Get Your Agent to Talk to My Agent

"Get your agent to talk to my agent" was once a Hollywood brush-off. In commerce it is about to become a literal description of how things get bought.

A buyer tells an AI agent what they need, what they will pay and what matters to them. The agent compares offers and, where permissions allow, acts on the buyer's behalf. Brands and retailers are beginning to deploy agents of their own, and direct negotiation between the two is the next step.

As purchasing authority shifts from people to software, brands must become credible choices to systems that evaluate offers on someone else's behalf.

How the conversation works

The buyer's agent starts with a brief: a running shoe for a first marathon, under a set price, delivered by Friday, from a brand the buyer has not had trouble with. It builds a shortlist from what it already knows and what it can retrieve. Where sellers have agents of their own, it can put questions to them directly.

Where one exists, the vendor's agent answers for the brand or the retailer. It knows the catalogue, the stock position, the margin floor and the offers it is allowed to make. When the buyer's agent asks whether a cheaper rival model would do the job, the vendor's agent has to make the case in terms the buyer's agent will accept: fit for the stated need, verifiable claims, total cost, delivery certainty.

Such an exchange could run in seconds and repeat across a dozen sellers at once. Price would matter, and so would evidence, consistency and trust. A buyer's agent designed to protect its principal should discount claims it cannot verify and treat a vendor that contradicts its own product data with suspicion. Card networks, payment firms and AI platforms are already building the payment and checkout rails, so brands have reason to prepare before the negotiation layer matures.

What changes for brands

Brand equity faces a new test. An agent does not experience your campaign or packaging as a human shopper does. It evaluates the representations of your brand available to it: learned associations, retrievable evidence, product data, reviews, terms and the fit between your offer and the buyer's stated needs. Brand recognition may get you considered. It does not guarantee that you survive comparison.

Shortlisting becomes the hardest step. A brand the buyer's agent never considers cannot be negotiated with. Our own work on how assistants recommend brands shows how easily a well-known name can be known and still not chosen. In agent-to-agent commerce, that gap can become a lost transaction with no visible trace.

Claims have to survive interrogation. A vendor's agent can expect to be asked to justify its product against specific criteria and against named rivals. Claims written to sound distinctive on a shelf can work against you when an agent is matching them to a buyer's literal requirements. Discounting is a blunt fallback. If the only lever your agent can pull is price, you have handed the negotiation to whoever has the lowest margin floor.

What changes for retailers

Retailers lose some of the moments they have spent decades designing. The store layout, the homepage, the end-cap and the sponsored listing all assume a human is looking. A buyer's agent looks at structured data, terms and reliability. Retail media that depends on human attention will need a counterpart that speaks to agents.

Retailers also gain a new role. A retailer's agent can bundle, substitute and guarantee in ways a single brand cannot. Its advantage may be less about persuading the buyer to want a product and more about making its offer the most practical choice: available now, delivered reliably, easy to return and competitively priced. The brand owns the product proposition. The retailer can control much of the transaction proposition.

That sets up a three-way tension. Brands want their product chosen. Retailers want the basket and the customer relationship, and a retailer's agent will represent the retailer's interests. A buyer's agent is designed to serve its principal, ahead of the retailer or the brand. Commercial terms between brands and retailers are likely to start covering how each party's agent represents the other.

Mapping and measuring the new market

Two questions follow for any brand or retailer. Who are the agents, platforms and protocols that will sit between me and my customers? And when a buyer's agent evaluates my brand, does it choose me?

The AIVO Atlas of the Agentic Economy maps the emerging market: the agents, protocols, payment infrastructure and relationships shaping how agentic commerce may develop. It answers the first question.

AIVO Meridian addresses the brand-level question. It measures how AI assistants evaluate brands across multi-turn buying conversations: which brands enter consideration, which survive comparison and which are ultimately recommended. Where it finds weaknesses, its remediation work helps close them. These tests provide a measurable baseline for the decision layer that agentic commerce will increasingly depend on. They do not, by themselves, establish how often a live autonomous agent completes a purchase.

What to do now

We describe agentic commerce through three gates: Access, Choice and Execution. Each creates a test brands can run now, an owner for the work and a decision the results should trigger.

The access test. Can an agent find and retrieve your relevant product information, claims, availability and commercial terms? Run realistic buying briefs through the main assistants and inspect what information they use. Digital, product-data and content teams own the work. Missing, contradictory or inaccessible information needs fixing before the brand can compete reliably.

The choice test. Does your brand make the shortlist, survive comparison with named competitors and win the recommendation? Repeat realistic buying conversations and record each stage. Brand and product teams own the work together. Weak consideration calls for better representation across the information sources models use. Claims that fail under comparison need stronger evidence, clearer qualification or a more precise target buyer.

The execution test. Can the agent complete the intended purchase under the applicable permissions and commercial terms? Check price, stock, delivery, returns, payment and transaction permissions across your own channels and your retail partners'. E-commerce and commercial teams own the work. Prioritise the discrepancies most likely to prevent a purchase, then define what your own agent may offer and concede.

Assistants provide a useful first baseline for Access and Choice. Execution needs more: a proper test checks real transaction capabilities and permissions, because an assistant saying it could complete a purchase proves little.

None of these tests requires autonomous negotiation to become routine. They establish whether your brand is accessible, competitive and transactable for software acting on someone else's behalf. Those are the conditions on which agentic commerce will depend.