Agent-to-agent commerce: what a business needs before its customers' AI agents come shopping
"Every consumer and every business will have an agent or set of agents that represent them." That was Alex Schultz's prediction at Meta's Conversations event in London in June 2026. Some people already hand an agent a credit card and let it shop. Meta's July 2026 report sets out five layers this agentic economy runs on: identity, relationships, messaging, commerce, and models and protocols. This page explains each layer in terms a business owner can act on, and what Learnmind puts in place so your business is ready when they arrive.
The short answer
Agent-to-agent commerce is a customer's AI agent dealing with a business's AI agent. It asks, compares and books, within limits a person sets.
It is early and uneven. In Schultz's words, the future is here but "not evenly distributed yet".
Agents read facts, not persuasion. Meta's report says businesses without machine-readable catalogues, verified credentials and accessible pricing "don't appear in the results".
Four things are worth doing now: a verified identity, structured prices and availability, an agent that can complete a booking, and written limits on what it may agree to.
Several layers are still to be built. Schultz lists verification for agents, an app store for them, payments and legal rules as things the economy is "going to need".
What it already looks like
Schultz described people at the leading edge who already use agents on WhatsApp to complete whole transactions. One example: renting a car from a business whose own agent handles the rental. He also described colleagues who point an agent at the cameras in their homes and let it work out what the household needs. Then they give it a credit card and let it buy the goods. "It is making their lives better when it goes well," he said.
The conversations on Meta's stage kept returning to the same picture.
- Cars. Motorway's Tanya Cordrey said she could see a future in car selling "where agents are dealing with agents".
- Food. Meta's Derya Matras told Deliveroo's VP of Engineering, Vaughn Washington: "I would love your agent to know my agent." The idea: it would know what to order for the family, and how many calories are in it.
- Parts. Cordrey pictured a future where she photographs a socket and asks a seller on WhatsApp for the right cable. Schultz said he already does it. That is exactly the kind of exchange an agent can handle on either side.
On the business side, none of this needs a new app. The conversations can happen in the messaging threads people already use.
The five layers, and what each asks of a business
Meta's report lists five layers that together move businesses "from chatbots that answer to agents that act". Most of the work in each is ordinary. That is good news.
| Layer | What Meta's report means by it | What a business can do now |
|---|---|---|
| 1. Identity | A verified identity behind the agent, so the other side can trust its commitments | Verify your business on WhatsApp, keep one number and one display name, and make the agent say who it acts for |
| 2. Relationships | Discovery runs on trust and networks; agents need a directory to find businesses | Keep listings, reviews and your own site consistent, so every source says the same thing about you |
| 3. Messaging | The interface already in use, provided it is conversational rather than one-way | Answer in seconds, and treat every thread as two-way |
| 4. Commerce | Catalogues, marketplaces and checkout that keep up with agents comparing at speed | Publish structured prices and availability, and let the agent complete a booking or send a payment link |
| 5. Models and protocols | Capable models, plus a marketplace of tools and other agents they can depend on | Little to do at protocol level yet. Choose an agent that can use tools, not one that can only talk |
Why machine-readable beats persuasive
The report puts it in three sentences. "When a buyer's agent evaluates options, it queries structured data. Businesses without machine-readable catalogs, verified credentials, and accessible pricing don't appear in the results. Discoverability to agents is now a competitive surface."
A customer's agent does not respond to adjectives. It needs the facts that decide a booking, stated plainly and kept current. For a Dubai dental clinic offering a hygiene appointment, that looks like this.
| Fact an agent needs | Stated so a machine can use it (illustrative) |
|---|---|
| Service | Scale and polish, 45 minutes, with a hygienist |
| Price | AED 450, including a check-up |
| Next availability | Read live from the diary, not typed into a page |
| Location | Area, building, parking, nearest metro station |
| Languages | The languages the clinic's staff actually speak |
| Cancellation | Free up to 24 hours before; the deposit is kept after that |
| Payment | Card on the day, or a deposit link when booking |
Every line in that table is also what a human asks on WhatsApp. Getting ready for agents and answering people well turn out to be the same job.
Trust is the layer that is still being built
Schultz was candid about what is missing. "You're gonna need trust. You're gonna need to know that that agent represents that business and you can trust it." Today, he said, when you download a skill for your agent it sometimes calls "a third-party URL and you don't know what it's doing with your data".
Alex Schultz, Chief Data Officer, Meta
"Where we built an app store for humans, we need an app store for agents. Where you have a payments infrastructure for consumers today, you are going to need that for agents. Where you have a legal structure today, you are going to need that for agents."
Meta, Beyond chatbots: The agentic economy is here, July 2026Until those layers exist, a business has to supply its own guardrails. These are the ones worth building into any agent a business runs.
- Limits on what it may commit toPrices and terms come only from your published rules. Discounts, refunds and exceptions go to a person.
- Commitments in writingEvery booking, price and condition is confirmed back in the thread, so either side can check it later.
- Payment through your own systemsA secure link from your booking or payment provider. The agent never collects card numbers in chat.
- A person for anything unusualAn agent asking for something outside the rules, human or not, gets a named member of your team.
- A record of every transactionThe conversation and the outcome sit in your CRM, whoever or whatever was on the other side.
An agent-ready business still answers people first
Most of your enquiries will come from people for now. The good news is that the same facts serve both. Schultz made a second point that applies to people too. An AI agent can respond to what the customer actually wants and "maybe not push an upsell", where a person might feel they need to push one.
- Be precise. Dates in full, prices with the currency, inclusions listed.
- Be consistent. The agent, the website and the catalogue say the same thing.
- Be quick. A comparing agent will not wait for office hours, and neither will a comparing person.
How fast this is arriving
Meta's report gathers the forecasts. McKinsey puts the global economic impact of agentic commerce at $3 trillion to $5 trillion by 2030. Gartner expects more than 60% of enterprise customer service interactions to be managed end to end by agentic AI by 2028, up from 20% in 2026. The report also cites Deloitte putting the share of consumers moving product discovery from search engines to AI at 55%.
Forecasts are forecasts. The line Schultz repeated on stage is more useful for a business deciding what to do this quarter.
Alex Schultz, Chief Data Officer, Meta
"Agents are the dumbest they are ever going to be. If you are building for where agents are today, you are building for the past six months from now."
Meta, Beyond chatbots: The agentic economy is here, July 2026The report draws the conclusion for small firms directly. "A three-person operation gets the same always-on capability as an enterprise." What separates one business from another, it says, "is infrastructure readiness".
What Learnmind puts in place
- A verified number on the official WhatsApp Business Platform, in your business name.
- Your prices, availability and policies as structured facts the agent answers from, kept current.
- The ability to act: book into your diary, send a payment link, update your AI CRM.
- Written limits and a named person for everything outside them.
- Replies in seconds, at any hour, in Arabic and English, to whoever is writing.
Most Learnmind agents are live within two weeks. If a customer's agent comes looking next year, it will find a business that already answers in facts.
Want your business ready for agents and people alike?
We build agents on your own verified number that answer from structured facts, act within written limits and hand the exceptions to your team.
Terms worth being precise about
- Agentic economy
- An economy in which AI agents act on behalf of people and businesses, not only answer questions.
- Agent-to-agent commerce
- A transaction in which both sides are represented by AI agents, within limits people set.
- Machine-readable
- Stated in a consistent structure a program can use without guessing: fields, units, currency and dates.
- Verified identity
- Proof, through the platform's business verification, that the account an agent speaks from belongs to the business it names.
- Guardrail
- A written limit on what an agent may do or agree to without a person.
Frequently asked questions
What is agent-to-agent commerce?
A customer's AI agent asking, comparing and booking with a business's AI agent, within limits people
set.
Is anyone actually doing this yet?
At the leading edge, yes. Meta's Alex Schultz described people using agents on WhatsApp to complete whole
transactions, such as renting a car. For most businesses it is still rare.
Do I need a separate website for AI agents?
No. You need your facts stated plainly and consistently, and an agent that can answer from them. The same
work serves human customers.
What should my agent never agree to on its own?
Anything outside your written rules: discounts, refunds, exceptions and unusual requests. Those go to a
named person.
Can my agent tell whether it is talking to a person or another agent?
Do not rely on it, and it should not need to. Give both the same precise facts and the same limits.
How we checked this, and what we could not settle
Checked: the five layers, the statistics and the quotes marked as from the report come from Meta's "Beyond chatbots: The agentic economy is here", published in July 2026. Other quotes come from the transcripts of Meta's Conversations 2026 sessions "The AI Shift: The New Era of Customer Engagement" and "Building the Customer Journey of the Future".
Not settled: no agreed standard yet exists for how a customer's agent should verify a business's agent. Schultz names verification as a layer still to be built, and says agents are not yet represented in the legal structure. The clinic table above is illustrative, not a real clinic's prices.
Sources
- PlatformMeta, Beyond chatbots: The agentic economy is here, July 2026, citing McKinsey, Gartner and Deloitte
- PlatformMeta, AI Agents and the Future of Customer Engagement, the on-demand Conversations 2026 sessions
Written by Edmund Gay, Learnmind.ai, Dubai. Figures are as published by the sources named; forecasts are the forecasters' own.