Two incompatible ideas about how advertising should be bought have coexisted for a decade. Television sells a quantity of audience, negotiated in advance, settled against a currency both sides agree to treat as true. Programmatic sells one impression, at auction, with a chain of files that establish who was authorised to sell it. Measurement published on June 22, 2026 found that buyers using artificial intelligence agents took part in 86% fewer auctions than conventional demand while clearing prices 13.4% lower. The agentic stack now being standardised does not extend the programmatic model. It reproduces the television one.

The two mindsets, defined precisely

The distinction is not about screens. It is about what the traded unit is and when it is verified.

The television mindset buys volume. A buyer commits budget months ahead against projected audience, the price is negotiated bilaterally, and delivery is reconciled afterwards against a third-party figure. Verification is periodic and statistical. Nobody inspects an individual exposure, because the market never promised that any individual exposure could be inspected. What holds the system together is a shared currency and the counterparties' willingness to accept it.

The programmatic mindset buys a unit. Each impression is offered separately, competing bidders establish the clearing price, and a set of public files establishes provenance. Ads.txt declares who may sell a domain, sellers.json identifies the seller, the SupplyChain object records every intermediary that touched the request, and the Open Measurement SDK reports whether the advertisement rendered. Verification is per transaction, and it is possible because the transaction is atomic and observable.

Each model has an obvious failure. Television cannot tell a buyer which household saw which advertisement. Programmatic generates enormous transactional waste: DataBeat has recorded 46% auction duplication at top supply-side platforms, and IAB Spain's supply-side platform guide found only 41% of budget reaching working media in the chains it examined.

The industry has spent 2026 building infrastructure to fix the second problem. The architecture it has chosen resembles the first.

The agentic transaction is a negotiated transaction

IAB Tech Lab released AAMP 2.0, the second major version of its Agentic Advertising Management Protocols, in April 2026. The release added five transaction types. Two of them are agentic private marketplaces and agentic Programmatic Guaranteed.

Programmatic Guaranteed deals are negotiated directly between a buyer and a publisher, with price and volume fixed in advance, and they bypass the open auction entirely. An agent that concludes one has not won anything. It has agreed terms.

Ari Paparo, the advertising technology veteran, identified exactly this as the protocol's strongest commercial application in his November 3, 2025 analysis of the Ad Context Protocol. Buying agents seeking well-defined signals, he argued, could pre-negotiate data-driven guaranteed deals across many publishers, because the agent "reduces the per-publisher cost of doing business" and the execution path can beat programmatic auctions. Augustine Fou, a fraud researcher, entered the opposite objection in the same coverage, arguing that "more automation means less transparency" and that agents still act for principals with their own incentives.

The strategic reading came from the company with most to lose. Jeff Green, chief executive of The Trade Desk, told the second quarter earnings call on August 6, 2026 that competitors had focused on guaranteed, fixed-price and simple deal products, and that "Some are even wrapping these in agentic technologies." He drew the binary explicitly: guaranteed transactions solve for certainty, simplicity and an upfront price, while decisioned buying optimises for business outcomes through data, measurement and real-time optimisation.

That is the television argument and the programmatic argument, stated by a named executive, about products shipping now. Rivals are offering the guaranteed-plus-agentic combination at fees near one percent, against roughly twenty percent captured by the incumbent platform. The Trade Desk reported second quarter revenue of seven hundred and fifteen million dollars, up three percent, and guided third quarter revenue to at least six hundred and fifty million dollars, roughly twelve percent below the same quarter a year earlier.

The number that measures the shift

DataBeat, a programmatic analytics provider, builds a monthly United States trends series from anonymised data across a network tracking more than fifty-five million dollars in monthly revenue, thirty-five billion monthly impressions and signals from over two hundred bidders. Its June 2026 edition, covering May data, compared agentic and conventional buying directly.

Conventional buyers cleared at an average of six dollars and ninety-five cents per thousand impressions. Agentic buyers cleared at six dollars and thirteen cents, a 13.4% premium for conventional demand. On fill rate, agentic buying led at 0.204% against 0.183%, an 11.5% edge.

The report then records that agentic buyers reached those results while participating in "86% fewer auctions" than their non-agentic counterparts.

Fewer, larger, pre-arranged transactions producing comparable delivery is a precise description of how television has always been bought. The measurement does not prove agents are worse. It proves they are transacting differently, and the direction of that difference is away from competitive per-impression pricing.

Verification becomes attestation

The clearest evidence that the agentic stack has inherited television's epistemology sits in what IAB Tech Lab shipped on July 30, 2026 as AAMP 2.3, in a release authored by chief operating officer Shailley Singh. Anthony Katsur, chief executive of the organisation, framed the version around "making them reliable in the environments where advertising actually happens."

A pricing provenance field was added to stop buying agents fabricating bid prices when real market data is unavailable. Server-side trust verification now applies on every price-moving path. Deterministic negotiation guardrails were tested against end-to-end negotiation. A vendor approval gate tied to the IAB Diligence Platform and SafeGuard Privacy was embedded in the Buyer Agent.

Consider what those controls concede. In an open auction, a price cannot be invented, because it is produced by other bidders. The need for a provenance field is a direct consequence of removing the mechanism that made prices self-evidencing. Trust that competition used to supply automatically must now be asserted, recorded and audited.

That is currency logic. Television did not verify prices per spot either; it agreed a number and built institutions to defend its credibility. The agentic market is building the same institutions, in software, at speed.

Distribution widened alongside. AAMP 2.3 supports deployment through Amazon Bedrock AgentCore and Databricks, adds Meta buying support and Google Ad Manager reporting integration, and incorporates a Deals Sync server donated by HyperMindz and a Content Taxonomy Parser donated by Mixpeek. Stephanie Layser, sub-industry leader for advertising at Amazon Web Services, characterised the barrier as shipping rather than building, saying organisations "don't struggle to build agents; they struggle to ship them."

Containerisation makes the same move underneath

The infrastructure programme runs in parallel and points the same way.

IAB Tech Lab launched its Containerization Project on June 4, 2025, with Katsur saying growth had "pushed the current framework upon which programmatic is built to its limits." The Agentic RTB Framework version 1.0 followed on November 13, 2025, defining container runtime behaviour and an interface for bidstream mutation. Participants included Index Exchange, OpenX, The Trade Desk, Chalice, Amazon Ads, Netflix, Paramount, Magnite and PubMatic.

Production followed. Zillow piloted containerised bidding with Chalice and Index Exchange in August 2025. Bedrock Platform became the first demand-side platform to run its bidder inside an exchange, each container cryptographically signed by the partner and verified by the exchange. PubMatic launched Decision Fabric on June 1, 2026, and Magnite subsequently offered two routes for partner models to run inside its auction.

Containerisation genuinely restores impression-level computation, which is the case Index Exchange has made publicly. What it does not restore is per-impression observability by anyone outside the runtime. The bid request stops being the unit of information exchange. Decisioning happens inside a signed container that neither counterparty inspects by design, and the guarantee is the signature rather than the record.

Michael Barrett, chief executive of Magnite, has described the coordination layer above the transaction as one that "also handles settlement, policy enforcement, pricing, yield management, identity" alongside payments and compliance, reaching for the New York Stock Exchange as an analogy in which floor trading gave way to electronic execution while clearing survived. Clearing houses are institutions of trust. They are not auctions.

Where the analogy breaks

The comparison has limits, and the evidence sets them.

Agentic guaranteed deals can be data-driven in ways television upfronts never were. Audience segments, contextual signals and outcome optimisation still apply inside a fixed-price transaction. AAMP 2.3 declared Agentic Audiences ready for programmatic and agentic transactions using OpenRTB-based embeddings with a Prebid module, which is a targeting capability television lacked entirely.

The efficiency case is also real, and Paparo's version of it is not a marketing claim. Removing duplicated auction paths and per-publisher negotiation overhead has measurable value against 46% duplication and a 41% working media ratio.

The honest formulation of the disagreement is narrow: whether removing per-impression comparison is a cost saving or a value transfer. Nothing published so far settles it. The DataBeat dataset measures price paid, not business result, and no comparable public outcome dataset exists.

Adoption remains thin where it would matter most. Research covered in April 2026 found agentic AI reaching 46% of surveyed agencies, concentrated in ideation at 86.9% and research at 84.0%, against media planning at 29.1% and media buying strategy at 22.1%. Katsur himself warned in late December 2025 that the industry was chasing agentic AI while transparency, privacy and measurement remained unresolved.

Connected television already ran this experiment

There is a control group, and it is instructive.

Streaming inventory moved onto programmatic pipes while retaining television's transactional habits. IAB data cited in a Gracenote report placed eighty-five percent of connected television inventory as programmatically purchased, up from seventy-five percent a year earlier. Yet a March 2026 Peer39 analysis found only 40% of CTV bid requests carry usable programme-level signals.

The consequences are documented. IAB research published in July 2026 found 43% of CTV buyers doubt where their advertisements actually ran, with confidence at 57% even for publisher-direct and guaranteed deals and 33% for open exchange. DoubleVerify measurement found 34 of every 100 monitored CTV impressions ran outside streaming contentand recorded CTV fraud schemes rising 140% year over year.

Connected television adopted programmatic plumbing and television verification norms simultaneously, and the verification norms won. Products such as IAS Total TV and FreeWheel's series-level tool exist to retrofit the transparency the pipes were supposed to provide.

Why this matters for the marketing community

Three exposures follow for buyers, and none requires taking a position on whether agents work.

The first is price. Agentic demand cleared 13.4% below conventional demand in the only independent comparison published. For publishers, a growing agentic share at that gap lowers average clearing prices even where fill improves. For advertisers, the same figure is either efficiency or under-competition, and the dataset cannot distinguish them.

The second is auditability. Contracts written against programmatic assumptions presume a bid request, a bid response and a supply chain object that third parties can inspect. Agent-negotiated guaranteed transactions executing inside signed containers satisfy none of those assumptions, and the compensating controls are attestations recorded by the counterparties rather than artefacts a buyer's verification vendor can independently read.

The third is fee structure. A one percent agentic guaranteed product against a twenty percent take rate is a repricing of the buy side, not merely a new feature. Whether it is also a downgrade in decisioning quality is the question Green posed and nobody has answered with data.

The television mindset was never irrational. It was a reasonable settlement for a market that could not observe individual delivery. Programmatic's claim was that observation had become possible, and that competition per impression would price inventory better than negotiation per campaign. The agentic stack is being built on the assumption that the observation was too expensive to be worth its cost. That assumption may prove correct. It has not yet been tested against outcomes, and the market is standardising it regardless.

Timeline

Summary

Who: IAB Tech Lab, led by chief executive Anthony Katsur, with AAMP 2.3 released in a note by chief operating officer Shailley Singh. Jeff Green of The Trade Desk and Michael Barrett of Magnite supplied the competing strategic framings. Ari Paparo and Augustine Fou provided the earliest structural critiques. DataBeat produced the independent performance comparison. Container work involves Index Exchange, OpenX, Chalice, Amazon Ads, Netflix, Paramount, Magnite and PubMatic.

What: Television advertising buys negotiated volume settled against a currency; programmatic buys one impression at auction with per transaction provenance. The agentic advertising stack standardised through 2026 reproduces the television model. AAMP 2.0 added agentic Programmatic Guaranteed and private marketplace transaction types, both of which fix price and volume in advance and bypass the open auction. AAMP 2.3 added a pricing provenance field, server-side trust verification and a vendor approval gate, controls made necessary precisely because competitive bidding no longer produces the price. Containerisation relocates decisioning into cryptographically signed runtimes that outside parties do not inspect.

When: The Containerization Project launched on June 4, 2025 and the Agentic RTB Framework entered public comment on November 13, 2025. AAMP was named on February 26, 2026, reached version 2.0 in April 2026 and version 2.3 on July 30, 2026. DataBeat published its comparison on June 22, 2026, and Green spoke on August 6, 2026.

Where: Specifications are maintained globally through IAB Tech Lab working groups and public repositories. The DataBeat measurement and the connected television research cover the United States market.

Why: Agentic buyers took part in 86% fewer auctions than conventional demand while clearing 13.4% lower, a transaction pattern closer to television than to real-time bidding. Connected television has already demonstrated the outcome of adopting programmatic pipes with television verification norms: 85% of inventory trades programmatically, only 40% of bid requests carry programme-level signals, 43% of buyers doubt where advertisements ran, and 34 of every 100 monitored impressions ran outside streaming content.