The deal infrastructure that governs most media buying today was designed for humans. Insertion orders are negotiated by humans, reviewed by humans, and signed by humans. When a dispute arises about what was agreed, it is resolved through correspondence, contract review, and, if necessary, legal process. The timescales involved in that resolution process are measured in days or weeks. The transactions being disputed may have been executed in milliseconds. That gap has always been a tension. In an agentic market, it becomes a structural problem.

What problem does current deal infrastructure not solve?

When a human buyer negotiates a media deal, the process produces several artefacts: email correspondence, a term sheet or proposal document, a finalised insertion order, and entries in the buyer's campaign management system and the seller's order management system. These artefacts are not a single shared record. They are parallel records, maintained by each party in their own systems, that should agree but often do not.

When a dispute arises, each party produces their version of the record. Resolving the dispute requires determining which version is authoritative, which typically depends on which document was countersigned by both parties, and on the precise language of the insertion order. This process works, after a fashion, when deals are negotiated slowly enough for humans to review and agree on documentation at each stage. It does not work when deals are negotiated by AI agents at machine speed, because the documentation process assumed in the resolution pathway does not exist.

An AI agent committing to a deal does not produce an email chain. It does not request a countersigned PDF. It produces the output of a negotiation: agreed terms, in whatever format the negotiation interface supports. If that output is not structured as a bilaterally owned record at the point of commitment, there is no shared evidence of what was agreed.

What is a DealSheet?

A DealSheet is a bilaterally owned deal record that governs an agent-negotiated media transaction. It is produced at the point of commitment, owned equally by both the buyer's agent and the seller's agent, and contains the agreed terms in a form that both agents and the humans they represent can read and verify.

The DealSheet contains three categories of information. The first is the deal terms: the inventory being purchased, the price, the volume commitment, the flight dates, and any quality or placement specifications that were part of the negotiation. The second is the brief mapping: a record of how the committed terms relate to the buying brief, demonstrating that the deal satisfies the brief's requirements or flagging where it deviates. The third is the approvals log: a sequential record of who authorised what at each decision point in the negotiation, including any points where a human principal reviewed the agent's proposed action before it was committed.

The DealSheet is not a feature. It is the accountability layer without which an agent-negotiated transaction has no governance infrastructure.

Why does bilateral ownership matter?

A deal record held by a platform intermediary is not the same thing as a bilaterally owned deal record. When the record is held by a third party, both the buyer and the seller are dependent on that third party's systems to access the evidence of what was agreed. If the intermediary's system is unavailable, the record is unavailable. If the intermediary's record-keeping practices change, the format or content of the record may change. If the relationship with the intermediary ends, access to historical records may become uncertain.

A bilaterally owned record removes that dependency. Both parties have the same document. Neither party is relying on a third party's system to access their evidence. When a dispute arises, there is a shared starting point for resolution rather than two parties pulling records from two different systems and hoping they agree.

There is also a governance dimension that goes beyond dispute resolution. In a governed agentic system, the human principal on each side of a transaction needs to be able to verify, after the fact, that the agent acted within the brief and the parameters it was authorised to operate within. That verification requires access to the deal record. If the record is held by a platform intermediary rather than by the principal directly, the verification process depends on the intermediary's cooperation and the format of its reporting. Bilateral ownership puts the evidence in the hands of the people who need it for accountability purposes, not just the people who processed the transaction.

What does the DealSheet replace in the current stack?

The DealSheet is not intended to replace the delivery infrastructure of programmatic advertising: the ad servers, the SSPs, the DSPs, the measurement systems. Those continue to function as they do now. What the DealSheet replaces is the informal deal record that currently governs direct deals between buyers and sellers.

The insertion order, as a mechanism for governing agreed deal terms, was designed for human negotiation. It is a static document produced at a point in time, reflecting the terms agreed at that point, without any mechanism for capturing the negotiation process that produced it or the approvals that authorised it. When the negotiation was conducted by humans over days or weeks, that limitation was manageable. When the negotiation is conducted by agents in minutes, a static document produced retrospectively is not adequate.

The DealSheet is produced at the point of commitment, captures the negotiation record as well as the agreed terms, and includes the approvals log that shows how the deal was authorised. It replaces the insertion order for the specific class of transactions that are agent-negotiated, without requiring any change to the downstream delivery and measurement infrastructure.

What does the DealSheet enable that current infrastructure does not?

Three things. First, dispute resolution from evidence rather than recollection. When a buyer and a seller disagree about what terms were committed to, the DealSheet provides the shared, authoritative record from which resolution can begin. Neither party needs to reconstruct what happened from system reports that may not agree.

Second, compliance demonstration from the deal record itself. A buyer who needs to demonstrate to a client, an auditor, or a regulatory body that their media buying was conducted within brief and within authorised parameters can produce the DealSheet as evidence. The approvals log shows who authorised what. The brief mapping shows how the committed portfolio relates to the original brief. The compliance case is in the record, not in a separately constructed narrative.

Third, the approval rights that make autonomous agent action governable. An AI agent operating without defined approval rights is not governed. An AI agent operating within a DealSheet framework, where the approvals log captures every point at which a human principal reviewed and authorised an action, is operating within an accountability structure. The DealSheet is the mechanism by which human oversight is preserved in a system where agents are making decisions at machine speed.

Why is the DealSheet not optional in a governed agentic market?

The argument for optional deal governance is that it adds friction to a process that gains its value from operating at speed. That argument misunderstands the nature of the friction. A deal record produced at the point of commitment adds no meaningful friction to the transaction. The model has already assembled the portfolio and determined that it satisfies the brief. Recording that determination, and the terms of the deals that constitute it, is not an additional step. It is the output of the commitment stage.

What the DealSheet prevents is not speed. It prevents the governance failure that occurs when autonomous action cannot be reviewed, verified, or disputed because no record of what was decided exists. That failure has real commercial consequences: unresolvable disputes, unverifiable compliance, and the erosion of trust in AI buying systems that could otherwise create substantial value. The DealSheet is not a nice-to-have feature of agent-to-agent trading. It is the accountability layer without which autonomous action is ungovernable.