Agentic Commerce
Agentic commerce is commerce in which an AI agent acts on behalf of a person or business across one or more stages of a shopping or purchasing journey, such as finding options, comparing them, coordinating with a merchant, preparing an order, or completing a transaction under defined authority. The term is broader than agentic payments: payment is one possible stage. It also does not require unrestricted autonomy. Current implementations can keep the user in control through explicit confirmation, scoped credentials, merchant acceptance, and recognizable agent-mediated transactions.
Origin and context
On 29 April 2025, Mastercard announced Agent Pay and described an agentic-commerce future involving tokenized credentials, registered agents, user-defined purchasing authority, and transactions recognizable across the payment chain. On 29 September, OpenAI announced Instant Checkout and the Agentic Commerce Protocol, allowing a user to proceed from product discovery to merchant checkout inside ChatGPT while explicitly confirming each step. McKinsey's October report used the same label for a wider intent-driven journey that can include research, comparison, negotiation, purchase, and coordination. These sources show convergence across payment, platform, and advisory organizations without supporting the base record's FIDO-origin attribution.
Why it matters
When software can advance a purchase rather than only recommend an item, product design must represent intent, authority, identity, price limits, merchant terms, confirmation, disputes, and audit evidence in machine-readable workflows. Merchants need to distinguish a trusted agent from abuse, while users need to understand what was proposed, approved, shared, and charged. This creates skill demand across agent design, commerce integration, authentication, payment operations, human-in-the-loop controls, and exception handling. It also changes discovery: an agent may compare offers or interact with merchant systems before a person visits a conventional storefront.
Example
A traveler asks an agent to find a refundable hotel within a stated budget. The agent compares eligible offers and prepares a booking with the merchant. Before purchase, it shows the hotel, dates, cancellation terms, total price, and payment method; the traveler confirms, the merchant accepts the order, and a scoped payment token is used. That is agentic commerce with human authorization. A list of hotel links with no ability to advance or coordinate the transaction is AI-assisted discovery, not the full pattern.
How it differs
Agentic AI
Agentic AI is the broader class of systems that pursue goals through multi-step actions. Agentic commerce applies that behavior to commercial journeys and introduces merchant, order, payment, consumer-control, and dispute requirements. Not every agentic system participates in commerce.
Agent Payments Protocol (AP2)
A payment protocol is one technical mechanism that may carry authorization or transaction information. Agentic commerce is the wider market and workflow category, spanning discovery through fulfillment and support. No single reviewed protocol defines the whole category.
Maturity and evidence
Maturity is rated 3. The exact label appears in dated materials from independent payment, AI-platform, and consulting organizations, and at least one reviewed service supported real merchant purchases with explicit confirmation. The category remains early: protocols, supported merchants, regions, transaction types, and delegation models are still evolving. The evidence does not justify maturity 4, universal interoperability, or claims that autonomous purchasing is already routine across commerce.
Limits and open questions
Announcements and forward-looking reports mix currently available features with planned capabilities and scenarios. A recommendation system, shopping chatbot, checkout integration, and independently acting procurement agent can all be marketed with similar language while granting very different authority. Teams should document which step the agent performs, what the user confirms, how credentials are scoped, who remains merchant of record, what data is shared, and how errors, fraud, returns, and disputes are handled. Market-size projections are intentionally excluded because they do not establish technical maturity or user outcomes.
Related terms
References
- Mastercard unveils Agent Pay, pioneering agentic payments technology to power commerce in the age of AIMastercard Newsroom · 2025-04-29 · class A
- Buy it in ChatGPT: Instant Checkout and the Agentic Commerce ProtocolOpenAI · 2025-09-29 · class A
- The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchantsMcKinsey & Company · 2025-10-17 · class B
Last updated: 2026-09-07
This term is also covered in the Skills Atlas as ai agent design skill.
This term is also covered in the Skills Atlas as human in the loop ai skill.
This term is also covered in the Skills Atlas as workflow orchestration skill.