From One-to-One to One-to-Many: The Topology Inversion at the Heart of A2A

Roughly seventy percent of mid-market and enterprise companies report deploying or piloting AI agents in customer-facing roles, with deployments touching chat, voice, web, and app surfaces. The numbers look impressive. The topology underneath them does not.

Almost every one of these deployments shares a single architectural pattern: a human customer interacts with that company's agent, on that company's surface. One consumer, one brand, one agent, one channel. We will call this the one-to-one pattern. Repeated across the brand-relationship surface of a typical consumer's life, fifty to two hundred recurring vendor relationships per household, the one-to-one pattern produces fifty to two hundred separate conversations, in fifty to two hundred separate interfaces, with no shared memory, no shared identity, and no aggregation of leverage. The consumer ends each interaction as a stranger to every agent they have spoken to.

This paper argues that the one-to-one pattern is not the natural shape of agentic customer engagement. It is an artifact of the deployment surface, not of the technology, and it is being inverted into a one-to-many architecture that places the consumer's agent, not the brand's, at the center. The inversion is the load-bearing change. Everything downstream of it is consequence.

The one-to-one pattern, in detail

The one-to-one pattern looks like this. A consumer arrives on a brand's website. The brand has deployed an agent on its support surface. The consumer asks the agent a question. The agent answers, optionally remembers the consumer's account context for the duration of the session, and resolves the conversation within the brand's policies. When the consumer leaves the surface, the agent forgets them. When the consumer arrives at a different brand's surface, the process repeats from scratch with no shared state.

This pattern is what every chatbot platform, every CX agent vendor, and every vertical assistant ships today. It is a continuation of the previous architectural era, the call-center era, translated into AI. The brand pays for the agent. The brand defines the agent's authority. The brand owns the conversation history. The consumer is, structurally, a guest on the brand's infrastructure for the duration of the exchange.

The one-to-one pattern is operationally sensible for the brand. It is also a dead end for the consumer. It does nothing to address the asymmetry we have written about elsewhere, the consumer's evening against the brand's dedicated infrastructure. It just gives the brand a faster, cheaper, slightly more personalized version of the same asymmetric infrastructure.

What the topology inversion changes

A one-to-many topology places a single agent in front of the consumer. That agent is not affiliated with any specific brand. Its loyalty, its memory, and its bounded authority all run to the consumer. When the consumer needs to negotiate a bill, dispute a charge, cancel a service, claim a benefit, or pursue any of the dozens of brand-mediated transactions a typical household conducts each year, the consumer's agent contacts the brand's agent and conducts the transaction.

This is one consumer agent talking to many brand agents, hence one-to-many, from the consumer's perspective. From the brand's perspective, the topology is the same: one brand agent talking to many consumer agents. The architecture is symmetric in shape but asymmetric in who pays for which agent. The consumer pays for theirs (subscription, success-fee, or marketplace-mediated). The brand pays for its own.

Three things change immediately.

The conversation actually happens. No hold queue, no menu maze, no scavenger hunt for the right contact channel. The consumer agent contacts the brand agent at machine speed and resolves the request in a single round trip. For the consumer, this collapses fifteen-minute interactions into seconds. For the brand, this drops per-contact cost to near zero.

Memory persists across brands. The consumer's agent knows the consumer's account history, payment patterns, prior disputes, preferences, and goals across every brand the consumer has ever transacted with. This is not the brand's CRM data, which is trapped inside each brand's silo. It is the consumer's own data, accessible to the consumer's agent. The consequence is that the consumer no longer has to re-explain themselves on every interaction with every vendor. The cumulative time savings, summed across a household's annual brand interactions, is meaningful, and the cumulative information advantage that accrues to the consumer's agent is structural.

Negotiation leverage is aggregated. When the consumer's agent negotiates with one brand, it brings to the table the consumer's full vendor portfolio. It can compare offers across competing brands in the same category in real time. It can credibly threaten churn, because it knows what the consumer would actually do. It can point to a specific competitor's price the consumer would accept. The consumer's agent has, for the first time, the same information advantage that the brand has had for thirty years.

The one-to-many inversion does not just speed up customer service. It moves the locus of leverage in the consumer-to-brand relationship from the brand to the consumer.

The marketplace condition

The inversion only works if there is a neutral layer where consumer agents and brand agents can find each other and transact under aligned incentives. Without that layer, the consumer agent has nowhere to go. With that layer, every brand agent that joins becomes available to every consumer agent on the platform, and vice versa. This is the classic two-sided marketplace dynamic, and it is the condition under which the one-to-many inversion becomes durable.

Marketplaces of this kind have specific structural properties. They aggregate value asymmetrically with scale: the value of joining is proportional to how many counterparties are already on the platform. They produce winner-take-most dynamics, where the platform that achieves liquidity first compounds away from competitors. They generate proprietary data: in this case, the corpus of signed agent-to-agent negotiation traces, that no entrant can replicate without years of operation.

The marketplace condition is what makes agent-to-agent commerce a category-defining opportunity rather than a feature improvement. The companies that build the marketplace layer first will define the protocols, the trust standards, and the economic model for the next decade of customer engagement.

Why the inversion is happening now

Three structural conditions had to be true for the inversion to begin, and all three landed in the last twenty-four months.

The first is agent capability. Until recently, consumer-facing agents could not credibly transact on a user's behalf. They could chat, summarize, retrieve, and route, but they could not negotiate, settle, or commit. The combination of better reasoning models, tool use, and structured-output discipline crossed the threshold of "agents that can actually act" some time in 2024. Once they crossed it, the consumer-side use case opened.

The second is protocol substrate. Agent-to-agent commerce requires a wire format. The combination of MCP, A2A, the Agentic Commerce Protocol, and the card-network entries (Visa TAP, Mastercard Agent Pay) created the protocol substrate in 2024–2025. Before that substrate, agent-to-agent transaction was custom integration work. After it, it is standard.

The third is financial-data connectivity. The consumer's agent is only useful if it knows the consumer's financial relationships: bank accounts, card accounts, subscriptions, insurance, recurring services. Bank-data connectivity through providers like MX, Plaid, Akoya, and Finicity has been productized over the last decade and is now a mature layer the consumer agent can build on without renegotiating individual bank integrations.

These three substrate conditions are why the inversion is happening now and not in 2018 or 2030. They are also why the window for category-defining companies is finite: substrate conditions stabilize protocols, and once protocols stabilize, the category structure crystallizes.

What incumbents and entrants should plan for

Two implications follow.

For brands: the one-to-one deployment is a transitional artifact, not a destination. Brand-side agents that were deployed in 2024–2025 to handle one-to-one customer interactions will need to be re-architected to interact agent-to-agent in a one-to-many topology. The brands that handle this transition gracefully will preserve customer relationships through the inversion. The brands that don't will discover that their customers' agents are negotiating with a platform that knows the brand's competitive position better than the brand does.

For entrants: build for the one-to-many topology from day one. The temptation to build a one-to-one consumer agent: a vertical assistant for shopping, or fashion, or travel: is real, because it is easier to ship and easier to fundraise. But the topology of value is one-to-many, and the company that builds across the consumer's full brand-relationship surface will compound away from any one-to-one competitor within three to five years.

The inversion from one-to-one to one-to-many is the architectural change that will define agentic customer engagement. Everything else: the protocols, the trust layer, the payment rails, the marketplace economics, is consequence. Mapping the inversion correctly is the difference between building infrastructure for the next decade and building a feature for the last one.