Nicola Sirchia

From B2B and B2C to A2A: what changes when the buyer is an agent?

By Nicola Sirchia. Published . 3 minute read.

One-minute version. AI avatar and AI voice, made from my own photo and voice.

The short answer

For twenty years we sorted marketing into B2B and B2C. A third case is arriving: the customer's AI agent talks to the brand's AI agent, and the two of them do most of the work before a person says yes. People call it A2A, agent to agent. The buyer is still a person or a company. What changes is who reads your brand first, and that reader has no patience for a slogan it can't check.

Is A2A my term?

No. A2A is the name of an open protocol, Agent2Agent, that Google announced in April 2025 and that the Linux Foundation has hosted since June 2025. It lets agents built by different companies find each other and hand work back and forth. At its one-year mark the foundation counted more than 150 supporting organisations. I use the term in its commercial sense: a sale prepared by two agents.

How far along are we?

The rails are built. The traffic is still light.

What Where it stands
The rails The Agentic Commerce Protocol from OpenAI and Stripe (September 2025) and the Universal Commerce Protocol from Google and Shopify (January 2026)
Choosing with AI 53% of consumers in 11 markets used AI for shopping in the past three months, in a McKinsey survey of about 21,000 people fielded in June 2026
Paying through AI About 1.5% of US online retail in 2026, in EMARKETER's forecast
Forecasts for 2030 15% to 25% of US e-commerce according to Bain. Forecasts differ widely because each firm defines "agentic" differently
53%of consumers used AI for shopping in the past three months (McKinsey, 11 markets, June 2026)
1.5%of US online retail is paid through an AI platform in 2026 (EMARKETER forecast)
Choosing with AI is common. Paying through AI is still rare.

People already let an assistant help them choose. Few let it pay. So the work today is about being chosen, and the checkout comes later.

What changes for a brand?

The first reader is a machine. Where a person looks at adverts, a machine compares data. Attributes, claims, prices, availability and terms have to be complete, and the same everywhere.

Your brand needs a voice that can answer. A website waits to be read. An agent gets asked. With my team I built an AI product advisor that answers from technical data sheets and product pages. That is the kind of source another agent can question.

Proof beats promise. A claim that carries its evidence travels from agent to agent. A claim without it gets dropped.

What does not change?

The reason people stay with a brand: it keeps solving their problem. An agent can put you on the shortlist, but it can't make the product work. B2B and B2C don't disappear either, because there is still a professional or a household at the end of the chain.

Where would I start?

  1. Ask an assistant twenty real customer questions in your category, and note where your brand is missing or wrong.
  2. Fix product data at the source before buying another tool.
  3. Give one person ownership of everything agents read about you.
  4. Decide what your own agent may answer, and when it hands over to a human.

This is the power of connected dots again. An agent can only use the dots you have connected.

Sources