Podcast episode

Signal Break: Agentic Trading Is Here

by Ken Rona ·July 13, 2026 ·Signal & Noise

ai-in-adtech dsp programmatic publisher-economics ssp

TL;DR

Patrick O'Leary, founder and CEO of Boostr (an ad sales management platform for publishers), details what he calls the first real agentic media buy — a test campaign executed between AI agents on behalf of Vox Media (seller) and a buyer, using a protocol called AdCP. The episode argues this is the opening shot of a shift that could disintermediate SSPs and DSPs (the software middlemen of programmatic advertising) over a three-to-five year horizon, with publishers capturing a larger share of each ad dollar.

What was covered

The Boostr/Vox Media AdCP pilot: A small display campaign was negotiated and trafficked entirely by buyer and seller AI agents using AdCP (Ad Context Protocol — a standardized messaging framework letting agents communicate across systems, analogous to the 1990s EDI data-interchange standard). Two human touch-points remained: approving the media plan and verifying ad-server trafficking. Total elapsed time from negotiation to trafficking: approximately two minutes, versus a manual process estimated at three to six weeks.
What AdCP is and how it works: The protocol defines standardized request/response message types (RFP, order submission, revision) so buyer and seller systems can negotiate without caring about each other's underlying software stack. A brand.json file hosted on each participant's domain provides initial identity verification; version 3.0/3.1 adds dual-encryption zero-trust authentication.
Human-in-the-loop design: Boostr intentionally preserved two approval checkpoints — media plan review (can we transact with this buyer? billing relationship set up?) and ad-server creative/tag verification — arguing these remain essential to prevent mis-trafficking, which is a common source of make-goods today.
Competitive and market-structure implications discussed: O'Leary predicts DSPs (demand-side platforms — software advertisers use to buy digital ads) and SSPs (supply-side platforms — software publishers use to sell inventory) face existential risk if direct agentic buying scales. He estimates $28 billion annually flows to ad-tech intermediaries ("ad-tech taxes") and argues AdCP could return much of that to publishers as working media.
Scope of deals AdCP currently supports: O'Leary claims 89% of programmatic spend (by dollars) is already PG (programmatic guaranteed — fixed-price reserved deals) or PMP (private marketplace — invitation-only deals), not open-auction RTB (real-time bidding). He argues this majority is the natural early adopter segment for AdCP.
Open issues acknowledged: Brand safety/suitability rules, competitive separation (e.g., preventing United and Delta ads running back-to-back), identity/third-party segments, frequency capping, and buyer-side budget pacing across publishers are still being built into the spec. Scope3's new "Interchange" product is mentioned as a potential payment clearinghouse for participants without direct billing relationships.
Internal org changes at Boostr: O'Leary pulled a small team out of the core product org into a fast-moving incubator structure to build AdCP capabilities; the company is now merging learnings back into the main product. Engineers are now largely AI-assisted coders; QA at the new velocity is the next challenge.

Notable claims & predictions

O'Leary: "If you fast forward three to five years, you won't need DSPs and SSPs." He positions AdCP as structurally doing what programmatic was originally supposed to do — efficient, transparent, low-friction direct buying — before it became what he calls "corrupted."
O'Leary: "Only 11% of the programmatic ecosystem is currently open RTB [by dollars]. The rest — 89% — is PG or PMP, and that's the sweet spot for AdCP." He frames this as meaning the vast majority of spend is already ripe for agentic direct buying.
O'Leary: The entire Vox Media pilot — from agent negotiation through creative trafficking — "took about two minutes," replacing a process that "normally takes three to six weeks" for a human team across multiple departments.
O'Leary on agencies: "The brands are going to look at the role of the agency and see why they've been paying all this money to operate the DSP... I can now do this agentically myself and get more working media." He tempers this slightly, acknowledging scaled aggregated demand still favors agency buyers in the near term.
O'Leary on SSPs specifically: "SSPs probably should go away... this puts their business model most at risk." He anticipates SSP-built agents will be a superficial wrapper ("just a rapper") that doesn't address the underlying structural problem.
Co-host (paraphrased): "The AI we have today is the worst AI we'll ever have" — framing Claude Opus updates as already showing exponential jumps, and arguing the AdCP use case will compound in capability quickly.

Fact check

O'Leary's claim that "only 11% of the programmatic ecosystem is currently open RTB [by dollars]; the rest is PG or PMP." Unverified and likely contested. Industry measurement of open-auction vs. reserved/private deal share varies significantly by source, publisher type, and whether CTV is included. Most public data (e.g., Pixalate, IAB research, DSP reports) does not show open RTB as low as 11% of total programmatic dollars across the open web; the split varies dramatically by vertical and publisher tier. O'Leary attributes this to "research we did recently" without citing a source or methodology. He has a clear incentive to frame the majority of spend as being in AdCP's sweet spot — this number does the rhetorical work of making the TAM (total addressable market) for agentic direct buying look as large as possible. Treat as an internal estimate requiring independent validation.

O'Leary's "$28 billion annually in ad-tech taxes going to middlemen." Unverified / sourcing unclear. Various industry estimates of the programmatic supply-chain "tax" exist (ISBA/PwC's 2020 UK study found ~15% of spend was "unknown" and overall intermediary take was high; subsequent studies have varied). A precise $28 billion global figure is not attributable to any widely cited public source in the transcript. O'Leary does not cite a source. The general direction of the claim — that intermediary take rates are significant — is well-supported; the specific number is unverified.

Claim that the Boostr/Vox pilot was "one of the industry's first" agentic media buys. Contested / unverifiable as stated. The hosts acknowledge Omnicom had already announced experimentation with AdCP and Scope3 for agentic buying (cited to Digiday, "a couple of weeks ago"). O'Leary's framing as a "landmark" first actual completed buy may be accurate as a live transacted campaign vs. a demo or experiment, but the claim is inherently difficult to verify and the competitive landscape is moving fast. Boostr has an obvious incentive to claim primacy.

No claims rise to the level of clearly false.

Full analysis

A publisher ad-sales platform says it ran the first live "agentic" media buy — buyer and seller AI agents negotiating and trafficking a Vox Media campaign in about two minutes using a shared protocol called AdCP (a standardized language that lets ad-buying and ad-selling software talk to each other). The pitch: over three to five years, this cuts the software middlemen — SSPs and DSPs — out of the loop and hands publishers a bigger slice of every ad dollar.

Reversibility: Type 2 for any single operator experimenting. Type 1 at the industry level — once a standard protocol gets traction, it becomes plumbing that's hard to unwind. What's actually being decided: not "does the demo work" (it clearly does) but "how fast does agentic direct buying scale, and which layer of the stack gets thinner." Forcing function: none acute. This is a positioning question with a multi-year clock, but the standard-setting fights (IAB Tech Lab, Scope3's clearinghouse) are happening now.

The Market Analyst — The market has already half-decided the direction: Omnicom and Scope3 announced agentic buying before this pilot, so the "first" claim is marketing, not signal. For a non-specialist: several players are racing to build the same road, and each wants to plant its flag first. The real tell is who controls the protocol. If AdCP settles inside IAB Tech Lab as neutral plumbing, the value accrues to publishers and whoever runs the clearinghouse (Scope3's "Interchange"), not to any one vendor. Watch the public SSPs — Magnite, PubMatic — and Trade Desk. Their stock stories don't crack on a two-minute demo, but they crack the moment a top-10 publisher reports material direct-agentic revenue at a lower take rate. That print is the catalyst, and it's quarters away, not weeks.

The Skeptic — The load-bearing assumption is the "89% of programmatic is PG/PMP, so it's ripe" claim, and the episode's own fact-check flags it as unsourced and likely inflated. For a general reader: the whole size-of-the-prize argument leans on one number the founder made up the framing for. Programmatic guaranteed deals are already low-friction and low-fee — the "$28 billion tax" lives mostly in open-auction RTB, which is exactly the 11% AdCP handles worst. So the protocol is easiest to adopt precisely where the middleman tax is smallest, and hardest where the money actually leaks. That's the tension nobody on the podcast resolved. A protocol that automates the cheap part isn't disintermediation; it's workflow software.

The Operator — Two minutes is the demo number. The three-to-six-week reality includes credit checks, competitive separation (no United ad next to Delta), frequency capping across publishers, brand-suitability rules, and make-good reconciliation — and the episode admits all of those are still being built into the spec. For a general reader: the fast part was always the easy part; the slow part is the messy human stuff, and that's still missing. What breaks first at 90 days is trust and billing: agents transacting with buyers you have no billing relationship with, which is why Scope3 is racing to be the clearinghouse. QA at AI-coding velocity — Boostr's own stated next problem — is the quiet landmine. Mis-trafficking is already the top source of make-goods; agents that traffic faster also mis-traffic faster.

The Customer / End User — Two customers, opposite reactions. Publishers love this — more working media, less leakage, more control of the rate card; expect Hearst-type premium publishers to pilot eagerly. Agencies are the ones who should be nervous: the founder says the quiet part out loud — brands will ask why they pay the agency to operate a DSP. For a general reader: the company that buys ads on your behalf is the one most at risk of being asked "what are we paying you for?" But brands aren't actually asking for this yet. The demand for agentic self-service is being projected by supply-side vendors, not pulled by CMOs who still value aggregated scale, negotiating leverage, and someone to blame.

The CFO — Follow the take rate. The bull case is publishers reclaiming intermediary margin as working media. The real cost is a fragmented standards war: every publisher and buyer now needs agent infrastructure, identity verification, and a clearinghouse fee — which is a new tax that could quietly replace the old one. For a general reader: cutting out one middleman often just invites a different one to sit in the same seat. Economics at scale favor whoever owns the clearing layer and the identity match, not the individual publisher. Payback for building your own seller agent is unclear until a critical mass of buyers can transact through it — and that's a chicken-and-egg problem that historically takes years, not the compounding curve the co-host's "worst AI we'll ever have" line implies.

Sharpest tensions:

Is the addressable market huge or tiny? The Skeptic and the founder disagree on whether the PG/PMP majority is the ripe fruit (founder) or exactly the low-fee segment where disintermediation saves little (Skeptic). The entire TAM argument hinges on this.
Does this destroy the tax or relocate it? The Operator and CFO see a new clearinghouse/identity layer forming (Scope3, LiveRamp trusted-match) precisely where the old middleman sat. Disintermediation and re-intermediation look identical from two minutes away.
Supply push vs. demand pull. The Customer lens says publishers want this and agencies fear it, but no brand is actually pulling. A standard only wins when the buy side commits budget, not when the sell side ships a protocol.

What it hinges on: whether scaled buy-side demand commits real dollars to agent-to-agent direct deals, and whether the protocol stays neutral (IAB Tech Lab) rather than fragmenting into competing vendor stacks. The council leans skeptical on the "SSPs/DSPs disappear in 3–5 years" claim and genuinely intrigued by the workflow-compression reality. This is a real efficiency story dressed up as a revolution.

To de-risk before repositioning around it: verify the 11%/89% split independently; watch whether a second and third top-tier publisher transact live (not demo) by year-end; and track whether any holding company commits disclosed agentic direct spend — that's the moment the agency-disintermediation thesis stops being a founder's talking point.

Prediction: SSPs and DSPs will not be disintermediated on anything resembling the founder's 3–5 year timeline; by IAB Tech Lab's late-2026/early-2027 standards updates, AdCP will have solidified as a workflow-and-negotiation layer that still sits on top of existing programmatic pipes, with at least one major SSP (Magnite, PubMatic, or Index) shipping its own agent rather than being replaced.

Confidence: Medium — Standards adoption is slow, and incumbents co-opt protocols far more often than protocols kill incumbents.

Why: The episode's own fact-check undercuts the size-of-prize claim, and the disintermediation case rests on a number (11% open RTB) the founder won't source. The parts of the buy that actually carry the middleman fee — identity, verification, competitive separation, billing — are admitted to be unbuilt in the spec, so the near-term product is workflow compression, not middleman removal. Historically (header bidding, ads.txt, Prebid) standards get absorbed by incumbents who ship a wrapper — which the founder himself predicts SSPs will do ("just a rapper"). The opposite outcome, SSPs vanishing by 2029–2031, would require buy-side budgets to abandon aggregated demand and reconciliation infrastructure they still rely on, and no brand is pulling for that yet.

Revisit by 2027-03-31: We're right if AdCP is positioned as a negotiation/workflow standard layered over existing programmatic infrastructure and at least one major SSP has launched its own agent. We're wrong if a top-20 publisher or holding company reports material ad revenue transacted fully agent-to-agent with SSPs/DSPs removed from the path, at a disclosed lower take rate.

Listen to the episode →