AdCP and the Agentic Reckoning: RIP RTB?

Why Advertising May Be Heading Toward a Negotiated, Agent-to-Agent Market, and what does this mean for Real Time Bidding (RTB)?

For the past fifteen years, AdTech has optimized relentlessly for one thing: speed. Emerging around 2010, Real-time bidding, or RTB, didn’t just enable digital advertising—it came to define it. As of 2025, close to 90 percent of all digital display ads are purchased programmatically, and within programmatic, RTB constitutes approximately 60-65 percent of all transactions.

RTB and its underlying auction became the norm for programmatic media because it elegantly solved a challenging coordination problem. When billions of impressions, millions of advertisers, and deeply fragmented supply paths all had to connect in milliseconds, there was no time for discussion. The auction is fast, neutral, and scalable, clears the market instantly, impression by impression, enabling a vast ecosystem of supporting players. 

Over time, that auction logic moved from being a technical necessity to an economic assumption. Advertising became something you bid on, not something you negotiated. Value was thus discovered moment by moment, detached from longer-term relationships or outcomes. Speed and efficiency were prioritized over everything else. RTB didn’t just power the market, one could argue it shaped how the industry thought about media itself.

RTB Explained, Courtesy of AppsFlyer

This framing holds if we believe the auction is the most efficient way to transact media, and systems lack memory, reasoning, or agency. Auctions make sense in a world of stateless decisioning. "Unified auction will eventually power 95% of all advertisements and transactions. Our future depends on it,’ said Jeff Green, CEO of The Trade Desk.

But this universally held premise is starting to show cracks.

As agentic AI enters the advertising stack—and as protocols like AdCP begin to standardize how autonomous systems communicate—the industry is rapidly laying the groundwork for a different kind of market. An agentic future points to a world where machines are no longer limited to submitting blind bids, but can reason, remember, and negotiate on behalf of buyers and sellers over time.

In an agentic environment, auctions stop being the default mechanism and become just one option among many. If agents can evaluate objectives, constraints, supply characteristics, and historical performance, arriving autonomously at agreements, then we’re looking at a future that look less like split-second bids and more like continuous, automated deal-making.

If this future materializes, the implications are uncomfortable. The center of gravity in advertising may shift toward dynamically negotiated, agent-executed direct deals—PMPs, programmatic guaranteed, preferred access, curated supply—while the open exchange increasingly serves as the place where unmatched or excess inventory goes to clear.

Hyperbole aside, this isn’t an argument that RTB disappears. It will certainly continue to play an important role in programmatic. It’s an argument that the auction—long the beating heart of online marketing—may no longer sit at the center of the system, replaced by a market that looks far more negotiated than bid.

What AdCP Actually Is (and What It Is Not)

To understand why this shift is plausible, we need to dig into what AdCP is, and just as importantly, what it is not. AdCP was introduced in 2025 and initially championed by Brian O’Kelley in collaboration with the IAB Tech Lab and a small group of senior AdTech operatives. AdCP is an open-source, industry-standardized protocol, not a standalone product or a platform, and is best understood as a common language for advertising agents to express intent, constraints, capabilities, and outcomes. It is infrastructure, not an application. 

That distinction matters because AdCP is not trying to make RTB smarter or work better. Rather, it is trying to make agents legible to one another. Instead of forcing all value discovery through impression-level auctions, AdCP ostensibly enables buy-side and sell-side agents to describe what they are trying to achieve, what they are willing to offer, and under what conditions a transaction makes sense—before a bid is ever placed.

In that sense, it is less about sheer execution speed and more about coordination. “RTB elegantly solves the matching problem for non-guaranteed inventory, said BoK himself recently on Twitter/X. “AdCP is trying to solve a different problem which is how to match supply and demand across the entire market.”

U of Digital's irreverent take on AdCP 🤘

In other words, this is where AdCP diverges sharply from other recent “agentic” efforts. Protocols like Google’s UCP are fundamentally commerce-oriented in that they focus on product discovery, pricing, and transaction fulfillment, with advertising barely considered.

Meanwhile, emerging frameworks like RTBF attempt to make auctions more agent-aware through a containerized approach—creating a secure environment so agents can collaborate on services critical to RTB like identity resolution pre-impression fraud detection. 

AdCP is disruptive because it questions the auction’s centrality altogether. Never shy to offer an opinion, Luma CEO Terry Kawaja has framed this moment as a structural inflection point in which the market evolves and coordination costs collapse. This is important because since the rise of RTB, auctions have minimized coordination by stripping transactions down to price and speed. But once agents can reason, remember, and negotiate, the coordination burden shifts again. The market no longer need to discover value purely through instantaneous price competition. In an agentic future, it can allocate value through coordination and negotiated agreement.

Hype aside, the current AdTech stack was never designed for this. It assumes humans sit at the center of every decision loop. People define strategies, configure campaigns, set guardrails, and interpret results. Sure, much of the process is supported by automation, and AI already assists in some tasks, but it does not act autonomously. Even today’s most advanced “AI-powered” tools remain human-authored systems operating inside human-defined workflows, I would argue.

AdCP challenges that assumption directly. It assumes the primary actors in the system are no longer people using tools, but agents interacting with other agents—continuously, autonomously, and across time. Once this assumption holds, auctions stop being the default organizing principle of advertising. It becomes one mechanism among many in a broader, negotiated marketplace. This shift that makes everything else in this conversation possible.

From RTB to Agentic—What Changes?

Turning back the clock a couple decades ago, all media—including digital—was negotiated. Negotiation was by definition slow, mostly because it involved humans and human interactions. Direct deals depended on trust, context, and coordination between multiple parties. Even once the terms were negotiated, IOs needed to be drafted and signed, pixels placed and tested, etc. It was a cumbersome process ripe for transformation. This is why RTB was invented.

In other words, RTB didn’t take over because the industry suddenly fell in love with auctions. Well, maybe some people did, but most people couldn’t care less and RTB took over because it offered an irresistible combination of automation, speed and scale. Auctions eliminated friction by design—heck, the term programmatic means “automated.” They compressed complexity into price, cleared the market instantly, and created liquidity across an otherwise unmanageable, fragmented ecosystem. But that efficiency came at a cost.

We can all agree that RTB is better than this. Pic courtesy of Yahoo.

Along with automation and efficiency, RTB stripped away memory and intent. Every impression became a standalone event. The system had no awareness of long-term objectives, no ability to reason across portfolios, and no concept of durable buyer–seller relationships. Optimization happened statistically—through signals and averages—not strategically. This tradeoff was rational when machines couldn’t do more than execute rules. 

Now with the emergence of agentic buying, this paradigm becomes harder to justify with machines that can reason, remember, and even act with intent. Definitionally, agentic systems introduce capabilities that auctions fundamentally lack. An agent can remember past performance across publishers, formats, and deal types. It can also reason about tradeoffs between price, reach, quality, or outcomes, and it can act with a defined objective that extends beyond a single impression.

Once these three capabilities exist, bidding blindly into an open auction starts to look less appealing. Instead of asking, “What is this impression worth right now?” an agent can ask, “Which combination of supply, context, timing, and price best satisfies my objective over time?” This is not an auction question—it’s a negotiation question.

The Agentic Deal Hypothesis

This leads to my hypothesis: In an agentified advertising market, more demand will flow into dynamically negotiated, direct-sold deals, executed autonomously by software agents, while the open exchange increasingly absorbs what remains, which will be mostly remnant inventory. In this manner, the agentic future will more resemble the TV landscape in which direct-sold deals hammered out during upfronts account for approximately 90% inventory sold, and what is left is purchased using the scatter market in which or remnant inventory is sold closer to program air date, usually for a higher price.

This doesn’t mean a return to the old days of manual IOs or annual upfronts à la broadcast TV. That ship has sailed and is not coming back to port. It will, however, mean loads of PMPs, programmatic guaranteed, preferred access, and curated deals, brokered as continuously optimized constructs rather than static contracts.

Agents will be able to negotiate price floors up or down based on performance, trade guarantees for flexibility, and prioritize certain environments based on brand fit or incrementality. And all of this should occur continuously, without much (if any) human intervention. The result is a market that looks less like a cacophony of isolated auctions and more like ongoing deal-making at machine speed.

What an Agentic Marketplace Looks Like in Practice

In an agentic marketplace, buyers deploy software agents with clear, explicit mandates. Those mandates may include reach, efficiency, incrementality, brand safety, sustainability, or some combination of all the above. On the other side of the market, sellers deploy agents that represent available inventory alongside yield requirements, customer experience constraints, and advertiser preferences. Each agent acts as a persistent proxy for its owner’s objectives.

Crucially, these agents do not simply submit bids into an exchange. They communicate, exchanging structured signals about availability, performance, pricing logic, and constraints, and they reason over that information across time. Access and pricing are negotiated dynamically, not preconfigured in static deal IDs. Agreements are created, adjusted, and dissolved continuously as conditions, performance, and objectives change.

This is what DALL-E thinks an "agentic marketplace" would look like

In this model, human teams do not disappear, though their role moves decisively upstream. People define strategy, set guardrails, establish governance, and remain accountable for outcomes. Day-to-day execution—pacing, optimization, deal construction, and adjustment—is delegated to machines that can operate continuously and at scale.

RTB still plays a role in this environment, but it is no longer the organizing principle of the market. Instead, it increasingly functions as a fallback layer: the mechanism through which supply that does not meet negotiated criteria clears efficiently, rather than the primary venue where intent-rich demand and premium supply first meet.

Why This Puts Structural Pressure on the Open Exchange

The open exchange was built for a world defined by volume. RTB marketplaces optimize for liquidity, speed, and continuous price discovery, assuming a steady flood of impressions and bids. Auctions work best when supply is abundant, participation broad, and competition constant. The arrival of an agentic market challenges those assumptions.

If a growing share of intent-rich demand flows upstream—through negotiated, agent-executed deals—then fewer transactions flow into the open exchange in the first place. But critically, it is not just demand that declines. Supply does too. Premium inventory, preferred contexts, and high-performing placements are increasingly reserved for agent-negotiated arrangements, leaving a smaller and more uneven pool available for open auction. The result is fewer auctions competing over less inventory.

When fewer auctions compete over less inventory, a marketplace doesn’t just “slow down.” Source.

That dynamic would have profound consequences. With lower transaction volume and less liquidity, auctions become more volatile, price discovery weakens, and CPMs may rise on certain scarce impressions as supply tightens, while collapsing on others as demand fragments. The open exchange thus becomes less of a stable market and more of a clearing mechanism—efficient, but increasingly exposed to imbalance and volatility

None of this suggests the auction breaks. To the contrary, it still performs exactly as designed. But as agents handle more coordination directly, the auction’s role shifts from primary allocator of value to secondary outlet for what remains. In this world, the open exchange does not disappear—but it becomes a smaller, noisier, and more economically fragile layer beneath a negotiated, agent-driven market.

Why This Time Really Is Different

The industry has heard many versions of this story before. Programmatic was supposed to eliminate direct sales. Header bidding was supposed to flatten differentiation. AI was supposed to automate media buying end to end. Each wave promised structural change, but ultimately stopped short.

What makes this moment different is this is not simply a change in tooling. It’s where decision-making authority is moving. This is not about giving humans better dashboards or faster optimization loops. The agentic reckoning is about delegating economic decisions—where to buy, whom to buy from, under what conditions—to autonomous systems operating continuously and at scale. This is a fundamentally different shift.

As Brian O'Kelley himself has argued, agentic advertising is not primarily about making bids more efficient. It is about allocation: deciding where value should flow before an auction ever takes place. Once that allocation happens upstream, the auction stops being the organizing principle of the market.

The Coming Reckoning

AdCP is a provocation, not a prediction. It forces the industry to confront a question it has largely avoided for the past decade: what happens to an advertising market built around auctions once machines no longer need them as the primary coordination mechanism? If agents can reason, remember, negotiate, and allocate value upstream, then auctions stop being destiny. They become a design choice.

This is the reckoning. For years, the industry treated RTB not just as infrastructure, but as inevitability. The auction became synonymous with efficiency, objectivity, and scale. Entire business models, take rates, and power structures were built around that assumption. AdCP quietly challenges it by suggesting that speed alone is no longer the highest-order constraint.

The reckoning cometh. Pic via Surfer Today.

If this is true, then some uncomfortable outcomes follow. Value will concentrate differently, intermediaries whose relevance depends on volume and opacity will be pressured, and direct relationships—once written off as unscalable—may return in machine-executed form. And the open exchange, while still necessary, may no longer be where the most important decisions are made.

This does not mean the auction disappears. It means it loses its monopoly on meaning. In an agentic future, the most consequential decisions in advertising will happen before a bid is ever placed. Allocation will precede execution. Negotiation will precede price discovery. And strategy will increasingly be expressed not in media plans, but in the objectives and constraints we give to machines.

The industry can pretend this is just another standards discussion. Or it can recognize what AdCP really represents: a shift away from advertising as a series of isolated transactions, and toward advertising as a continuously negotiated market between autonomous actors. Once this shift begins in earnest, there is no going back to a world where bidding is the center of gravity. The question is no longer whether the auction survives. It’s whether the rest of the ecosystem is ready for what comes after.