What is agent-to-agent (A2A)? A plain-English guide for business owners

Agent-to-agent (A2A) is when a customer's AI assistant talks directly to a business's own AI agent to ask questions, compare options, and book or buy something, with the customer approving the final choice. Instead of a person clicking through websites and phoning around, two pieces of software have the conversation, and the person makes the decision.

This guide explains agent-to-agent in plain English: what it is, how it works, which standards make it possible, what's real today and what's still emerging. It's written for business owners, not developers. Where we talk about the future, we say so clearly.

What is agent-to-agent, in one sentence?

Agent-to-agent means your customer's AI assistant can talk to your business's AI agent directly, the same way two people might talk at a front desk.

Today, most people find a business by searching, reading a few websites, and maybe calling. More and more, people are instead asking an AI assistant such as ChatGPT, Gemini, Claude or Siri to do that work for them. The assistant reads the web, summarizes options and makes a recommendation.

Agent-to-agent is the next step. Rather than only reading your website, the customer's assistant can ask your business's own agent questions and get accurate answers straight from the source: "Do you have Saturday openings?", "Is this covered under warranty?", "Can you deliver to this postal code by Friday?" Your agent answers from your real business information, and when the customer is ready, the two agents can complete a booking, quote or purchase.

The short version: AI assistants are becoming your customers' front door, and agent-to-agent is how they knock.

What does agent-to-agent look like in real life?

Picture a customer asking their assistant to book a dental cleaning, and the assistant sorting out insurance, availability and timing with the clinic's agent in a few seconds.

Maria tells her phone's assistant: "Book me a teeth cleaning in the next two weeks, near work, after 4pm. My dental plan is through work."

1. The assistant shortlists clinics. It looks for dental clinics near Maria's office whose facts are clear: location, hours, services, and whether they accept her insurance.
2. It asks each clinic's agent. For clinics that publish a business agent, the assistant asks directly: "Do you have a hygiene appointment after 4pm in the next two weeks? Do you bill my insurer directly? What's the cost for a new patient cleaning?"
3. The clinic's agent answers from real information. "Yes, we bill most workplace dental plans directly. We have Tuesday the 18th at 4:30pm or Thursday the 20th at 5:15pm. New patient cleaning and exam is typically covered; any balance is explained before treatment."
4. Maria approves. Her assistant shows her the two best options. She taps Tuesday.
5. The booking is made. The clinic's agent confirms the appointment, sends intake forms, and puts it in the clinic's calendar. The front desk sees a new booking with all the details.

Notice what happened. The clinic that had clear, structured facts and a working agent got the booking. A clinic with an outdated website and no agent might never have been considered, even if it was closer.

What's the difference between an AI assistant and a business agent?

An AI assistant works for the customer; a business agent works for your business. Agent-to-agent is the conversation between the two.

A general AI assistant can only guess about your business from what's on the web. A business agent knows the details: your real price ranges, your actual service area, what's covered under warranty, which days you're booked solid. That's why the two working together is so useful.

How does agent-to-agent work?

It works in three parts: the business publishes an agent and clear structured facts, shared standards let agents understand each other, and the customer's assistant asks, checks and acts within limits the customer sets.

1. The business publishes an agent and structured facts

First, your business needs to be readable by machines. That means your key facts (services, prices or ranges, hours, areas served, policies and FAQs) are published clearly and consistently, using schema markup. Schema markup is a standard vocabulary from Schema.org that labels information on a web page so software knows "this is a price", "these are opening hours" and "this is a service area". Google explains how it uses this in its structured data documentation.

Then your business publishes its own agent, along with an agent card. An agent card is a small public file that acts like a digital business card for your agent: its name, what it can do, and how to reach it. Under the A2A specification, it lives at a standard address on your website, /.well-known/agent-card.json, so other agents know exactly where to look (A2A agent discovery).

2. Shared standards let agents talk

Agents built by different companies need a common language, the same way email works no matter which email app you use. That's what open standards (also called protocols) do. They define how agents introduce themselves, ask questions, hand off tasks, and complete purchases safely. We explain the main ones below.

3. The customer's assistant asks, checks and acts within limits

Finally, the customer's assistant does the legwork. It finds suitable businesses, reads their agent cards, asks their agents questions, compares the answers, and brings the best options back to the customer. For anything that matters (spending money, booking time, sharing personal details), the customer approves. Payment standards are being designed around exactly that kind of verifiable approval.

Which standards make agent-to-agent possible?

Four open standards matter most for businesses right now: Google's Agent2Agent (A2A) for agents talking to agents, Anthropic's Model Context Protocol (MCP) for connecting AI to tools and data, OpenAI and Stripe's Agentic Commerce Protocol (ACP) for checkout, and Google's Agent Payments Protocol (AP2) for payments.

You don't need to understand the technical details. You just need to know what each one is for, and that they're open: no single company owns the conversation. Status notes below were checked against official sources in November 2026; this space moves quickly.

You'll also hear about the Universal Commerce Protocol (UCP), an open commerce standard co-developed by Google and Shopify that is designed to work alongside A2A, AP2 and MCP (Google). In November 2026 Google began rolling out UCP-powered checkout to shoppers in Canada on AI Mode and the Gemini app (Google Canada).

How do A2A and MCP fit together?

MCP connects an AI agent to tools and data; A2A connects one agent to another agent. Most business agents will use both.

Think of your business agent as a new front-desk employee. MCP is how that employee reaches the calendar, the price list and the inventory system. A2A is how that employee talks to visitors' assistants at the door. We wrote a short explainer on this: A2A vs MCP: what a business owner needs to know.

What's real today, and what's still emerging?

The standards are real and the groundwork pays off today; everyday customers using assistants to book through business agents is still early and growing.

We don't publish adoption numbers we can't verify. Where we've cited figures above, they come directly from the organizations that run each standard.

Is agent-to-agent safe? Who stays in control?

The customer stays in control of decisions, and your business stays in control of what its agent can say and do.

On the customer's side, assistants are designed to act within limits the customer sets and to ask for approval on things that matter, like spending money. Payment standards such as AP2 record that approval in a verifiable way.

On your side, a well-built business agent follows clear rules: what topics it covers, which prices it can quote, what it should never promise, and when to hand the conversation to a real person. It answers only from information you've approved. That's a big part of what we set up for you.

What should a business do about agent-to-agent right now?

Start with what pays off today: make your business facts clear and machine-readable, then add your own agent on your website, built on open standards.

1. Find out where you stand. Ask a few AI assistants about businesses like yours and see what they say. Or take our 2-minute readiness check.
2. Publish clear facts. Services, prices or ranges, hours, areas served, policies and FAQs, consistent everywhere, with schema markup.
3. Answer real questions. Write direct answers to the questions customers actually ask, near the top of each page.
4. Make it easy to act. Online booking, quote requests or checkout, so there's something for an assistant to do.
5. Add your own agent. Trained on your real information, live on your website first, with a published agent card.
6. Keep it current. Outdated facts are worse than none; an assistant may repeat them confidently.

That's exactly what our done-for-you services cover, in that order.

What happens if a business waits?

Assistants recommend and book the businesses they can read and talk to, so waiting risks being quietly left off the shortlist.

There's no single day when agent-to-agent "switches on". It grows gradually, one assistant feature and one platform at a time. Businesses with clear facts and a working agent get included as each new capability arrives. Businesses without them may never know what they missed, because the customer never saw them. See examples by industry in what agent-to-agent means for your business.

What do the key terms mean?

AI assistant

Software like ChatGPT, Gemini, Claude or Siri that helps a person find information and get things done.

AI agent / business agent

Software that can take actions toward a goal. A business agent represents your business and answers from your real information.

Agent card

A public file describing an agent: its name, skills and how to reach it. In A2A, found at /.well-known/agent-card.json.

Protocol / standard

An agreed set of rules that lets software from different companies work together.

Schema markup

Labels added to web pages so machines know what each piece of information means.

llms.txt

A proposed plain-text file at the root of a website that summarizes the site for AI tools (llmstxt.org).

AEO (answer engine optimization)

Making your business easy for AI answer engines to understand, trust and cite.

Want to see these in action? This site is agent-ready, and we explain exactly how.