Episode 142: Programmatic Was Just the Beginning with Adam Soroca

Full analysis

Your draft

Adam Soroca (nToggle founder, longtime Magnite product chief) lays out a thesis that AI won't just automate programmatic workflows, it will force the industry to ask why each "lane" between an ad buyer and an ad seller exists at all. His answer: only two moats survive. Proprietary data and owned supply. Everything else (the commodity pipes connecting demand-side platforms and supply-side platforms) gets squeezed out as AI buyer-agents route budgets automatically.

What's actually being weighed here: not a single decision but a structural bet about where value concentrates when AI can discover and transact against supply without a human in the loop. For operators, the live question is: does my business have a defensible reason to exist in an agent-routed market, and how much time do I have before that gets tested? This is a Type 1 (hard-to-reverse) strategic posture, but Soroca himself says the forcing function is slow. Real dollars through agentic pipes are "very little" today.

The Market Analyst: Soroca is describing the intellectual justification for the DSP/SSP convergence that's already underway. When The Trade Desk builds toward the sell side and Amazon/Google run both ends, the "buy-side vs. sell-side software" distinction he says will "blur" is a merger of margin pools, not a friendly handshake. The tell worth watching is buyers taking ownership stakes in supply infrastructure. That's a different animal from supply-path optimization fee pressure. If a holdco puts equity into a pipe, it's declaring the independent middle uninvestable. For the generalist: the toll-booth operators of digital advertising are being told the highway is getting rebuilt around them, and the biggest platforms already own both the on-ramp and the destination.

The Skeptic: The load-bearing assumption is that buyer-agents will "discover and route budgets automatically" well enough to strip out commodity lanes. That's the part furthest from reality. Soroca concedes autonomous negotiation is "a long way" off. "Position and data are the only moats" is also a conveniently timeless slogan; it was true in 2010 and it'll be true in 2035. A thesis that survives any decade isn't predicting a change. It's restating a truism that was old before AI existed. And note the disclosure: Soroca is an investor in the podcast's parent company. The framing is directionally smart but not neutral testimony.

The Operator: Strip the thesis and the practical to-do is unglamorous: get your data pipes and APIs clean enough for a machine to transact against you. That's where most of the room dies quietly. The internal signals here are instructive. A Google PPS (publisher-provided signals, a way to pass audience context) A/B test netted a +1.3% price lift against a -1.44% impression drop: basically flat, run anyway to close a "table-stakes gap." That's the real texture of this transition. The wins aren't transformational; they're the cost of staying eligible. Retail media below Walmart/Amazon (Ace Hardware's ~5,100 stores) only becomes a "lane" once the connectivity plumbing exists, and that plumbing is years of unsexy integration work, not a strategy slide.

The Customer / End User: Two customers, opposite reactions. Large buyers are the ones driving this. The Cannes signal is agencies furious about platforms "steering" budgets and now willing to buy equity to reclaim path transparency. That's a customer voting with a checkbook, and it's real. But the mid-tail retailer and the SMB CTV advertiser aren't asking for agentic routing; they're being told it'll rescue them. The SMB CTV story (AI creative removing the production barrier) is a genuine unlock. The thing that actually stopped a local advertiser from running connected-TV ads was never the media, it was making the ad. That one's underrated in the thesis. For the generalist: the giant advertisers want to see exactly where their money goes; the small ones just want a cheap way to make a TV commercial.

The CFO: Soroca's own history is the cautionary footnote. JumpTap sold for ~$239M in stock, and Millennial Media's shares then cratered. "We sold for $250M" and "we realized $250M" are very different numbers, a lesson for anyone eyeing an exit into a consolidator's paper. On the integration point he's right and it's expensive: Magnite's CTV roll-up (Telaria, SpotX, SpringServe) was hard not because of price but because of choosing which stack dies. Any operator betting on "consolidation of commodity lanes" as their exit should model that the acquirer pays for data and supply position and treats the pipe as a cost to be rationalized. Your team, not your multiple.

Where the council splits:

1. Timing. The Market Analyst sees convergence happening now; the Skeptic and Operator note the agentic mechanism that supposedly forces it is years out. The thesis is right about direction and vague about the clock, and the clock is where money is made or lost.

2. Is "data and position" actionable or just true? The Skeptic calls it a truism. The Operator counters that the actionable version (clean APIs, clean rooms, PPS support) is concrete, expensive, and separates survivors from the merely eligible. The disagreement is whether Soroca gave a strategy or a mood.

3. Who's the real disruptor? Soroca frames AI agents as the force. The Customer lens says the actual near-term pressure is human buyers demanding path ownership: a governance and equity fight, not a technology one. The agents are the story; the angry CMOs are the deadline.

What it hinges on: whether buyer-side agents mature fast enough to route budgets without humans before the current wave of supply-path pressure and DSP/SSP convergence plays out on its own. If agents are slow (likely), the consolidation happens for ordinary competitive reasons and "AI" is the narrative laid over it. Either way, undifferentiated middle-of-the-stack players lose. The mechanism just changes who writes the obituary.

Where the council leans: the structural direction is sound and largely already in motion; the specifically-agentic causation is oversold for the next 12 to 18 months. Operators should treat "make yourself machine-transactable and own some data or supply" as the correct posture, and treat "agents will do this to you by next year" as marketing.

Prediction: Through year-end 2026, spanning the Q3 and Q4 earnings calls of the major independent ad-tech platforms (Trade Desk, Magnite, PubMatic), no more than a low-single-digit share of programmatic spend will flow through fully autonomous "agentic" buying, and management commentary will continue to frame agentic buying as pilot/roadmap rather than revenue.

Confidence: High. Soroca, an insider bull, already concedes agentic dollars are "very little."

Why: The one person most invested in this thesis says actual money through agentic pipes is minimal today and full autonomous negotiation is "a long way" off, and the internal experiment texture (a PPS test netting basically flat results, run only to close a table-stakes gap) shows the industry is still fighting over incremental plumbing, not autonomous routing. Buying infrastructure adoption in ad tech historically lags hype by years. Header bidding and SPO both took multiple budget cycles to move real dollars, and there's no evidence buyer-agents can yet transact reliably at scale. The opposite outcome (a sudden agentic surge in under six months) would require both the technology and agency trust to leap simultaneously, and the Cannes signal shows buyers are still fighting for basic path transparency: the precondition, not the arrival.

Revisit by 2027-02-28: We're right if the Q3/Q4 2026 earnings calls of Trade Desk, Magnite, and PubMatic describe agentic buying as pilots/roadmap with no material disclosed revenue line. We're wrong if any of the three reports agentic/autonomous buying as a meaningful, quantified share of spend on those calls.