies) mTLS and/or OAuth service creds Pilot Google + partners
AP2 Payment mandates & multi-rail REST/JSON Signed mandates (e.g., JWS) Draft/early pilots Google Cloud + PSPs

Merchant Readiness Checklist

Is your store ready for agentic commerce? Use this checklist to assess your current capabilities.

Requirement Status Priority
Structured product catalog with SKU + global IDs (GTIN/UPC/EAN/ISBN/MPN) and pricing Required Critical
Real-time inventory status API (or cache TTL ≤ 5 min) Required Critical
OAuth 2.0 auth with granular scopes (or mTLS service creds for S2S) Required Critical
Idempotent checkout and order write endpoints Required Critical
PCI-DSS compliant processor (e.g., Stripe, PayPal) with off_session + mandates Required Critical
ACP-compliant product & checkout endpoints (via AgenticCart) Required Critical
Webhook events for order lifecycle with signature verification + retries Required High
Detailed product metadata (variants, dimensions, materials, images) Recommended High
Per-agent rate limiting + fraud/risk checks Recommended High
Audit logs & consent/mandate IDs for agent transactions Required High
Ecosystem

The Agent Landscape

Multiple agent types serve different use cases. Knowing who builds and controls the agent determines how you integrate and where data/consent lives.

General AI Assistants

ChatGPT, Claude, Gemini. Multi-purpose assistants that can include shopping as a capability via partner integrations.

OpenAI Anthropic Google

Best for: Broad adoption, cross-category purchases, users who prefer one agent for "everything"

Merchant-Defined Agents

Store/brand-owned agents with deep catalog understanding and policies tuned to a single merchant.

Brand-specific Deep catalog knowledge

Best for: Brand loyalty programs, specialized catalogs, high-touch CX

Vertical Shopping Agents

Purpose-built for product research and comparison (e.g., AI shopping/search apps). Often aggregate across merchants; monetization may include affiliate links.

Product research Price comparison

Best for: Complex product research, first-time purchases, price-conscious shoppers

Enterprise Procurement Agents

Company-specific agents integrated with ERP/PO systems and approval workflows; enforce budget/vendor policies.

ERP integration Approval workflows

Best for: High-volume B2B, regulated industries, complex compliance rules

Personal Shopping Agents

User-trained agents that learn preferences, sizes, budgets, and reorder rhythms. Operate with explicit permission and scoped access.

Personalized Learns preferences

Best for: Recurring purchases, subscription management, gifting

Trust & Compliance

Ethics, Trust & Compliance

Agentic commerce introduces new trust and compliance considerations. Addressing these proactively builds user confidence and reduces regulatory risk.

01

Transparency & Explainability

Users must understand why an agent made specific purchase decisions. Provide comprehensive audit trails and clear explanations that build trust and enable informed oversight. Every autonomous action should be traceable, interpretable, and justifiable.

Product options considered — Which alternatives were evaluated and why they were included
Selection criteria applied — User preferences, price limits, delivery requirements used
Price comparisons made — Historical data, competitor analysis, value assessment

"I bought this because you ordered it 3 times previously, it's in stock, and price is 10% below average."

02

User Approvals & Consent

Explicit consent for agent capabilities with risk-based approval thresholds. Users should be able to set spending limits, category restrictions, and approval thresholds. Provide a consent/approval receipt for each purchase (e.g., per Kantara CR pattern).

Low risk: View products
Medium risk: Place orders under $50
High risk: Orders over $50 (require confirmation)
03

Audit Trails

Log every agent action: authentication events, product queries, cart operations, checkout attempts, payment authorizations. Logs must be tamper-evident and available for user review and regulatory audits.

Retention: Per PCI DSS (≥1 year, 3 months online) and applicable financial/legal requirements (often 5–7 years for business records)
04

Data Minimization

Agents should request minimal data: product preferences, budget constraints, delivery address. Avoid collecting unnecessary personal information.

Purpose limitation: Collect only for stated purpose

Storage limitation: Delete when no longer needed

Data accuracy: Keep information current and correct

05

Regional Compliance

Different jurisdictions have varying rules on automated transactions, electronic signatures, age verification, and consumer protection.

EU: GDPR Consent & DPIA

Requires lawful basis, transparency, and right to object. Conduct DPIA when processing is likely high risk. Art. 22 safeguards apply for solely automated decisions with legal/significant effects.

EU: eIDAS

eIDAS provides a trust-services framework, but ordinary ecommerce purchases do not require qualified electronic signatures.

EU: PSD2 SCA & EMV® 3DS

Strong Customer Authentication required for most card payments in the EU/EEA, with exemptions (low value, low risk, trusted beneficiary). Implement EMV® 3-D Secure.

US: State-by-State

CCPA for California, varying age verification laws, state-specific consumer protection rules.

Consult legal for regulated verticals: healthcare, alcohol, finance. See GDPR, eIDAS.

06

High-Discretion Purchase "Red Lines"

Certain purchases should always require explicit human approval. Agents request approval, explain rationale, and wait for user confirmation.

Purchases over a configurable threshold (e.g., policy-based limit)
Age-restricted items (where statutory age-verification is required)
Prescription medications (where statutory approval is required)
Financial products with legal liability

Ecosystem Security

To distinguish legitimate shopping agents from bots, card networks and PSPs are piloting verification frameworks (e.g., Visa's Trusted Agent Protocol). Mention this in your risk and fraud controls.

Read more
Common Questions

Frequently Asked Questions

Quick answers to the most common questions about agentic commerce.

How is agentic commerce different from chatbots or human checkout?

Agents complete the purchase end-to-end; humans still browse/decide for complex buys. This supplements, not replaces, human checkout.

How do I get started as a merchant?

Focus on product data quality, real-time stock/pricing, and agent-ready APIs. If you're on WooCommerce, try AgenticCart; AgenticCart enables ACP-compliant product feeds and checkout endpoints that can be submitted to OpenAI's Instant Checkout Programme.

How are trust, fraud, and approvals handled?

Scoped OAuth permissions, spending/category limits, SCA/3-D Secure where applicable, consent receipts, rate limiting, anomaly detection. Set approval thresholds (e.g., >$500 or age-restricted items require confirmation). Clear dispute paths.

What about returns and refunds?

Same policies as human purchases. Agents can initiate RMAs via API, attach consent/audit refs, and track refunds end-to-end.

Which protocols and payments are supported?

Agents can work with multiple API styles. AgenticCart enables ACP-compliant endpoints for OpenAI's Instant Checkout Programme. Payments via compliant processors (e.g., Stripe/PayPal). AP2 can cover card rails and A2A/open-banking under one mandate; MCP connects tools; A2A handles agent interoperability.

Will this work for my industry (and is it only subscriptions)?

Strong fit for repeat or low-ambiguity purchases (beauty, electronics, fashion, F&B, B2B). Not just subscriptions but also price-watch, replenishment, gifting, bundles.

What if an agent makes a mistake?

Audit trails show what was considered and why. Users can dispute via normal channels; agents improve quickly with feedback loops.

Do customers really trust agents with purchases?

Trust grows with transparency, scoped consent, clear limits, and easy reversibility. Start with low-risk use cases to build confidence.

Key Terms

Glossary

Essential terminology for understanding agentic commerce.

Agent
AI software that acts for a user: discovers products, evaluates options, and completes purchases end-to-end.
AgenticCart
WooCommerce plugin that enables ACP-compliant product feeds and checkout endpoints for OpenAI's Instant Checkout Programme.
Delegated Purchase
An agent completes a transaction using a user-authorized mandate with tokenized payment credentials held by a PCI-compliant PSP.
Autonomous Checkout
Checkout executed by an agent without human clicks beyond initial consent/limits; may invoke SCA/3-D Secure when required.
Machine Customer
An AI "buyer" acting as the customer of record (software, not a human).
Price Watch
Agent monitors price (and constraints like stock/shipping/total) and purchases when conditions are met.
Approval Flow
Human-in-the-loop confirmation for high-value/risk orders based on configurable thresholds.
Idempotency
Endpoint behavior ensuring retries don't duplicate orders/charges, typically via an idempotency key.
Consent Receipt (Proof of Consent)
Durable record tying a purchase to the user's authorization (e.g., OAuth grant ID/mandate, timestamp, scope).
Scopes (OAuth 2.0)
Granular permissions granted to the agent (e.g., "view:products", "order:≤$100").
Webhook
Signed, retryable callbacks (e.g., shipped/delivered/failed) that keep agents in sync.
Product Metadata
Structured attributes (title, specs, variants, media, price, stock) used for automated decisioning.
Sessionless Requests
Each API call carries auth/context (no fragile browser sessions); resources like carts remain server-side state.
Dispute / Chargeback Evidence
Package (consent, logs, delivery) used to contest chargebacks; not a guarantee of "protection."
Replenishment Agent
Agent that predicts consumption and reorders to avoid stockouts.
ACP (Agentic Commerce Protocol)
Merchant API surface (catalog/search, pricing, inventory, cart, checkout, webhooks). AgenticCart enables ACP for WooCommerce.
MCP (Model Context Protocol)
Standard interface connecting agents to tools/data.
A2A (Agent-to-Agent)
Interoperability/coordination between agents for multi-party flows.
AP2 (Agent Payments Protocol)
Signed mandates and multi-rail payments (cards + A2A/open-banking).
Mandate
The user's standing authorization specifying spend/merchant/category limits the agent must honor.
SCA / 3-D Secure
Strong Customer Authentication mechanisms that may be triggered during payment.
Tokenization
Replacing PANs with tokens so agents/merchants never handle raw card numbers.
PSP (Payment Service Provider)
Compliant processor (e.g., Stripe, PayPal) that stores payment credentials and processes charges.
Cart Abandonment
When a human starts checkout but doesn't complete. Agents reduce this dramatically because there's no multi-step form to abandon.
External Resources

Technical Resources

Authoritative specifications, documentation, and industry developments for implementing agentic commerce.

AI-Powered Commerce

Make Your WooCommerce Store
AI-Ready with AgenticCart

Transform your online store into an agentic commerce platform. Enable AI assistants to browse, recommend, and purchase products directly from your WooCommerce store.

14-day free trial
Cancel anytime