Most of the discussion about agentic media buying focuses on the buy side. An agent interprets a brief, identifies suitable inventory, and executes deals on behalf of an advertiser. The sell side is treated as passive: inventory presented, bid received, impression served.

That framing becomes wrong as soon as the sell side is also agent-driven. If a publisher deploys a sell-side agent to manage deal negotiation, yield optimisation, and compliance monitoring, the architecture of the sell side changes in ways that matter for how the overall system works.

What does a DSP actually do that a sell-side agent needs to do differently?

A demand-side platform operates on policy rules. It applies targeting criteria to available inventory, bids according to a defined algorithm, and executes against the rules its operators have configured. It does not negotiate. It does not reason about whether a particular deal is in the best interests of the buyer in the context of the specific campaign brief. It executes policy.

A sell-side agent in an agent-to-agent market needs to do something categorically different. It represents the publisher's inventory and interests in a negotiation with a buyer agent that is itself making contextual judgments about the brief. The sell-side agent needs to reason about what it can offer, at what price, under what compliance terms, and why a specific deal would be good or bad for the publisher's yield and relationship with the advertiser.

This is not a matter of faster policy execution. It requires a different kind of system.

How should the sell-side agent structure deal proposals differently from a DSP?

A DSP produces a bid: a price against an impression. The proposal is unilateral and contains only the information needed to clear an auction. The DSP does not propose deal terms, because there are no deal terms in a programmatic auction.

A sell-side agent in an agent-to-agent negotiation needs to produce a deal proposal. That proposal includes price, but also the audience it can evidence, the inventory specifications, the delivery window it can commit to, the compliance parameters it will hold to, and the verification data it can provide post-delivery. The proposal is structured so that the buyer agent can evaluate it against the brief and respond with a counter-proposal or acceptance.

The format of the proposal matters. A buyer agent processing dozens of concurrent negotiations needs the sell-side proposal to be structured in a consistent, machine-readable format that makes direct comparison possible. A DSP optimised for auction clearing does not produce this format. A sell-side agent designed for bilateral negotiation must.

How should approval logic work on the sell side?

A DSP executes within the policy bounds its operators have set. If a bid meets the floor price and clears the targeting filters, the impression serves. There is no per-deal approval step because the policy is the approval.

A sell-side agent operating in a bilateral negotiation needs a different approval logic. Deals above a certain value, deals with non-standard compliance terms, deals that involve novel audience parameters, or deals with advertisers whose category is sensitive: these should trigger a human review before the deal is committed. The sell-side agent should be able to identify which deals fall into these categories and route them accordingly.

This is the earned-autonomy model applied to the sell side. The agent handles routine deal execution within defined parameters. Non-routine deals, or deals that exceed the parameters the publisher has approved for automatic execution, are escalated for human review before the commitment is made.

What does bilateral record-keeping require from the sell side?

In a bilateral deal, the sell-side agent is a co-owner of the deal record. This is different from the current position of an SSP in programmatic trading, where the deal record is the SSP's own log and the publisher's accountability is to the SSP rather than directly to the buyer.

A sell-side agent in an agent-to-agent market needs to maintain a record that matches the buyer's record of the agreed terms. When delivery data is checked against the deal, both sides should be able to do that check against the same record. If there is a discrepancy, the discrepancy is visible in the record, not hidden in each party's separate accounting.

This requires the sell-side agent to maintain deal records in a format that is compatible with the shared deal record infrastructure. Alkimi's DealSheet provides that infrastructure. The sell-side agent writes to the DealSheet at the point of deal agreement, and the DealSheet becomes the reference record for both parties throughout delivery and post-campaign reconciliation.

Why does the sell-side architecture matter for how the market develops?

An agentic market in which only the buy side has agent infrastructure is a partial market. The buyer agent negotiates against a sell side that is still operating on policy rules, which means the negotiation is asymmetric. The buyer is trying to specify deal terms; the seller can only offer what its policy rules allow.

A fully bilateral agent market, where both sides have agents capable of contextual reasoning, structured proposal generation, and bilateral record-keeping, is a qualitatively different market. Deal terms become genuinely negotiable. Compliance parameters become contractually specific. The accountability structure is symmetrical.

Publishers who invest in sell-side agent infrastructure that can operate in this way will be positioned to participate in the bilateral agent market as it develops. Publishers who do not will find themselves defaulting to the programmatic clearing mechanism for an increasing share of deals that the buy side is trying to execute through bilateral negotiation. The sell-side architecture is not a downstream consideration. It is a strategic one.