Agentic commerce describes a shift from user-initiated digital shopping to AI-initiated shopping, where software agents perform tasks on behalf of customers and merchants. Instead of a person manually searching, comparing, selecting products, and checking out, an agent handles these steps autonomously based on preferences, context, and structured data.

Agentic commerce does not replace ecommerce. It changes who performs the work or "toil". The interface moves from screens and clicks to intent and delegation.

1. The Core Idea: Delegated Digital Work

In traditional ecommerce the user carries out every step. In agentic commerce:

• A customer agent evaluates options, interprets preferences, checks inventory, and selects products.

• A merchant agent provides structured data, prices, availability, shipping options, and payment capabilities.

• Payment and identity flows are handled through a standardised commerce protocol that agents can interpret.

• The user reviews or approves the outcome, or receives the result with minimal interaction.

This shifts the transaction from a manual sequence to a machine-to-machine workflow.

2. The Protocols That Enable Agentic Commerce

Agentic commerce relies on interoperable standards that allow agents to discover products, initiate payments, and interact with merchant systems. Five protocols matter most.

AP2: Discovery and Ranking

AP2 defines how merchants supply structured product information that agents can read and compare. It establishes a common language for:

• Product attributes

• Price and availability

• Variants

• Delivery options

• Metadata alignment across merchants

AP2 effectively becomes a new ranking system similar to SEO, where data quality determines visibility within agent results.

ACP: Checkout and Payments

ACP standardises how agents create carts, select SKUs, perform identity checks, and submit payments. It covers:

• Item selection

• Order construction

• Payment initiation

• Authentication

• Confirmation

Instead of bespoke integrations, merchants expose a consistent agent-friendly checkout surface.

Network Protocols: Approvals, Trust, Routing, and Identity

Card networks and domestic payment networks are modernising the underlying messaging standards that support agent-led transactions. This includes:

• Richer transaction metadata

• Stable identity tokens

• Continuous messaging rather than simple request-response patterns

• Multi-rail orchestration across cards, wallets, and account-to-account flows

• New authentication pathways that do not rely on device signals

Network upgrades enable secure approval decisions in an environment where agents, not human users, initiate transactions.

MCP: Tool Use and Enterprise Integration

The Model Context Protocol (MCP) allows agents to interact with business systems safely. It defines how models call tools, databases, and APIs through a secure and auditable layer. MCP is already being adopted as a standard for enabling:

• Inventory lookups

• Pricing checks

• Account actions

• Workflow automation

MCP is particularly important on the merchant side, where agents need controlled access to operational data.

A2A: Direct Bank-to-Bank Rails

Account-to-account payment systems provide an alternative to card rails for agent-initiated payments. They offer:

• Lower costs

• Near-instant settlement

• Strong identity frameworks

• Clear user consent flows

A2A will sit alongside card networks as a core option, and agents will choose between them based on cost, speed, approval likelihood, and incentives.

3. Customer Agents and Merchant Agents

Agentic commerce involves two interacting systems.

Customer Agents

Customer agents identify what the user wants and evaluate options based on:

• Price

• Preferences

• Delivery speed

• Availability

• Merchant reputation

• Historical behaviour

They produce a ranked set of outcomes and may act directly depending on the user’s level of delegation.

Merchant Agents

Merchant agents respond with:

• Structured product data

• Accurate pricing

• Real-time stock information

• Fulfilment options

• Payment methods and routing preferences

• Identity and token capabilities

Their role is to ensure the customer agent receives predictable, high-quality information that supports the best outcome.

4. Why Agentic Commerce Matters for Merchants

a. Higher Approval Rates

Agents send cleaner, more predictable data, and network protocols provide more stable identity signals. This can improve:

• Authorisation rates

• Fraud detection

• Challenge rates

• End-to-end conversion

b. Smarter Payment Routing

Agents allow dynamic selection of the optimal payment method based on:

• Likelihood of approval

• Cost to serve

• Network incentives

• Customer preferences

This creates a natural alignment between approvals and cost optimisation.

c. Better Cost Efficiency

Standardised protocols let merchants influence how payments are made. For example:

• Steering to cheaper payment methods

• Adjusting incentives in real time

• Choosing the lowest-cost capable rail

d. Migration From Device-Based Signals to Network-Based Signals

As agents replace browsers, traditional device and behavioural signals disappear. Fraud and identity systems shift to:

• Network-level metadata

• Token quality

• Historical consistency

• Structured authentication flows

This requires merchants to invest in data consistency and robust identity frameworks.

5. How the Agentic Commerce Stack Is Forming

The emerging architecture can be expressed in five layers.

Layer 1: Agents

ChatGPT, Gemini, Claude, merchant-built systems.

Layer 2: Protocols

AP2 for discovery
ACP for checkout
Network protocols for approvals and trust
A2A as an alternative settlement pathway
MCP for tool use and system integration

Layer 3: Merchant Systems

Product data, pricing engines, orchestration, fraud tools, gateways, and settlement platforms.

Layer 4: Data Quality

Structured product data
Pricing consistency
Token and identity integrity
Inventory accuracy

Layer 5: Governance

Transparent logic
Auditability
Rules and human oversight
Security and compliance frameworks

6. What Agentic Commerce Is Not

Agentic commerce is not:

• A chatbot

• A UI redesign

• A new payment method

• A marketing channel

• A browser extension

It is the automation of the work that currently sits between intent and purchase.

Final Definition

Agentic commerce is a model where AI agents perform discovery, evaluation, and checkout on behalf of customers and merchants using standardised protocols that make transactions predictable, secure, and optimisable.