There is a slide that appears in nearly every agentic advertising deck. On the left, today's supply chain: a tangle of boxes. DSP, SSP, exchange, verification, resellers, arrows everywhere. On the right, the agentic future: brand, agent, publisher. Three boxes, two arrows. The tangle is gone.
The slide never says where the tangle went. It is worth asking, because the industry has seen this slide before.
Supply path optimisation made the identical promise in 2018: fewer hops, cleaner paths, less leakage. The chain's response was instructive. It did not get shorter. It grew curation, an entirely new layer that charges a fee for organising the other layers. Resellers multiplied, ads.txt files bloated into the thousands of lines, and by the time the ANA ran its 2023 transparency study, the average advertiser was still transacting through thirteen SSPs per campaign, with some touching fifty-eight. Every reform aimed at removing intermediaries has so far concluded with an additional intermediary whose product is managing the rest.
Agentic buying will be sold as the reform that finally works, because this time the middle does not get pruned, it gets replaced by intelligence. Look closely at what actually happens. The agent does not remove the hops. It absorbs them. The path from budget to impression still exists, every auction, every fee, every reseller relationship, but it now runs inside the agent's decision-making rather than across systems a human can pull apart. Today, a buyer who suspects the path is leaking can demand log files and trace it, slowly and expensively, but it can be done. The ISBA and ANA studies were both built by doing exactly that. Tomorrow, the answer to the same suspicion is a sentence: the agent optimised it.
A fee that lives inside an optimisation is the most defensible fee ever invented. It never appears on a rate card. It cannot be benchmarked, because no two agent decisions are alike. It does not even need to be hidden in the old sense, because there is nothing left to hide: there is simply no artefact a third party can request. ISBA found 15p in every pound unaccounted for across a chain where every hop at least produced an invoice. The agentic version of that delta will not be a discrepancy in the records. It will be an absence of records to be discrepant.
This is the distinction the three-box slide quietly elides: abstraction is not disintermediation. Wrapping a supply chain in a model does not shorten it. It moves it somewhere audits cannot follow.
There is a one-question test for any agentic stack a vendor shows you this year. Can a third party reconstruct the path the money took without asking the agent's owner? Not a dashboard the owner built. Not a report the agent writes about itself. An independent reconstruction, from records the owner does not control. If the answer is yes, the chain has genuinely changed. If the answer is no, you are looking at the old chain with better conversation skills.
Alkimi runs an on-chain marketplace where that reconstruction is the default, because every transaction settles on a public ledger and the path is a matter of record rather than a matter of disclosure. Alkimi also publishes this blog, which makes this the least neutral paragraph in the piece. It is, deliberately, also the shortest.
Ask the question anyway, of every vendor who presents you the three-box slide. The length of the pause before the answer is a reasonable estimate of the length of the chain.
Related reading: Everyone Built the Easy Half. The Agent Learned It From Watching You. Measurement Isn't Verification: Fifteen Years of Self-Graded Essays.
Pilot enquiries: lauren@alkimi.org or marco@alkimi.org