A Simple Guide to Agentic Commerce

Agentic commerce is a model in which AI agents autonomously execute purchases on behalf of users - searching, comparing and completing transactions without continuous human input. This isn't chatbots suggesting products or recommendation engines nudging you toward purchases. AI agents in agentic commerce handle the entire workflow: understanding your intent, querying product catalogs across retailers, evaluating options and executing payment. You delegate the task; the agent completes it. This represents a fundamental shift from human browsing to machine reasoning. Traditional e-commerce is built for human eyes - product pages designed for visual appeal, persuasive copy, conversion-optimized layouts. Agentic commerce is built for machine parsing - structured data, API endpoints and machine-readable catalogs that agents query and compare at scale. The agent doesn't see your product page. It reads your API response. The term "simple" in this guide isn't about oversimplification - it's about clarity. Agentic commerce involves protocols, APIs, structured data, payment delegation and authorization frameworks. These are technical concepts, but they're not inaccessible. This guide explains what agentic commerce is, how it works technically, why it differs from traditional e-commerce and what businesses must do to prepare. By the end, you'll understand the core mechanisms, the infrastructure required and the timeline for adoption - whether you're a merchant, developer, finance professional, or consumer evaluating whether to delegate purchasing authority to an AI agent.

What Is Agentic Commerce? (Definition and Core Concept)

Agentic commerce is a model in which AI agents autonomously search, compare, negotiate and purchase products on behalf of users without requiring continuous human input. Unlike chatbots that suggest options or recommendation engines that nudge you toward products, AI agents in agentic commerce execute complete transactions based on delegated authority.

The core distinction lies in delegation versus assistance. When you use a traditional shopping assistant, you remain in the loop - reviewing suggestions, clicking through pages and completing checkout yourself. With agentic commerce, you delegate the entire process. You might tell an agent, "Get me camping gear for next week, budget £500," and the agent handles product discovery, cross-retailer comparison, vendor selection and payment - reporting back only when the transaction is complete.

This autonomy requires three defining characteristics:

• Multi-step reasoning: Agents break down complex requests into searchable parameters, evaluate trade-offs and make decisions across multiple steps.
• Cross-platform operation: Agents query multiple retailers, marketplaces and vendors simultaneously, comparing options that no single human could efficiently evaluate.
• Transaction completion: Agents don't just recommend - they execute purchases using pre-approved payment credentials and authorization rules.

What agentic commerce is not helps clarify the boundary. It's not a recommendation engine showing you "customers also bought" suggestions. It's not a chatbot answering product questions. It's not a voice assistant adding items to your cart for manual checkout. These tools assist; they don't execute. Agentic commerce shifts the human role from operator to supervisor - from browsing and clicking to setting boundaries and auditing outcomes.

The mental model that best captures this shift: from human browsing to machine reasoning. Traditional e-commerce is built for human eyes - product pages designed for visual appeal, persuasive copy, conversion-optimized layouts. Agentic commerce is built for machine parsing - structured data, API endpoints and machine-readable catalogs that agents can query and compare at scale. The agent doesn't see your product page. It reads your API response.

How Agentic Commerce Works: Step-by-Step Process

Understanding how an agent transaction unfolds reveals why agentic commerce requires fundamentally different infrastructure than traditional e-commerce. Here's the complete workflow from user intent to completed purchase:

Step 1: User Delegates Intent to Agent

The process begins when a user expresses a purchasing goal to an AI agent. This isn't a search query - it's a delegated task with context and constraints. Example prompts might include:

• "Reorder office supplies from approved vendors, prioritise next-day delivery"
• "Buy noise-canceling headphones when the price drops below £250"
• "Get me camping gear for next week, budget £500"

The user provides intent, constraints (budget, timing, preferences) and authorization (pre-approved spending limits, vendor whitelists, category restrictions). The agent now has both the goal and the boundaries within which to operate.

Step 2: Agent Reasoning and Query Planning

The agent breaks down the user's intent into structured, searchable parameters. For "camping gear for next week, budget £500," the agent might identify:

• Product categories: tent, sleeping bag, portable stove, headlamp
• Constraints: total cost ≤ £500, delivery within 7 days
• Optimization goals: maximise quality within budget, prioritise items user doesn't already own

This reasoning step is where AI agents differ from keyword search. The agent constructs a multi-part query plan, anticipating that it will need to compare products across categories and retailers to meet the user's goals.

Step 3: Multi-Site Product Discovery via APIs

The agent queries product catalogs across multiple retailers - not by scraping web pages, but by accessing structured APIs. Using protocols like the Universal Commerce Protocol (UCP), the agent retrieves machine-readable product data: specifications, availability, pricing, delivery options.

For each product category, the agent might query five retailers simultaneously, receiving structured responses that include:

• Product attributes (weight, materials, specifications)
• Real-time inventory status
• Current pricing and any active promotions
• Delivery timelines and costs

This is where traditional e-commerce infrastructure breaks down. If your product catalogue isn't exposed via API with structured data, agents can't discover it. Visual product pages, persuasive copy and conversion-optimized layouts are invisible to agents.

Step 4: Cross-Retailer Comparison Using Structured Data

The agent compares options across retailers using machine-readable attributes. For a tent, it might compare:

• Capacity (2-person vs 4-person)
• Weight (important for backpacking)
• Waterproof rating
• Price per feature set
• Delivery speed

Because the agent has structured data from multiple sources, it can perform comparisons no human could efficiently execute - evaluating dozens of products across multiple dimensions in seconds. The agent selects products that collectively meet the user's budget and timing constraints while optimizing for quality.

Step 5: Authorization Check and Spending Limit Validation

Before executing any transaction, the agent validates against the user's pre-approved authorization rules:

• Is the total cost within the user's spending limit?
• Are the selected vendors on the user's approved retailer list?
• Do the product categories fall within allowed purchase types?

If the transaction exceeds pre-approved limits - say, the optimal camping gear set costs £520 instead of £500 - the agent either requests user approval or adjusts the selection to meet constraints. This authorization layer is what makes autonomous purchasing safe and controllable.

Step 6: Payment Delegation and Transaction Execution

Once authorized, the agent executes the purchase using delegated payment credentials. This isn't the agent "knowing" your credit card number - it's using tokenized payment methods and protocols like the Agent Payments Protocol (AP2) to initiate transactions on your behalf.

The agent places orders with multiple retailers if needed, coordinating delivery timelines and confirming that all items will arrive within the user's "next week" constraint. Payment providers like Stripe and Visa are building agent-specific APIs that enable this delegation while maintaining security and fraud protection.

Step 7: Fulfillment Tracking and User Notification

After purchase, the agent monitors fulfillment status and notifies the user with a summary:

• Items purchased and from which retailers
• Total cost and breakdown
• Expected delivery dates
• Tracking information

The user receives a complete transaction record, including the agent's reasoning: why these products were selected, what trade-offs were made and how the budget was allocated. This transparency is essential for trust and auditability.

Real-World Agent Transaction Examples

Example 1: B2B Procurement
User prompt: "Reorder office supplies from approved vendors, prioritise next-day delivery"
Agent reasoning: Queries inventory management system to identify low-stock items (printer paper, toner cartridges, pens). Checks approved vendor list (three office supply companies). Compares pricing and delivery options across vendors. Vendor A has lowest price but 3-day delivery; Vendor B has next-day delivery at 8% premium. Agent selects Vendor B to meet priority constraint.
Outcome: Order placed for £340 worth of supplies, next-day delivery confirmed, user notified with itemized list and vendor selection reasoning.

Example 2: Subscription Optimization
User prompt: "Cancel unused streaming services, keep only those I used in last 30 days"
Agent reasoning: Accesses user's subscription management data and usage logs. Identifies five active subscriptions: Netflix (used 12 times), Spotify (used daily), Disney+ (zero usage), Apple TV+ (used once), HBO Max (zero usage). Checks cancellation policies for unused services (Disney+ and HBO Max have no cancellation fees; Apple TV+ requires 24-hour notice).
Outcome: Three subscriptions cancelled (Disney+, Apple TV+, HBO Max), saving £28/month. User receives summary with usage data and cancellation confirmations.

Example 3: Price Monitoring and Opportunistic Purchasing
User prompt: "Buy noise-canceling headphones when price drops below £250"
Agent reasoning: Sets up continuous monitoring across five electronics retailers. Tracks pricing APIs for three headphone models matching user's specifications (active noise cancellation, over-ear, wireless). After six days, detects price drop at Retailer C: Sony WH-1000XM5 reduced from £279 to £239.
Outcome: Purchase executed automatically at £239, user notified of £40 savings and 2-day delivery timeline.

Agentic Commerce vs. Traditional E-Commerce: What's Changed

The shift from traditional e-commerce to agentic commerce isn't incremental - it's architectural. Understanding what changes reveals why businesses optimized for human conversion will struggle in an agent-driven environment without significant infrastructure updates.

Why Traditional SEO Becomes Less Relevant

In traditional e-commerce, search engine optimization focuses on ranking product pages for human search queries. Businesses invest in keyword research, content optimization, backlink building and technical SEO to appear in Google's top results. The goal: get humans to click through to your site.

In agentic commerce, AI agents don't click search results. They don't browse product pages. They don't care about your site's visual design or persuasive copy. Agents query APIs directly, parse structured data and compare machine-readable catalogs. The optimization target shifts from human eyes to AI data parsing - a discipline some are calling Agent Engine Optimization (AEO).

Traditional SEO metrics like click-through rate, dwell time and bounce rate become irrelevant. Agents don't "dwell" on pages or "bounce" from sites. What matters instead:

• API response speed: How quickly can your catalogue respond to agent queries?
• Structured data completeness: Are all product attributes exposed in machine-readable format?
• Schema accuracy: Does your markup correctly represent product specifications, availability and pricing?
• Protocol compliance: Have you implemented standards like UCP that agents use for product discovery?

The Conversion Layer Transforms from Persuasion to Trust

Traditional e-commerce conversion optimization focuses on persuading humans to complete purchases: urgency signals ("Only 3 left in stock!"), social proof ("10,000+ reviews"), visual merchandising and frictionless checkout flows. These tactics work because humans make emotional, often impulsive purchasing decisions.

AI agents don't respond to urgency signals or social proof. They evaluate products based on structured criteria: specifications, pricing, availability, delivery timelines and merchant trust signals. The new conversion layer is trust and authorization:

• Is the merchant verified and trustworthy?
• Are return policies clearly stated in structured format?
• Does the transaction meet the user's authorization rules?
• Is the pricing transparent and competitive?

Merchants must prove legitimacy to agents, not just humans. This means exposing trust signals in machine-readable format: verified business credentials, structured return policies, transparent shipping terms and reputation data that agents can parse and evaluate.

The Technical Infrastructure: Protocols, APIs and Structured Data

Agentic commerce requires a new technical foundation - one built for machine-to-machine interaction rather than human browsing. Three components form this infrastructure: commerce protocols, API-first architecture and structured data standards.

Why Protocols Matter: The Common Language for Agent-Commerce Interaction

Protocols establish standardized ways for AI agents to discover products, access context and execute transactions across different platforms and retailers. Without protocols, every agent would need custom integrations with every merchant - an impossibility at scale. Protocols enable interoperability: one agent can transact with thousands of merchants using a common language.

Three protocols are emerging as foundational infrastructure:

Universal Commerce Protocol (UCP)

Developed by Google and Shopify, UCP defines how AI agents discover and access product data across e-commerce platforms. It standardizes:

• Product catalogue structure (attributes, specifications, categories)
• Inventory and availability data
• Pricing and promotions
• Delivery options and timelines

UCP ensures that when an agent queries "noise-canceling headphones under £250," it receives consistent, machine-readable responses from every UCP-compliant merchant. The protocol handles product discovery - the equivalent of a human browsing a product page, but structured for machine parsing.

For merchants, implementing UCP means exposing your product catalogue via API endpoints that conform to the protocol's data schema. This isn't a website redesign - it's infrastructure work: building API endpoints, mapping your product data to UCP's schema and ensuring real-time inventory sync.

Model Context Protocol (MCP)

Created by Anthropic, MCP is an open standard that gives AI agents a universal way to connect to external tools and data sources in real time. In agentic commerce, this means:

• Connecting to a retailer's inventory, pricing and order systems through a common interface
• Pulling in live data the agent needs to reason about a purchase
• Letting one agent integrate with many systems without bespoke connectors for each

If a user tells an agent, "Reorder what I bought last month but upgrade the shipping," MCP is what lets the agent reach into the retailer's order and product systems to retrieve that history and act on it. MCP handles the connection between the agent and external data; it is not itself a payments or authorization layer.

Agent Payments Protocol (AP2)

AP2 handles transaction authorization and payment delegation. Developed by Google with more than 60 partner organizations, it defines:

• How users grant agents payment authority through cryptographically signed "mandates"
• Authorization rules, spending limits and how long that authority lasts
• Transaction approval workflows
• Security and fraud prevention mechanisms

Payment partners including Mastercard, American Express, PayPal and Adyen back AP2, which lets agents initiate transactions while respecting user-defined spending constraints. A separate checkout protocol, the Agentic Commerce Protocol (ACP) — built by OpenAI and Stripe and used in ChatGPT's Instant Checkout — handles the agent-to-merchant checkout flow. Together, AP2 (payment authorization) and ACP (checkout) form the technical implementation of the trust layer.

Structured Data: Making Products Machine-Readable

Protocols define the rules; structured data is the content agents consume. Structured data uses standardized schemas (like Schema.org) to mark up product information in machine-readable format:

• Product attributes: Name, brand, model, specifications, materials, dimensions
• Pricing: Current price, currency, any active promotions or discounts
• Availability: In stock, out of stock, backorder status, quantity available
• Delivery: Shipping options, delivery timelines, costs, restrictions
• Returns: Return policy, return window, conditions, refund process

Structured data allows agents to parse and compare products without human interpretation. A human can look at a product page and infer that "lightweight" means the tent weighs 2kg. An agent needs the weight explicitly stated in a structured field: "weight": {"value": 2, "unit": "kg"}.

Implementing structured data means adding Schema.org markup to your product pages and exposing the same data via API. Many e-commerce platforms (Shopify, WooCommerce, BigCommerce) support structured data markup, but full agent readiness requires API-first exposure - agents need to query your catalogue programmatically, not scrape your website.

APIs as the New Storefront

In traditional e-commerce, your storefront is your website - the visual interface where humans browse and buy. In agentic commerce, your storefront is your API - the programmatic interface where agents query and transact.

Agent-ready businesses expose core commerce functions via API endpoints:

• /products: Returns product catalogue with full attributes, specifications and availability
• /inventory: Provides real-time stock levels and availability status
• /pricing: Returns current pricing, promotions and any dynamic pricing rules
• /checkout: Accepts transaction requests and returns order confirmation

These APIs must be fast, reliable and well-documented. Agents won't tolerate slow response times or inconsistent data. If your API returns outdated inventory information, agents will route transactions to competitors with accurate real-time data.

Payment Systems and Trust Layers: How Agents Transact Safely

The most common concern about agentic commerce is also the most fundamental: how can autonomous agents make purchases without compromising security or user control? The answer lies in payment delegation frameworks and multi-layered authorization systems.

The Trust Challenge: Autonomy Without Risk

For agentic commerce to work, users must grant AI agents access to payment credentials. This creates an inherent tension: agents need enough authority to execute transactions autonomously, but users need assurance that agents won't overspend, make unauthorized purchases, or be exploited by bad actors.

The solution isn't binary trust - it's trust as a designed system. Payment delegation frameworks allow users to grant limited, rule-bound authority that agents can use within defined boundaries. The system is built on three pillars: authorization rules, audit transparency and revocation controls.

Payment Delegation Mechanisms

Rather than giving agents direct access to credit card numbers or bank accounts, payment delegation uses tokenized methods and pre-approved credentials:

• Tokenized payment methods: Agents receive temporary tokens that represent payment authority without exposing actual account details. If a token is compromised, it can be revoked without changing underlying payment information.
• Spending limits: Users set maximum transaction amounts (per purchase, daily, monthly) that agents cannot exceed without requesting approval.
• Merchant restrictions: Users can whitelist approved vendors or blacklist specific merchants, ensuring agents only transact with trusted retailers.
• Category constraints: Users can limit agent purchases to specific product categories (groceries, office supplies) while blocking others (luxury goods, electronics).

Payment providers like Stripe are building agent-specific APIs that implement these delegation mechanisms. When an agent initiates a transaction, the payment API validates against the user's authorization rules before processing. If the transaction violates any rule - exceeds spending limits, uses an unapproved vendor, falls outside allowed categories - the payment is blocked and the user is notified.

Authorization Frameworks: User-Defined Rules

Users configure authorization rules that define what agents can and cannot do. These rules act as guardrails, allowing autonomy within safe boundaries:

• Pre-approved auto-execution: Transactions below a certain threshold (e.g., £50) that meet all other criteria are executed automatically without user approval.
• Approval-required transactions: Purchases above the threshold trigger a notification requiring user approval before execution.
• Recurring order rules: For repeat purchases (office supplies, household consumables), users can set "always approve" rules that allow agents to reorder without intervention.
• Time-based restrictions: Users can limit agent purchasing to specific times (business hours only, weekdays only) or disable agent access entirely during certain periods.

These frameworks mirror how businesses handle expense approvals and procurement rules. The difference is that rules are enforced programmatically by the payment system, not manually by finance teams.

Audit and Transparency: Seeing What Agents Do

Trust requires transparency. Users need visibility into what agents are doing, why decisions were made and how money is being spent. Agent transaction systems provide:

• Transaction logs: Complete records of every agent-initiated purchase, including what was bought, from which merchant, at what price and when.
• Agent reasoning trails: Explanations of why the agent selected specific products or vendors - what criteria were prioritized, what trade-offs were made.
• Real-time notifications: Instant alerts when agents execute purchases, request approval, or encounter errors.
• Spending analytics: Dashboards showing spending patterns, category breakdowns and comparisons to budgets or historical spending.

This transparency serves two purposes: it builds user confidence in agent decision-making and it enables users to refine authorization rules based on actual agent behaviour. If an agent consistently selects vendors with slower delivery, the user can adjust priorities to favor speed over cost.

Revocation and Correction: Maintaining Control

Users must be able to revoke agent access instantly and correct mistakes without friction:

• Instant revocation: Users can disable agent payment authority immediately, blocking all pending and future transactions until access is re-enabled.
• Transaction disputes: If an agent makes an unwanted purchase, users can flag it for review and initiate refunds or returns through standard merchant processes.
• Rule adjustments: Users can modify authorization rules at any time - tightening spending limits, adding vendor restrictions, or changing category permissions.

The goal is to make agent delegation feel safe and reversible. Users should feel confident that they can grant authority, monitor outcomes and pull back control if needed - without complex processes or delayed responses.

Trust as the New Conversion Layer

In traditional e-commerce, merchants optimise for human trust through visual signals: professional design, customer reviews, security badges, clear return policies. In agentic commerce, trust must be machine-readable. Agents evaluate merchant trustworthiness using structured data:

• Merchant verification: Is the business registered and verified by trusted authorities?
• Return policies: Are return terms clearly stated in structured format (return window, conditions, refund process)?
• Reputation data: What are the merchant's transaction volumes, dispute rates and resolution times?
• Security compliance: Does the merchant meet payment security standards (PCI DSS) and data protection regulations?

Agents prioritise merchants with strong trust signals. A merchant with verified credentials, clear return policies and low dispute rates will win agent transactions over a competitor with incomplete or opaque data - even if the competitor has lower prices. Trust becomes the conversion layer because agents optimise for user satisfaction and risk minimization, not just cost.

Market Outlook and Adoption Timeline

Agentic commerce is transitioning from concept to reality, but mainstream adoption remains years away. Understanding the adoption timeline, market forecasts and leading indicators helps businesses decide when to invest and what signals to watch.

Market Forecasts: How Big Will Agentic Commerce Become?

Industry analysts project significant growth for agentic commerce over the next decade:

• McKinsey: Estimates agentic commerce could generate around $1 trillion in US retail revenue, and $3-5 trillion globally, by 2030
• Morgan Stanley: Projects that 10-20% of US e-commerce sales will be agent-driven by 2030
• Gartner: Forecasts that AI agents will intermediate more than $15 trillion in B2B spending by 2028, and that around 20% of digital commerce transactions will run through AI platforms or agents by 2030

These forecasts suggest agentic commerce will capture a meaningful share of total e-commerce, but not replace traditional models entirely. Human-driven shopping will remain dominant for discretionary, high-involvement purchases (fashion, home decor, gifts), while agents will excel at routine, rules-based and optimization-heavy purchasing (groceries, supplies, travel, subscriptions).

Phased Adoption Timeline: When Will Agentic Commerce Go Mainstream?

2025-2026: Protocol Launches and Early Adopters

The foundational infrastructure for agentic commerce is being built now. Key milestones in this phase:

• Google and Shopify launch Universal Commerce Protocol (UCP)
• Anthropic expands Model Context Protocol (MCP) adoption
• Payment providers (Stripe, Visa, Mastercard) release agent payment APIs
• First agent-native commerce startups emerge
• Major platforms (OpenAI, Google, Amazon) announce agent commerce capabilities

Leading indicators to watch: UCP adoption announcements from major retailers, MCP integration by agent platforms, payment provider agent API launches, early agent transaction volume data.

2027-2028: Platform Integration and Infrastructure Buildout

Mainstream platforms begin exposing catalogs via agent-accessible APIs. E-commerce infrastructure providers add agent readiness features. Key milestones:

• Amazon and Shopify expose full product catalogs via UCP-compliant APIs
• WooCommerce, BigCommerce and other platforms add native agent support
• Payment delegation becomes standard feature in payment gateways
• Agent transaction volume reaches 1-5% of total e-commerce
• First consumer-facing AI shopping agents gain significant user bases

Leading indicators to watch: Platform announcements of agent API availability, percentage of major retailers with UCP implementation, agent transaction volume growth rates, consumer agent adoption surveys.

2029-2030: Mainstream Adoption and Market Maturity

Agentic commerce becomes standard practice for routine purchasing. Consumers expect agent options for common transactions. Key milestones:

• Agent transactions exceed 15% of e-commerce volume
• Majority of households use AI agents for at least some purchasing
• Traditional e-commerce platforms add agent-first interfaces
• AEO (Agent Engine Optimization) becomes standard practice alongside SEO
• Regulatory frameworks for agent commerce are established

Leading indicators to watch: Agent transaction volume crossing 10% threshold, consumer surveys showing majority adoption, regulatory proposals addressing agent commerce, shift in platform investment priorities from human UX to agent APIs.

What Could Accelerate or Delay Adoption?

Accelerators:

• Major platform announcements (Google, Amazon, Apple launching consumer agent commerce features)
• Breakthrough agent capabilities (significant improvements in reasoning, reliability, or user experience)
• Economic pressures driving demand for optimization (recession, inflation increasing focus on price comparison and savings)
• Protocol standardization reducing implementation friction

Inhibitors:

• High-profile agent failures or security breaches eroding consumer trust
• Regulatory restrictions on autonomous transactions or payment delegation
• Slow protocol adoption by major retailers creating limited agent inventory
• Consumer resistance to delegating purchasing authority

Industry-Specific Adoption Curves

Not all sectors will adopt agentic commerce at the same pace. Adoption will follow predictable patterns based on purchase characteristics:

Early adoption (2026-2027):

• B2B procurement (routine, rules-based, high repetition)
• Subscription management (clear optimization criteria, low risk)
• Digital goods and services (instant delivery, no physical logistics)
• Household consumables (frequent repurchase, low involvement)

Mid-stage adoption (2028-2029):

• Travel and booking (complex but structured, high optimization value)
• Electronics and appliances (specification-heavy, comparison-intensive)
• Grocery and meal planning (routine but requires preference learning)

Late adoption (2030+):

• Fashion and apparel (high personalization, fit uncertainty)
• Home decor and furniture (aesthetic judgment, high involvement)
• Gifts and special occasion purchases (emotional significance, personalization)