Podcast episode

Fruit of the Poisonous Tree

by Ken Rona ·July 13, 2026 ·AdTalk

dsp programmatic publisher-economics ssp walled-gardens

TL;DR

Adam and Gareth (Chalice, an ad optimization company) dig into Google's new "Buyer Direct" product inside Google Ad Manager (GAM, the dominant publisher ad server), arguing it could render agentic ad-buying workflows irrelevant by offering a simpler, cheaper direct-deal pipeline — all built on infrastructure cemented by Google's now-adjudicated monopoly. The episode also covers Warner Bros. Discovery building an ad stack on AWS, publishers blocking Google AI crawlers, and CTV pricing inefficiencies.

What was covered

AI "slop" on LinkedIn: Both hosts open with frustration at the volume of LLM-generated takes flooding professional social media, arguing it's now harder to identify genuine expertise than ever; they frame it as a "zone flooding" problem structurally identical to disinformation tactics.
Alex Karp / Palantir ontology angle: The hosts walk through Palantir CEO Alex Karp's academic background — doctorate in social theory under Jürgen Habermas in Frankfurt — and connect it to Palantir's core business: arriving at enterprises to build unified data taxonomies (ontologies) before any AI model is deployed. Key claim: a huge, underreported portion of frontier-model training spend goes to manual data labeling by knowledge workers, not compute.
Jounce Media and Sincera: Praised for creating digestible, standardized quality taxonomies for ad placements (e.g., MFA — Made for Advertising — classifications), described as translating raw ad-placement data into executive-readable frameworks. Gareth credits Sincera with building the underlying data set first; Jounce with making it publicly legible.
Google Buyer Direct in GAM: Google has a live (though limited-beta, agency-only) product that lets buyers select publishers and run guaranteed deals directly inside GAM — no DSP required. Reported take rate: ~10%, far below typical programmatic fees (SSP fees alone often 10–20%). Gareth's thesis: because GAM is installed at ~90% of publishers as a result of the antitrust-adjudicated tying of GAM and AdX, Buyer Direct has a structurally insurmountable distribution advantage over any agentic workflow built outside Google's stack.
"Automation doesn't beat architecture": Gareth's central argument — LLM-based agentic media-buying tools requiring multiple API calls, MCP handshakes, and cross-system negotiation will always lose to a unified architecture like GAM + Buyer Direct that eliminates those steps entirely.
Warner Bros. Discovery building ad stack on AWS: Under-the-radar story (sourced from Tip Sheet AI): WBD is constructing an ad-tech stack on AWS infrastructure, potentially including agentic buying capabilities via AWS Bedrock. Hosts connect this to Jeff Green's prediction that Amazon may not have a standalone DSP in five years, but will own critical cloud infrastructure for ad decisioning.
CTV mispricing / information asymmetry: CTV is described as structurally immature — still largely lacking show-level data in programmatic bids, similar to early open-web programmatic that withheld page-level data. As show-level signals proliferate, pricing should tighten; IP-address-based audience targeting is flagged as a recurring bad habit being re-imported into CTV.
Publishers opting out of Google Search / AI crawlers: AdWeek story cited: publishers increasingly using tools (including Cloudflare-based crawler controls) to block AI scrapers. Gareth cites Newsweek down ~80% in traffic. Adam reframes it: Google's AI Overviews are the symptom, but Google has been displacing publisher traffic with on-SERP answers (timers, date lookups) long before LLMs. Publishers now have no traffic "gift" left to justify giving Google their content for free.

Notable claims & predictions

Gareth: "Agentic will be killed in the crib by Google's Buyer Direct product, which has basically all the promise of agentic and infinitely less friction… automation doesn't beat architecture." — Core thesis of the episode; stakes are high if correct.
Gareth: "Google's 10% take rate [on Buyer Direct] is fruit of the poisonous tree — a massive competitive advantage born directly from Google's monopolistic tying of AdX and GAM." — Frames low pricing not as pro-competitive but as predatory advantage from already-adjudicated illegal conduct.
Adam (paraphrasing Jeff Green): "Jeff Green said he doesn't think Amazon will have a DSP in five years" — Cited as context for why WBD is building on AWS cloud rather than buying into a DSP relationship.
Gareth: "We are currently building all of our agentic tech in the image of Google until the antitrust suit is done and the tying is obviated. The last mile is owned by GAM in this world." — His own tweet, read on air.
Adam: "Chalice is spending about $5,000 a month on Anthropic and doing a ton with it" — Positioned as a reality check against narratives of massive AI infrastructure costs for mid-sized ad-tech companies.
Gareth: "CTV is very immature — it's mispriced right now because there is tremendous information asymmetry. As show-level data pervades the market, pricing will get better and more direct-sold deals will face pressure from programmatic."

Fact check

Gareth's claim that GAM is installed at "~90% of publishers" Unverified, but widely cited as an industry approximation. Independent audits vary; the claim is directionally consistent with the DOJ's own findings in the Google ad-tech antitrust case, which described GAM as dominant among large publishers. Treat as plausible but imprecise.

Adam's claim that Newsweek is "down 80%" due to Google AI Overviews Unverified as stated. Several large publishers have reported significant organic-search traffic declines attributable to AI Overviews (Dotdash Meredith and HouseFresh have reported steep drops publicly), but the specific "Newsweek 80%" figure is not corroborated in any named study in the transcript. Listeners should note Adam doesn't cite a source.

The framing of Karp as having studied under Habermas and written his doctorate in German True and verifiable. Karp received his doctorate from Goethe University Frankfurt; his adviser was Jürgen Habermas. This is well documented.

Adam's claim that Google Buyer Direct is "only available to some ad agency customers right now" Consistent with publicly available GAM documentation as of early 2025, but the product's exact availability status evolves. Readers should verify current access tiers directly with Google.

Chalice's position: Both hosts are co-founders or senior executives at Chalice, a company that sells custom optimization algorithms to advertisers — a product that competes with both Google's unified architecture and the "overlord" model they critique. Nearly every structural argument in this episode (fragmented programmatic needs custom algorithms; GAM/Buyer Direct is dangerous; agentic workflows need independent optimization layers) directly supports Chalice's commercial interests. Their critiques of Google may be accurate, but the incentive to frame Google consolidation as dangerous is substantial and should be weighted accordingly.

Full analysis

Google has quietly shipped a product — Buyer Direct, inside its dominant publisher ad server (Google Ad Manager, the software ~90% of large publishers use to sell their inventory) — that lets an agency pick publishers and run guaranteed deals directly, with no demand-side platform (DSP, the software buyers normally use to purchase ads) in the middle. Reported take rate: about 10%, versus the 20–40% that stacks of middlemen typically skim. The claim on the table: this could smother the whole "agentic" ad-buying wave — AI agents that automate media buying across systems — before it grows up, because a single unified pipe beats a chain of AI handshakes on both price and friction.

This is a Type 1 question for anyone building outside Google's stack (hard to reverse — you're picking an architecture), and a Type 2 for buyers (easy to test a Buyer Direct deal and walk away). What's actually being decided across the ecosystem: whether independent optimization, agentic tooling, and the SSP layer have a durable role, or whether Google's already-adjudicated dominance simply migrates from open programmatic into direct deals under a cheaper label.

One caveat load-bearing enough to state up front: both hosts run Chalice, which sells independent optimization. Their "Google consolidation is dangerous, independent layers are essential" framing is exactly what their business needs to be true. Weight accordingly.

The Market Analyst — The interesting tension is that Buyer Direct at a 10% take rate is deflationary for the middle of the stack, and the middle is where the public independents live. If guaranteed direct deals move inside GAM at half the cost, the squeeze lands on SSP fees first — the layer PubMatic, Magnite and Index charge for. The Trade Desk's counter-thesis (independent buyers on the open web, away from walled gardens) gets harder to sell if Google offers "direct, cheap, no DSP" as the default. Plainly: Google is offering to do the same job for less, using rails a court already said it built illegally. That's a pricing weapon, not a product feature.

The Skeptic — The load-bearing assumption is "architecture always beats automation." That's a slogan, not a law. Buyer Direct is limited-beta, agency-only, guaranteed-deals-only — it doesn't touch the messy, biddable, audience-driven buying that's most of programmatic. And "10% take rate" is an introductory number from a monopolist that has raised rates every time it cornered a market. The louder problem: the people making this call sell the alternative. Every structural claim here — fragmentation needs custom algorithms, consolidation is dangerous — happens to require buying what Chalice sells. In plain terms: the fire alarm is being pulled by the people selling extinguishers.

The Operator — Tuesday-morning reality: a media buyer tries Buyer Direct and it just works — pick publisher, set price, done, no trafficking sheet, no DSP seat fee. That's genuinely seductive, and adoption won't wait for a philosophy debate. But it also means the buyer hands Google even more of the workflow, and loses cross-publisher optimization and independent measurement in the process. The 90-day surprise: reporting and controls that live only inside Google's four walls, and no clean way to compare a Buyer Direct deal against anything bought elsewhere. Everyday version: it's easy to move in, hard to check whether you're getting a fair deal, and harder to move out.

The Customer / End User (publisher) — Publishers are the ones actually holding leverage here, and the episode buries the real story: they're turning off Google's AI crawlers (Cloudflare-style blocking, Newsweek reportedly down sharply). The traffic "gift" Google used to justify taking content free is gone. So publishers face two Googles at once — one draining their audience via on-page AI answers, one offering a cheaper sales pipe via Buyer Direct. Meanwhile Disney, Netflix and Warner Bros. Discovery (now building on AWS) have already walked away from GAM entirely. Plainly: the biggest publishers are voting with their engineering budgets to not depend on Google.

Where they part ways:

Is 10% a floor or a bribe? The Operator says the low price and zero friction drive real adoption now. The Market Analyst and Skeptic say a monopolist's introductory rate is the setup, not the equilibrium — the whole "fruit of the poisonous tree" point is that the price is cheap because the distribution was won illegally, and cheap prices from monopolists don't stay cheap.
Does Buyer Direct actually kill agentic? The hosts say yes. The Skeptic and Customer say it kills agentic for guaranteed direct deals inside Google's stack — a real but narrow slice. The largest CTV publishers building their own stacks on AWS are the living counterexample: the premium end of the market is fragmenting away from Google, which is precisely where agentic coordination across systems has value.
Who's the victim? Framed as "agentic startups." The more exposed party is the independent SSP layer, whose fees Buyer Direct directly undercuts.

What it hinges on: whether Google holds the take rate low as adoption grows (a monopoly-pricing question), and whether guaranteed direct deals are a big enough slice of spend to matter (they're a minority of programmatic today). Before anyone re-architects around this: run a live Buyer Direct deal, measure the all-in cost and the reporting you lose, and watch whether the antitrust remedy forces GAM/AdX apart — because the entire distribution advantage rests on that tie surviving.

The council leans skeptical of the "killed in the crib" headline but takes the pricing threat to the SSP middle seriously. The real signal isn't in the drama; it's that the biggest publishers are already leaving.

Prediction: Google will not expand Buyer Direct to general availability (open, self-serve access beyond invited agency accounts) before its next major GAM/Marketing Live announcement cycle in mid-2026 — it stays gated while the ad-tech antitrust remedy is unresolved.

Confidence: Medium — Google won't scale a monopoly-leverage product while a judge weighs breaking up that exact stack.

Why: The product's whole advantage comes from GAM's court-adjudicated dominance and the AdX tie, and that tie is the live subject of the remedies phase; broadly rolling out a product that visibly monetizes the tie invites the court to cite it as ongoing harm. That legal exposure is a strong reason to keep Buyer Direct in a controlled, deniable beta rather than push it to every publisher — the opposite of what you'd do with a clean product. The less likely outcome — a full open launch — would hand plaintiffs a gift-wrapped exhibit, which is why Google's own lawyers make it improbable.

Revisit by 2026-12-31: We're right if Buyer Direct remains limited/invite-only through year-end. We're wrong if Google opens it to general self-serve availability or announces broad rollout before then.

Listen to the episode →