Browse

AI agents, explained

Agent-to-agent communication, explained for business owners.

Agent-to-agent communication is how one AI agent hands a job to another AI agent, even when different companies built them. An open standard called A2A lets agents find each other, ask for work and hand back results without a custom connection for every pair. A second one, MCP, connects an agent to its tools. It is early. This page says what is real, with dates, and what a service business should do now.

Get ahead

Free strategy call.

What changes

You will know what A2A and MCP are, what is real today and what is not, and what to do now, whether or not you ever build anything.

What is agent-to-agent communication?

Agent-to-agent communication is a way for AI agents to find each other, ask each other to do a job and hand back the result. Inside one business it can already look like this: one agent reads a request, another drafts the quote and a third books the visit. Between companies, an open standard called A2A lets an agent from one company work with an agent from another, without a custom connection built for every pair.

See the agents we build and run

At a glance

The two standards to know
A2A connects an agent to other agents. MCP connects an agent to tools and data.
Who published them
MCP: Anthropic, November 25, 2024. A2A: Google, April 9, 2025.
Who looks after them now
The Agentic AI Foundation, part of the Linux Foundation. MCP joined on December 9, 2025. A2A joined in August 2026.
How big it is
The Linux Foundation reports more than 150 organizations back A2A (April 9, 2026) and 305 members in the Agentic AI Foundation (October 22, 2026).
Ready for a service business?
Agents working together inside your own business can already work. Agents from different companies working together: it is starting.
What to do now
Five steps below. Four of the five need no developer.
How to start
A free strategy call about the one job you would hand to an agent first.Get ahead

What does agent-to-agent communication look like in a business I know?

It looks like a repair shop where each worker is an agent. The A2A project explains it that way in its own documentation. A customer, or the customer's own assistant, talks to the shop's Shop Manager agent. The Shop Manager holds the back-and-forth about what is wrong and hands the diagnosis to a Mechanic agent. The Mechanic uses MCP to reach its tools: a diagnostic scanner, a repair manual database and a lift. When it needs a part, it uses A2A to ask a Parts Supplier agent whether the part is in stock, and that can lead to an order.

  • The same pattern in your business. An intake agent reads the request, a quote agent drafts it from your price list, a scheduling agent finds the slot and a follow-up agent chases the answer. Inside one business, that can work today.

  • The rule that does not change. A person approves what matters at every step.

What is the difference between A2A and MCP?

MCP connects an agent to tools and data. A2A connects an agent to other agents. They are not rivals. The A2A project puts it this way: "Used together, MCP gives each agent depth, and A2A gives your system reach."

What is the difference between A2A and MCP?
QuestionMCPA2A
What it connectsAn agent to a tool or a set of data, like a calendar or a customer list.An agent to another agent, even one built by a different company.
The question it answersHow does my agent use this tool?How does my agent ask that agent to do a job?
Who published itAnthropic, November 25, 2024.Google, April 9, 2025.
Who looks after it nowThe Agentic AI Foundation, part of the Linux Foundation, since December 9, 2025.The Agentic AI Foundation, part of the Linux Foundation, since August 2026. The Linux Foundation hosted it from June 23, 2025.
Where it standsThe current version of the standard is dated July 28, 2026.Version 1.0, the stable release, came out on March 12, 2026.
In the repair shopThe mechanic uses the diagnostic scanner and the repair manual.The mechanic asks the parts supplier whether a part is in stock.
You need it whenYou want an agent to work inside your tools.You want agents from different companies to work together.

How does one agent find and trust another?

Each agent publishes an Agent Card, a small file that works like a digital business card. It says who the agent is, what jobs it does, where to reach it and how to prove who is asking. Another agent reads the card, proves who it is the way the card asks and sends a job. The agent doing the work can ask questions, report progress and hand back the finished piece. A job can take a minute or days, and a person can step in. Agent Cards can also be digitally signed, so an agent can check who published one.

  1. Introduce. The asking agent reads the other agent's Agent Card, usually found at a standard address on the company's website.

  2. Ask. It sends a task, which has its own ID and status.

  3. Hand back. The other agent returns the result, such as a document or a filled form.

When did all of this happen?

When did all of this happen?
DateWhat happenedWho
Nov 25, 2024MCP is published as an open standard for connecting AI to the systems where data lives.Anthropic
Apr 9, 2025A2A is published so agents built by different companies can work together.Google
Jun 23, 2025A2A is given to the Linux Foundation to run in the open.Google and the Linux Foundation
Aug 25, 2025IBM's agent protocol, ACP, is folded into A2A.IBM's ACP team
Sep 16, 2025AP2 is published: rules for agents that make payments.Google
Sep 29, 2025The Agentic Commerce Protocol is published: rules for agents that buy from businesses.Stripe and OpenAI
Dec 9, 2025MCP is given to the new Agentic AI Foundation, part of the Linux Foundation.Anthropic
Jan 11, 2026UCP is published: an open standard for agents that shop.Google, with its retail partners
Mar 12, 2026A2A version 1.0, the stable release.The A2A project
Apr 9, 2026The Linux Foundation reports more than 150 organizations back A2A.Linux Foundation
Aug 2026A2A joins the Agentic AI Foundation, the home of MCP.Agentic AI Foundation
Oct 22, 2026The Linux Foundation reports 305 members in the Agentic AI Foundation.Linux Foundation

Dates and versions on this page were checked against each standard's own announcement or documentation on October 29, 2026.

Sources: Anthropic, Nov 25, 2024 · Google, Apr 9, 2025 · Linux Foundation, Jun 23, 2025 · ACP team, Aug 25, 2025 · Google, Sep 16, 2025 · Stripe, Sep 29, 2025 · Anthropic, Dec 9, 2025 · Google, Jan 11, 2026 · A2A releases · Linux Foundation, Apr 9, 2026 · Agentic AI Foundation, Aug 17, 2026 · Linux Foundation, Oct 22, 2026 · MCP versions · A2A and MCP · Agent discovery · UCP

Is any of this real for a service business today?

Part of it. Agents can already work together inside one business. Agents from different companies working together through A2A is starting. The buying and paying standards began with shopping, and UCP's own site now lists a draft for booking hotel rooms. Nobody can say when a customer's assistant will book a plumber for them. What we can say is what an assistant needs to do it: clear, current information it can read, and a way to book.

What should a service business do now?

Do five things that help whether or not agents from other companies ever reach you.

  1. Make your basics match everywhere. Your hours, services, service area and contact details should be the same on your website and on every listing. Assistants read all of them.

  2. Give customers a real way to book or ask for a quote online. A booking page or a short quote form is something software can use. A PDF or a line that says "call us" is not.

  3. Decide what a machine may promise for you. Write down which prices, dates and discounts an agent may confirm, and which need your yes.

  4. Start one agent on one job inside your business. Learn what it gets right and wrong before you connect it to anyone else's.

    AI agents for service businesses
  5. Keep a record of what every agent does. What it did, when, and who approved it. That record is what lets you trust it with more.

Who can build and run agents like this for my business?

FlowChainLabs can. We build and run AI agents for the parts of a service business that eat the week, done for you, and we follow these standards as they change. We start inside your business: one job, one agent, a person approving what matters. That has to be right before any agent of yours deals with anyone else's. On the free strategy call you leave with the first agent we would build and what it would take, whether or not we work together.

Get ahead

Free strategy call.

What do the words mean?

AI agent
Software that does a job from start to finish and asks a person when a person should decide.
Agent Card
A small file an agent publishes that says who it is, what it can do, where to reach it and how to prove who is asking.
Task
One job that one agent asks another to do. It has its own ID and status, and it can take a minute or days.
Artifact
The finished piece of work an agent hands back, like a document or a filled form.
MCP server
A connection that lets an agent use one tool or one set of data, like a calendar or a customer list.
Open standard
Rules anyone can use, published in the open, so one company's agent can work with another's. A2A and MCP are both hosted by the Agentic AI Foundation, part of the Linux Foundation.
Human in the loop
A person who can approve, change or stop what an agent does.

Questions owners ask

How do AI agents talk to each other?

One agent sends the other a job as a short message in an agreed format. The other agent works on it, asks for what it needs and sends back the result. The format is public, so any company's agent can read it.

Can two AI agents talk to each other?

Yes. Inside one business, agents can already hand work to each other. Between different companies' agents, A2A is the open standard built for it, and the Linux Foundation reports more than 150 organizations back it.

Do I need A2A for my business?

Probably not today. You need a good agent doing a real job first. A2A matters when an agent you run has to work with an agent from another company.

Is A2A better than MCP?

They do different jobs, so you do not have to pick one. MCP connects an agent to tools and data. A2A connects an agent to other agents.

Who controls A2A?

It is open. Google published it on April 9, 2025, and the Linux Foundation has hosted it since June 23, 2025. In August 2026 it moved into the Agentic AI Foundation, which also hosts MCP. It is free to use under the Apache 2.0 license.

Is it safe to let agents deal with each other?

The standard includes ways to prove who an agent is, such as signed Agent Cards. Safety still depends on how each agent is built and what it is allowed to do. That is why a person should approve anything that moves money or makes a promise.

Will my customers' AI assistants book my business?

For products, standards for agents that buy and pay have started to arrive: AP2 from Google on September 16, 2025 and the Agentic Commerce Protocol on September 29, 2025. For services it is early. An assistant can only use what it can read, so accurate hours, services and a real booking page help either way.

How do we get started?

It starts with a free strategy call. We look at where work waits in your business and pick the first agent. You leave with that answer whether or not we work together.

Start with the one job you would hand to an agent first.

Get ahead

On the free strategy call we look at where work waits in your business and pick the first agent. You leave with that answer, whether or not we work together.

Founder experience before FlowChainLabs

Meet the founder