Media trading has always depended on trust. A buyer trusts that the inventory is what the seller claims. A seller trusts that the buyer will pay. The human relationships that underpin those trust assumptions have been managed through agency relationships, contracts, and the reputational consequences of getting it wrong.

When both sides of a deal are operating through agents, none of those trust mechanisms apply in their current form. The agents do not have reputations in the way that humans do. They do not have long-term relationships that create accountability over time. And they operate at a speed that makes the conventional contract-and-dispute process unworkable as a primary trust mechanism.

What does trust mean in a transaction between two agents?

In a human transaction, trust is a judgment made on the basis of past behaviour, reputation, contractual obligation, and the expectation of future dealings. These are social and institutional mechanisms. They work because humans have identity, memory, and stakes in outcomes beyond the immediate transaction.

An agent has none of these in the same sense. It does not carry a reputation that accumulates across transactions in the way a business relationship does. It acts on instructions. It can be reconfigured. Its counterparty in a given transaction cannot assume that the agent representing a business today will behave consistently with the agent representing the same business yesterday.

Trust between agents therefore needs to be structural rather than relational. It cannot rest on the expectation that the other party has a stake in the long-term relationship. It needs to be built into the transaction itself.

Who verifies the deal when neither party is human?

In a human deal, verification happens at several points: during the negotiation, through the contract review process, and through post-campaign reporting. Each of these points involves a human making a judgment about whether what was agreed matches what was delivered.

When the negotiation and execution both happen between agents, the verification question changes. The buyer agent and the seller agent can agree on terms. They can record those terms. They can even check delivery data against those terms automatically. But the question of whether the agreed terms were the right terms for the campaign, whether the deal represented value, and whether a compliance deviation is significant enough to trigger a remedy: these are judgment calls that require human input.

The answer is not to remove agent involvement. It is to design the system so that human judgment is applied at the points where it matters most: at deal approval, and at the point where compliance data triggers a decision. Everything else can be handled by the agents.

Why is a shared deal record the foundation of trust in agent-to-agent trading?

A shared deal record changes the accountability structure in a specific way: it makes the agreed terms visible to both parties simultaneously and permanently. There is no room for the buyer and seller to hold different versions of what was agreed.

In current media trading, the deal record problem is real. A buyer and seller can have different records of the terms they agreed, especially when the deal was negotiated over multiple conversations and email threads. When delivery data arrives, disputes often turn on what was actually committed to rather than whether delivery met a clear standard. Each side produces its own version of the agreement.

A bilaterally owned deal record eliminates this ambiguity. Both parties see the same terms. Both parties approved the same terms before the buy was committed. When delivery data arrives, it is checked against a record that neither side can retrospectively amend.

How does Alkimi's DealSheet address the trust problem?

The DealSheet is Alkimi's deal record layer. It is designed specifically for agent-to-agent transactions where neither party to the negotiation is human, but both businesses behind those agents need to be able to trust that the deal was executed on the terms they approved.

The DealSheet captures the deal terms as agreed between the agents. It requires human approval from both sides before the buy is committed. It holds the approval record alongside the deal terms. And it provides the audit trail that either party can reference when delivery data is reviewed.

The human approval step is not a compromise with the agentic model: it is a feature of it. Earned autonomy, as a principle for agentic deployment, means that agents operate within boundaries that have been explicitly approved by the humans who are responsible for the outcome. The DealSheet makes those boundaries visible and enforceable.

What breaks without a shared deal record?

Without a shared deal record, agent-to-agent trading reduces to a faster version of the existing problem. Agents negotiate at machine speed, but the deal terms that result from that negotiation are held separately by each party. Each party's systems record what their agent agreed to from their perspective. When those records diverge, there is no neutral reference.

The volume and speed of agent-executed deals compounds this problem. Human-negotiated deals involve enough friction that disputes are proportionally infrequent. Agent-executed deals run at a volume where even a small proportion of ambiguous or disputed terms represents a significant operational problem.

The trust infrastructure for agent-to-agent trading is not a nice-to-have layer on top of the transaction. It is the precondition for the transaction being a real agreement rather than two agents independently recording their version of the same event. Without it, the speed advantage of agentic execution creates a scale of ambiguity that the market cannot absorb.