Agentic Commerce: Preparing for AI-Driven Transactions

The simple act of buying online is shifting from human-driven, manual transactions to autonomous, conversational exchanges between intelligent agents.

This new landscape of agentic commerce is not theoretical; it’s already taking shape through emerging protocols like OpenAI’s ACP (Agent-to-Commerce Protocol) and the foundations of full Agent-to-Agent (A2A) interoperability, largely led by Google.

Early adopters stand to benefit from what analysts project to be a $1.7 trillion opportunity by 2030, spanning both physical and digital goods. For brands and technology leaders, understanding these models and preparing their systems to participate in this new economy is now a strategic priority.

Understanding the Models

MCP: Model Context Protocol

MCP is a universal way for AI systems to connect to and use external tools. It's a standardized way for models like ChatGPT, Claude, or Gemini to access data, perform actions, and interact securely with external systems.

MCP bridges the gap between large language models and business systems. It allows AI to not just reason about information, but to take meaningful action. Every organization should be thinking about MCP-compatible APIs. This is the first step in making systems “AI-ready” and discoverable by intelligent agents.

A2A: Agent-to-Agent

A2A allows agents, built by different companies or operating in different ecosystems, to recognize, authenticate, and exchange information. It’s the framework that will let a buyer agent (for example, ChatGPT acting for a customer) and a seller agent (built by your organization) interact with one another enabling automated quoting, booking, and negotiation.

AP2: Agent Payment Protocol

AP2 is an extension of A2A that introduces mandates and cryptographic proof for secure transactions. In simpler terms, its how an agent confirms its authority to act on a user’s behalf. It ensures that when an agent buys something, it does so transparently and verifiably.

We can expect to see AP2 increasingly embedded into payment gateways, fraud systems, and digital wallets.

ACP: Agent-to-Commerce Protocol

ACP, which is an Agent-to-API protocol (not an Agent-to-Agent protocol) powers instant checkout directly within ChatGPT and other AI platforms. It connects buyer agents to merchant APIs using structured product feeds. If a merchant’s feed includes the enable_checkout flag, ChatGPT can complete the purchase instantly; fetching inventory, processing payment, and updating order status automatically.

ACP is the first live implementation of agentic commerce and it applies to both physical goods and time-based experiences like events and travel.

How to get started

The near-term play is to expand your sales channels to include selling inside conversational interfaces (ACP/instant checkout) like ChatGPT. The mid-term play is to operationalize true agent-to-agent (A2A) transactions by standing up a “seller agent” and a set of agentic workers across the order lifecycle.

We see this in phases:

Phase 1: What’s Possible Now (next 3 months)

If a good percentage of your audience resides in ChatGPT today, consider publishing an ACP-compatible product feed so AI agents can find and purchase directly from your catalog. You needn't expose your entire inventory at once; consider launching a limited pilot (a single product or experience) to measure conversion lift and gather user data. Use OpenAI's flags (enable_search and enable_checkout) to control whether a product is discoverable and/or purchasable inside ChatGPT. These fields do not affect how the product is displayed on your own site, they simply enable or disable the ChatGPT integrations.

Phase 2: Build a Seller Agent (next 12 months)

To be prepared for agent-to-agent commerce, it's recommended to develop a “seller agent,” capable of negotiating and completing orders/bookings via standardized APIs, as well as "agentic workers" to automate your most repetitive commerce tasks such that an order could be triggered out of business hours and still be fulfilled.

A good starting point is to conduct a data audit. Agents are only as good as the information they use, so a crucial first step is to develop a single, reliable source for all product details (prices, dates, inclusions, policies) and ensure this information is readily available.

As part of this effort, standardize the language you use; if one file says "Cancellation Fee" and another says "Penalty for Withdrawal," your agents will get confused. Agree on one term for every concept and apply it everywhere. Centralizing customer records is also imperative (past bookings, dietary needs, special requests).

From there, identify the repetitive, rules-based tasks that an agent can handle from start to finish. Identifying processes that involve multiple hand-offs between systems or people (e.g., a customer asks a question, Operations checks availability, Accounting sends an invoice) and focus your automation efforts on low-hanging fruit. Good first candidates for businesses such as Tour Operators might be:

1. Availability: "Is Tour X available for 12 people on these dates?"
2. Invoice Generation: "Generate a final invoice for booking #123."
3. Simple Customer Support: "What is the luggage allowance for this tour?"

Create an "Agent Card"

Every agent has a digital "Agent Card" that advertises its capabilities. It defines what your Seller Agent does. It's core capabilities should be things such as:

1. Tell a prospective guest what tours you offer.
2. Provide a final price for a specific group size and date.
3. Secure the reservation.
4. Modify an existing booking (e.g., add a passenger).

Create Agentic Workers

These are the internal specialists that the Seller Agent delegates its tasks to. They handle your important logistical and operational workflows. These are likely to include:

• An Availability Agent that manages real-time inventory (seats, rooms, guides). It eliminates overbooking errors and provides instant, accurate availability to the Seller Agent.
• A Logistics Agent might create the daily manifest, allocates guides, books local suppliers (e.g., transfers). This saves time for your Ops team and ensures no detail is missed during high-volume periods.
• And, a Payments Agent that manages payment schedules, sends reminders, and flags late payments.

Importantly, don't try to build all three at once. By way of example, start with the Availability Agent because it's the most rules-driven and provides instant, measurable value to your front-line sales.

Strategic Lens

The future of commerce is clearly conversational, automated, and agent-to-agent. Brands that build and adopt AI-ready systems, design for transparency, and partner within evolving AI ecosystems will stay visible and trusted as machines begin to buy, sell, and negotiate on our behalf. The shift has already started, the question is whether your business is ready to transact in it. Organizations that modernize their APIs, publish structured data, and begin experimenting with MCP, A2A, AP2, and ACP will not only remain discoverable in the next generation of AI interfaces but lead it.