Jeff Green used The Trade Desk's August 6, 2026 earnings call to describe a category of competitor building fixed-price and guaranteed transaction products, wrapping them in agentic technology, and competing on fee rather than outcome. He compared the result to ad networks of 2006. The standards bodies had already written guaranteed deals into the agentic specification four months earlier.

The passage arrived roughly two thirds of the way through prepared remarks, immediately after a section on product roadmap and immediately before an explanation of why decisioned buying and guaranteed buying are different products.

"Some of our competitors, big and small, have been focused on producing products for Programmatic Guaranteed, fixed price transactions, and simple deals with approaches that do not leverage buyer decision," Green said. "Some are even wrapping these in agentic technologies. They argue that their cheap fee will compensate for their lack of objectivity and their decisions that favor publishers rather than the buyer who is footing the bill. These approaches look more like ad networks of 2006 than reflect the progress that our industry has made in the last 20 years."

No competitor was named in the passage. The categories described, however, map onto identifiable products, several of which have shipped within the past twelve months, and one of which sits inside a standard published by the industry's own technical body.

What the claim actually asserts

The statement makes four separate assertions, and they are worth separating because they carry different evidentiary weight.

First, that competitors are building products oriented toward Programmatic Guaranteed and fixed-price transactions. Second, that some of those products are being packaged with agentic technology. Third, that the vendors involved are arguing a low fee compensates for a lack of objectivity. Fourth, that the decisions those systems make favour publishers over the buyer paying for the media.

The first two are documented. The third is a characterisation of an argument. The fourth is a claim about how competing systems allocate spend, and it is the one that carries the most commercial weight while being the hardest to verify from outside.

Green expanded on the distinction shortly afterwards: "Decision buying and Programmatic Guaranteed are fundamentally different products. Programmatic Guaranteed can solve for certainty, simplicity, and upfront price, but decision buying optimizes for the highest business outcomes utilizing data, measurement, and real-time optimization."

That framing sets up a binary. Guaranteed transactions deliver certainty at a known price. Decisioned buying delivers outcomes through real-time evaluation. The argument for a demand-side platform's fee rests entirely on the second column being worth more than the first.

Why this appears in an earnings call

The passage is not an abstract industry critique. It sits directly downstream of the quarter's results.

The Trade Desk reported second quarter revenue of $715 million, up 3% year over year, and guided third quarter revenue to at least $650 million, roughly 12% below the same quarter a year earlier. Earlier in the same remarks, Green attributed part of the shortfall to advertisers under budget pressure changing how they buy.

Some brands, he said, are "falling prey to low-cost, low decisioning methods like Programmatic Guaranteed and fixed price." He continued: "Doing so essentially means buyers will give away their decisioning in a great buyer's market to the sellers in exchange for lower cost of transactions. This approach is often deliberately short-sighted."

That is a direct link between the transaction-type argument and the revenue line. If pressured advertisers are moving budget from auction-based decisioned buying into guaranteed and fixed-price arrangements, a platform that monetises decisioning captures less of that spend. The competitive critique and the growth explanation are the same argument.

The fee gap Green is describing

The company does not disclose its take rate. Reporting has placed Amazon DSP at approximately 1 to 2% of media spend against a 12 to 15% range typically associated with The Trade Desk. More specifically, Amazon charges a 1% fee for open web publisher ads through its DSP and no fee at all for programmatic guaranteed deals on its own owned-and-operated media.

That last detail is the sharpest version of what Green is describing. On a guaranteed deal against Amazon's own inventory, the platform fee is zero. There is no fee to compare against a decisioning fee, because there is no fee.

Green addressed the arithmetic later in the call, arguing that platform rates cannot be compared in isolation because competitors bundle their fee with media. His illustration used an 8% fee against a 4% fee, producing an impression cost of $1.08 against $1.04 on underlying media of $1. The narrower the fee spread, the less incremental performance a decisioning platform must deliver to justify itself. At 1% against 12%, the comparison becomes $1.12 against $1.01, and the required performance edge rises accordingly.

Guideline data tracked through the first quarter of 2026 showed Amazon moving from under 10% to just under 20% of tracked programmatic spend in roughly fifteen months, with The Trade Desk growing at approximately the market rate.

The one exchange where competitors were named

Prepared remarks kept the criticism anonymous. The question-and-answer session did not.

Dan Salmon of New Street Research put the two largest names directly to Green, asking why an independent platform built on objective decisioning can continue winning share against walled gardens combining exclusive live sports inventory, as Amazon does through its own demand-side platform, or offering simple guaranteed-style transactions at low pricing, as Salmon characterised Google's buyer direct programme.

Green's answer began by conceding the market has room for several winners. "There will be many winners. There have to be," he said, adding a structural point about auction dynamics: "An auction with only one bidder isn't an auction."

He then disputed the significance of walled-garden volume. "We have by far the highest market share in the programmatic space, especially as we're looking at the open internet," he said. "If you want to say, 'Well, yeah, but Google buys a lot of YouTube,' I would say, 'Well, I would hope so.' Amazon buys a lot of Amazon Prime and sponsored listings. Well, I would hope so. They're the only place where you can buy them."

The substantive part of the answer was an argument about margin structure, and it is the mechanism underneath the assertion that competing systems favour publishers over buyers. In Green's framing, the favoured publisher is the platform itself.

"When you have something like YouTube, where their cost of goods sold is almost zero," he said. "If they get $1, they can spend it on YouTube and keep the whole dollar. They can spend it on Disney+ and they keep $0.10."

The conclusion followed: "Of course, it would be better for a Google or an Amazon to buy their owned and operated inventory. That's what they do, because they make more money that way. None of these companies are in the community service business, if you will. They're not giving things away. When things come out as free, they are not really free. They're just moving the cost somewhere else."

Green did not claim the competing products are poor. "It doesn't mean that those companies can't have great products," he said. "It doesn't mean that they won't have an ancillary business. At the core, they are selling owned and operated inventory. Both of them make most of their money from their owned and operated inventory."

That is a narrower and more defensible claim than the one in prepared remarks. It does not assert that guaranteed transactions are technically regressive. It asserts that a platform earning a full margin on its own inventory and a thin margin on everyone else's has a measurable financial reason to allocate spend toward the former, and that a zero fee on a guaranteed deal against owned inventory is where that incentive is most visible. The economics are verifiable from public filings. The allocation behaviour that would follow from them is not disclosed by any platform.

The agentic wrappers already exist

The second assertion, that competitors are wrapping guaranteed transactions in agentic technology, describes something that has been built in public over the past ten months.

The Ad Context Protocol launched on October 15, 2025 with six founding members: PubMatic, Scope3, Swivel, Triton Digital, Optable and Yahoo, plus 23 additional participants, built on Anthropic's Model Context Protocol. Its nine core tasks span inventory discovery through campaign activation and reporting. Magnite embedded a seller agent into SpringServe and completed its first agentic test in December 2025 with Scope3 as buyer agent. PubMatic launched AgenticOS on January 5, 2026 with live campaigns running. Prebid.org took stewardship of an open-source sales agent on January 29, 2026, on the argument that publishers without representation in machine-to-machine buying protocols risk becoming price-takers.

Several of these are supply-side businesses building agents that represent sellers. That is the structural basis for Green's fourth assertion, that the decisions favour publishers. A seller agent is, by construction, operating on behalf of the seller.

The more direct evidence sits in the standards themselves. IAB Tech Lab formally named its umbrella initiative the Agentic Advertising Management Protocols, or AAMP, on February 26, 2026, in a post by chief executive Anthony Katsur, covering three pillars: execution, protocols and an Agent Registry. In April 2026, AAMP 2.0 added five new transaction types, among them agentic private marketplaces and Programmatic Guaranteed deals.

That is the specific thing Green described. Guaranteed transactions, executed by agents, written into a standard by the industry body, four months before the earnings call. The architecture is not a competitor's proprietary idea. It is the published direction of travel.

The registry side is smaller than the rhetoric on either side implies. The AAMP Agent Registry reached 10 active entries on March 11, 2026, adding Amazon, Burt Intelligence, Optable, Dstillery and HyperMindZ.ai, all classified as MCP servers with no agent-to-agent entries at that point. Kochava opened its StationOne platform to public beta on March 25, 2026 as the first accessible sandbox for testing standards-compliant agentic workflows without real spend.

Green is not arguing against agentic technology

The passage is frequently readable as hostility to agentic advertising. On the call, it was not.

Asked by Matt Swanson of RBC whether AI threatens the demand-side platform model, Green called agentic "one of the biggest opportunities that advertising will ever see" and said the company is "already seeing massive advantage from it." His objection was narrower: "Some are using agentic to just build yesterday's business model all over again, if you will. They're building ad networks out of agentic-like technologies. That's not going to work because it doesn't leverage the very best of decisioning, which is what a DSP does."

He also redefined the category to absorb the threat: "A DSP is a platform built to decide which of those impressions you buy and which you don't. Of course that is enhanced by AI, and that's why we've been investing in AI for years. I wouldn't say that the DSP model, if you will, is going to be disrupted by AI. It is AI."

The distinction being drawn is between agents that decide and agents that transact. An agent negotiating a guaranteed package at a fixed price is performing procurement. An agent evaluating impressions against each other is performing decisioning. Green's claim is that the first is being marketed as the second.

Testing the 2006 comparison

An ad network in 2006 aggregated publisher inventory, sold it to buyers at a rate the network set, and kept the difference. Buyers did not see impression-level pricing, could not compare one opportunity against another in real time, and had limited visibility into which sites their money reached. The network's incentive was to fill its own inventory at the highest achievable margin.

The comparison holds on two points and fails on two others.

It holds on decisioning granularity. A guaranteed deal at a negotiated price removes impression-level comparison by design. That is the product's purpose. It also holds on incentive alignment where the seller operates the transaction layer, which is the case for owned-and-operated inventory sold through a platform's own buying tools.

It fails on transparency. A 2006 network disclosed almost nothing. Contemporary guaranteed deals carry deal IDs, log-level reporting in most environments, and verification tooling. It also fails on the direction of standardisation. AAMP 2.3, released in late July 2026, added a pricing provenance field designed to prevent AI agents from inventing prices, alongside a vendor approval gate tied to the IAB Diligence Platform and SafeGuard Privacy. That is the opposite of a black box, and it is a direct response to the concern Green articulates.

The most substantive challenge to the binary comes from a different direction. Amazon Publisher Cloud introduced Outcome Optimizer on June 19, 2026, applying shopping, browsing and streaming signals to programmatic guaranteed deals inside FreeWheel's ad server, with tests showing a 33% improvement in on-target reach. That product adds decisioning to a guaranteed transaction. If signal-driven optimisation can operate inside a guaranteed deal, the categorical separation between certainty-buying and outcome-buying narrows, and the argument shifts from a difference in kind to a difference in degree.

The evidence on price

One dataset speaks to whether agentic demand behaves the way Green describes.

DataBeat's June 2026 report, drawn from May 2026 data across a network tracking more than $55 million in monthly revenue, 35 billion monthly impressions and signals from over 200 bidders, found conventional programmatic buyers clearing at an average CPM of $6.95 against $6.13 for agentic buyers, a 13.4% premium for conventional demand.

That figure is genuinely ambiguous, and both sides can read it favourably. Agentic buyers acquiring impressions at lower clearing prices is consistent with Green's claim that these systems optimise for cost rather than outcome. It is equally consistent with the opposite claim, that agentic systems are simply more efficient at avoiding overpayment. The dataset measures price paid, not business result, and no comparable outcome dataset exists publicly.

The same report characterised agentic demand as tracking close to conventional programmatic buying rather than beating it. On volume, agentic buying remains small relative to the rhetoric surrounding it.

What Green did not address

Three counterpoints went unmentioned in the remarks.

The first is buyer demand. The third annual State of Programmatic Report, surveying November 2025 and released January 20, 2026, found three quarters of respondents considering curated marketplaces important to 2026 strategies, alongside 82% citing AI-powered optimisation as essential when evaluating partners. Movement toward curated and deal-based supply is not solely a vendor push. Buyers report wanting it.

The second is that the largest competing platform has been narrowing manual deal creation rather than expanding fixed-price simplicity. Google deprecated manual creation of non-Programmatic Guaranteed deals in DV360 around April 2026, making the Deal Sync API the exclusive route for new deals. That is workflow automation of the deal layer, not a retreat from decisioning.

The third is the company's own participation in the same transaction types. The Trade Desk operates OpenPath, its direct publisher integration, at a flat 4.5% fee, which Green told investors in February is meant to be nearly breakeven to slightly profitable. It launched Deal Desk in June 2025 to manage private marketplace deals inside Kokai and secured publisher backing for the OpenAds auction platform in January 2026. When Samsung Ads opened Smart TV home screen inventory through The Trade Desk and DV360 on June 10, 2026, the announcement did not specify whether the inventory would transact through open auction, private marketplace or programmatic guaranteed. Green also disclosed on the call that the company is building white label agentic AI products with agency partners.

The company sells guaranteed and deal-based transactions. Its argument is about where the decisioning sits, not about the existence of the transaction type. The remarks did not make that distinction explicit.

The critique has independent support

Green is not alone in questioning whether agentic infrastructure addresses the industry's actual problems.

Debate over the Ad Context Protocol divided the industry within weeks of launch. Katsur of IAB Tech Lab questioned whether another trade body was needed, noting that existing open-source structures had already solved some of the problems the initiative targeted. Fraud researcher Augustine Fou warned that more automation means less transparency and that agents can still act for people with bad incentives. Lindsay Rowntree of ExchangeWire described agentic AI as a giant black box, according to reporting on the protocol's reception. David Kohl characterised the protocol as the tail wagging the dog, arguing it addressed automation before fixing structural issues.

Ari Paparo's November 3, 2025 analysis is the closest to Green's structural point while reaching a different conclusion. Paparo described a model in which cross-publisher entities package inventory, negotiate with buy-side agents and assume delivery and outcome risk, a pattern he called curation by wire, and identified pre-negotiated data-driven programmatic guaranteed deals across many publishers as a genuinely valuable agentic application because it reduces per-publisher cost of doing business. Paparo had argued in July 2025 that autonomous systems could automate the campaign setup, targeting and optimisation functions demand-side platforms currently perform.

That is the same architecture Green describes, assessed as efficiency rather than regression. The disagreement is not about what is being built. It is about whether removing per-impression comparison is a cost saving or a value transfer.

What would settle it

The argument reduces to a measurable question: whether decisioned buying produces enough incremental business outcome to justify a fee that may be ten times higher than the alternative.

Green acknowledged on the same call that the company's ability to demonstrate that is incomplete. Describing measurement as possibly the biggest problem in the industry, he said that while last click and last view remain the standard, "the most premium parts of the open internet will always look expensive and ineffective." He added later: "It's that last piece where I think we've missed a little bit in the last few years."

The measurement framework intended to close that gap is in alpha, built with unnamed media, measurement and data partners, with no disclosed timeline.

Until it produces comparable outcome data, the disagreement runs on assertion from both directions. Vendors selling guaranteed transactions at near-zero fees point to cost. A platform selling decisioning points to outcomes it cannot yet fully evidence. Advertisers under margin pressure, in the categories Green identified as most affected, are choosing between a number they can see today and a benefit they are asked to accept on argument.

The passage about 2006 ad networks is, read that way, less a competitive dismissal than a description of the conditions under which the company's pricing has to be defended.

Timeline

Summary

Who: Jeff Green, Chief Executive Officer and co-founder of The Trade Desk, Inc. (NASDAQ: TTD), speaking on the company's second quarter 2026 earnings call. The remarks describe unnamed competitors building guaranteed and fixed-price transaction products, a category that includes supply-side platforms, walled-garden demand platforms and agentic protocol participants including PubMatic, Magnite, Scope3, Amazon and Google.

What: A prepared-remarks passage asserting that competitors are producing Programmatic Guaranteed, fixed-price and simple deal products that do not use buyer decisioning, that some are wrapping these in agentic technology, that they argue a cheap fee offsets a lack of objectivity, and that their decisions favour publishers over the buyer paying for the media. Green compared the result to ad networks of 2006.

When: The remarks were delivered on August 6, 2026, at 2:00 P.M. Pacific Time, during the earnings call for the quarter ended June 30, 2026. IAB Tech Lab had added Programmatic Guaranteed to its AAMP agentic transaction types in April 2026, four months earlier.

Where: The argument concerns the global programmatic advertising market and specifically the open internet inventory pool where The Trade Desk competes, against walled-garden platforms selling owned-and-operated inventory and supply-side platforms operating seller agents.

Why: The passage connects directly to the quarter's results. Green attributed part of a deceleration to 3% revenue growth to advertisers under budget pressure moving toward what he called low-cost, low decisioning methods, naming Programmatic Guaranteed and fixed price. A platform that monetises decisioning captures less value when spend shifts into guaranteed transactions, making the transaction-type argument and the growth explanation the same argument. The underlying question, whether decisioned buying produces enough incremental outcome to justify a fee that may be an order of magnitude higher, remains unresolved because the measurement framework intended to demonstrate it is still in alpha.