A buyer agent confirms a CTV impression delivered. A seller agent confirms it was delivered. Three intermediaries between them each clip a fee. The buyer's report says one number. The seller's report says another. Both reports are internally consistent. Neither party can verify the other. Both campaigns close out clean.
That happens in programmatic every working day, with humans in the loop. What the humans were doing in the loop was absorbing the discrepancy. The relationship phoned the other relationship, the two reports got close enough, the campaign closed out. Agentic systems do not have those relationships. They have logs. The version of programmatic where two agents shake hands on a number while a third party clips a fee neither agent can see is not stable. It is the substrate the next generation of buying will refuse to run on.
What follows is the operational version of "on-chain" for the programmatic veteran who has heard the word a hundred times and never seen anyone explain what it buys them. No blockchain background is assumed and none is required. The point is structural, not technical, and the operative example is sitting in production: 20 million daily impressions on Alkimi's exchange, three billion cumulative transactions, brands including AWS, TikTok, Currys, Kraken, Ripple and Polestar running against premium publisher inventory from The Wall Street Journal, The Guardian, Bloomberg, Bauer Media and News UK.
What Actually Changes
Strip the language down. The thing an on-chain advertising layer does is move the transaction record off the platforms that profit from the transaction and onto a shared ledger that all parties can read but none of them control.
That sounds modest. It is not. Almost every operational pain point in programmatic that the industry has spent a decade trying to fix traces back to the fact that the transaction record sits on the platforms rather than between them. Fees are reconciled after the fact because the fee record lives on the intermediary's books. Supply paths are opaque because each hop holds its own portion of the data. Discrepancies happen because the buyer's record and the seller's record are different documents recorded at different times by different systems with different incentives. Fraud is profitable because the surface where fraud is detected is owned by parties who have economic reasons to detect less of it.
Move the record off those systems and the operational consequences are direct. Fees become a number visible at the point of transaction rather than a discovery at the post-campaign audit. The supply path becomes a structured field any party can read, with the number of hops, the identity of each hop and the fee at each hop available without permission. The buyer's record and the seller's record stop being two documents because they are the same record. Reconciliation collapses from a process to a query.
That last point matters most to the operator. The current reconciliation industry exists because the transaction layer is unverifiable. MRC-accredited measurement vendors, IAS, DoubleVerify, fraud detection platforms, brand safety verification providers. Each one charges a fee. Each one operates on data supplied by the systems whose claims they are verifying. The industry has built a verification layer on top of an unverifiable transaction layer, which is structurally the same problem as auditing a bank by asking the bank for the figures. Remove the underlying problem and a meaningful fraction of that stack is restructured, not augmented. That is the part of the story the verification industry is quietly tracking and not loudly endorsing.
Why This Was Not Operationally Feasible Until Recently
The reason "put it on a blockchain" has been a punchline for most of the last decade is that the infrastructure could not handle the workload. Real-time bidding operates inside a 100-millisecond window. Ethereum settles in approximately 12 minutes. Most chains that were faster offered probabilistic finality, meaning a transaction could technically be reversed under specific conditions. Fast enough but not certain enough. Cheap enough but not deterministic enough. The maths did not work for transactions worth $2 to $3 each, settling at millions per day, on infrastructure that had to be both fast and final.
Sui's Mysticeti consensus settles in approximately 390 milliseconds with deterministic finality. Fast enough for the RTB window, certain enough to settle without reversal, with per-transaction network fees low enough that the economics of micro-transactions work. That combination is recent. The Alkimi exchange processes roughly 20 million daily impressions on it. Cumulative transactions on the network have passed three billion. The infrastructure works at the operational scale programmatic actually runs at, which is the point at which "on-chain advertising" stops being a thought experiment and starts being a stack.
What This Means for Agentic Programmatic
The relevant point in 2026 is what happens when the agents arrive.
Agentic systems are optimisers. An optimiser is only as intelligent as the data it can verify. An agent operating on a stack where agreement is provisional, transactions are reconciled afterwards and verification is supplied by the parties being verified is an agent optimising on assumptions. An agent operating on a stack where the record is independent of the parties to the transaction is an agent optimising on evidence. Both agents run at the same speed. They produce different outcomes, because they are routing money against different inputs.
Which means in an agentic environment, the infrastructure choice stops being an ideological preference and starts being an optimisation gradient. The platforms with verifiable records become structurally more useful to agents than the platforms with cleaner dashboards and better self-reporting. Not because verifiability is virtuous. Because agents are optimisers and opacity is friction they route around.
The downstream evidence sits in campaign data. The Polestar connected-TV activation running on this stack produced 34% sales intent lift and 24% lift in "100% electric" brand association, measured by Nielsen rather than self-reported. The way those numbers were reached was unremarkable: working media reached the screen because the supply chain stopped consuming it on the way. The reason the numbers are credible is structural. The campaign data was queryable independently throughout, which is the part the brand uplift study could rely on and the part most platforms cannot offer.
That is the part of the conversation the agentic discourse keeps deferring. The agents themselves are not the variable. The substrate underneath them is.
What This Leaves the Operator
The supply path that has worked through the era of human reconciliation is not the supply path that will hold up to agentic verification. The gap between what the campaign reported and what the campaign delivered is now a structural risk rather than a tolerable inefficiency, because the agents being deployed this year will surface it whether the operator wants it surfaced or not.
The fee structures that survived the era of post-campaign audit do not survive an agent reading the fee at the point of bid. An agent will route toward the cleaner number. The platforms with transparent take rates win that routing by default, not because anyone is persuaded to use them.
The question to ask of any agentic capability being pitched in the next two budget cycles is whether the agent can independently verify what it bought, against a record neither party controls. If it cannot, the agent is running faster against the same opacity that has been costing the industry money for a decade. Speed without verification is not progress. It is just losing money faster with better dashboards.
---
Related reading from Alkimi Research
What Your Clients Are Actually Asking When They Ask About Agentic. How to position the verification conversation with agency clients.
Test a direct supply path against your current stack
Side-by-side pilots through the Alkimi research team compare a current supply path against a direct alternative across three to four weeks. Working media, cost efficiency, brand uplift and carbon footprint measured on the same campaign. Contact lauren@alkimi.org or marco@alkimi.org to get started.