TL;DR: Agent-to-agent negotiation reduces intermediaries by allowing a buy-side agent to reach a deal directly with a sell-side agent, cutting out the ad exchange, SSP reseller layers, and the bid optimisation tools that sit between them. The result is a shorter supply path, a lower total technology fee, and a deal record that both parties hold in shared state rather than reconciling separately after the fact.

Conventional programmatic buying routes an impression through multiple layers before a buyer wins it. The buy-side starts at a DSP, which connects to one or more SSPs, which connect to the publisher's ad server. Each layer adds a fee, introduces latency, and holds a partial record of what happened. By the time a campaign is running, the buyer's impression log and the publisher's delivery log are two separate documents that rarely reconcile cleanly.

This architecture was designed for auction-based buying at scale. It works tolerably well for that purpose. It works less well when a buyer wants to negotiate terms directly with a publisher, agree a CPM, and hold a shared record of what was agreed. That kind of bilateral deal has been possible in theory for years, through programmatic guaranteed and private marketplace products. In practice it has involved the same multi-layer infrastructure, with the SSP managing the deal setup and the DSP managing delivery, and two separate records of what each party agreed to.

Agent-to-agent negotiation changes the architecture, not just the execution method.

What changes structurally

When a buy-side agent negotiates directly with a sell-side agent, the path from buyer to publisher is shortened to two parties plus a marketplace that holds the shared deal record. The buy-side agent carries the buyer's mandate. The sell-side agent carries the publisher's parameters. The two agents reach agreement and write the agreed terms to a bilateral deal record, the DealSheet, that both parties hold simultaneously.

There is no SSP intermediary managing the deal setup. There is no DSP bid optimisation layer. The marketplace provides the negotiation infrastructure and the deal record service. Both parties connect to it once, rather than each connecting to a chain of intermediaries that connect to each other.

This matters for fees. In conventional programmatic, the buyer's DSP fee, the SSP fee, and any data or verification layer fees are stacked. Industry estimates have consistently put the proportion of media spend that reaches the publisher at between 50 and 60 per cent. Agent-to-agent negotiation, running through a neutral marketplace that charges a fixed deal fee, replaces a percentage-of-spend stack with a flat service cost.

What changes for the deal record

In conventional programmatic, the buyer holds an impression log and the publisher holds a delivery log. These are generated by different systems and are not the same document. Reconciliation is a post-campaign exercise that often produces discrepancies, which are then resolved through a credits and adjustments process.

A bilateral deal record is a single document, written at negotiation time, held by both parties in shared state. The agreed CPM, the inventory scope, the audience data permissions, the delivery commitments: all of these are captured in the DealSheet before a single impression is served. Post-campaign reconciliation becomes a comparison of delivery data against a shared reference document, rather than a comparison of two separate logs.

This matters for audit. When a buyer asks "what did we agree, with whom, and at what terms?" they are looking at the same record as the publisher. There is no asymmetry between what the buyer's systems say happened and what the publisher's systems say happened.

What does not change

Agent-to-agent negotiation does not remove the need for planning, targeting, creative, and measurement. Those functions remain in the buyer's and publisher's existing systems. The agents handle one specific function: negotiating and documenting the deal. Everything else stays where it was.

It also does not remove the need for human oversight. The buyer's agent operates under a mandate that a human has approved. Deals above a defined financial threshold require explicit approval before they are executed. The agent is not autonomous in the sense of operating without human governance. It is autonomous in the sense of executing the negotiation without requiring a human to manage each individual bid or proposal.

The supply chain after agent adoption

The supply chain does not disappear; it narrows. Premium inventory that is valuable enough to justify bilateral negotiation migrates to agent-to-agent channels. Open exchange remains for inventory that does not meet the threshold for direct negotiation. The two channels run in parallel, as they do today with programmatic guaranteed alongside open auction. The difference is that agent-to-agent bilateral deals are cheaper, faster to execute, and better documented than their manual equivalents.

For buyers focused on supply chain efficiency, the question is not whether agent-to-agent negotiation reduces intermediaries. It demonstrably does. The question is which inventory categories justify the switch first.