If coordination is already proving problematic across platforms, the challenge becomes significantly more complex once additional AI agents begin operating across the advertising stack.
Modern programmatic campaigns rely on multiple specialised systems, each designed to optimise one particular aspect of performance. These include:
Bidding agents: identify the optimal price for each impression
Brand safety filters: evaluate whether inventory meets suitability requirements
Fraud detection systems: monitor traffic quality and protect media budgets
Each agent performs its task with precision. Each has its own dashboard, its own metrics, and its own definition of success. The problem is that there is little to no coordination between them. Instead, they analyse different datasets, optimise towards different objectives, and record their own interpretations of the same transaction.
This is where structural friction emerges at the boundaries between agents. A brand safety filter may block an impression, preventing a bidding agent from placing a bid. The fraud detection system may subsequently flag a discrepancy, while measurement tools record a shortfall in delivery against the target.
One underlying event. Four different accounts of what transpired.
Each agent's decision is rational. But without a complete view of the transaction, fragmentation occurs. And this cannot be solved by simply adding more dashboards.
Dashboards may show that brand safety filters blocked 2% of impressions and that delivery hit 97% of the target. Both are technically accurate. What they fail to reveal is that those blocked impressions triggered pacing adjustments elsewhere that shifted cost per thousand impressions (CPMs) and forced bidding systems to compensate. In other words, dashboards add more correct-but-incomplete views of a system that nobody is seeing as a whole.
The impact of this fragmentation is often felt during reconciliation. This has traditionally been perceived as a back-office function, where invoices and trafficking reports are reviewed, discrepancies are identified, and human teams work to align buyer and seller records.
This is manageable when transactions take place at human speed, as teams have sufficient time to investigate, negotiate, and adjust. A 3-4% discrepancy on monthly insertion orders is considered an acceptable margin.
But what happens when these transactions scale? When an AI-to-AI marketplace facilitates thousands of daily transactions, even the smallest discrepancies compound rapidly.
Now consider where the industry already is before autonomous agents enter the picture. ISBA and PwC's 2020 programmatic supply chain study found that only 51% of advertiser spend reached publishers. Around 15% of that spend was an unattributable "unknown delta" that no party in the chain could explain. The follow-up in 2022 closed some of the gap, but the core finding held: even at human transaction speeds, in a market with established contracts and quarterly reconciliation cycles, a significant slice of programmatic spend cannot be matched between buyer and seller records.
That is the floor. Not the ceiling.
Layer autonomous agents on top of that infrastructure, and the failure modes are predictable. Fraudulent buyer agents impersonating legitimate buyers. Seller agents misrepresenting inventory quality or availability. And critically, even when individual agents detect bad actors, there is no shared mechanism for that knowledge to propagate across the marketplace. Each agent works from its own ledger. None of them can warn the others. The same structural gaps that ISBA documented compound rather than resolve, because the speed of transactions has changed, but the underlying reconciliation logic has not.
If 51% of spend struggles to reach publishers at human speed, the question is not whether agent-to-agent transactions will improve that figure. It is how much further it falls before the structural problem is addressed.
The fact that agents can transact faster than humans can verify is irrefutable. Implementing additional agents will only increase coordination complexity. And while dashboards are effective for visualising conflict, they do not resolve structural issues.
Human oversight alone cannot scale orchestration. The coordination problem will only escalate as more autonomous systems are introduced into the advertising ecosystem, unless changes are made.
About Alkimi
Alkimi is the agentic marketplace for digital advertising. Agents work from the same deal sheet, one live document that updates in real time as agents negotiate, agree terms, and optimise campaigns. Advertisers brief campaigns in plain language, publishers list their inventory, their agents match them, negotiate on their behalf, and close deals automatically, every step logged and verifiable. No reconciliation or middlemen. Just a shared record both sides can trust. Read more about Alkimi here: www.alkimi.org