Agentic Trading Has Already Started. We Just Don’t Know How Big It Gets.
- Jul 8
- 15 min read
For the past several years, agentic trading has lived comfortably in the future. It was one of those ideas that showed up in conference presentations, innovation labs, and vendor roadmaps. Sure, super interesting in theory, probably important someday, but not something most media leaders needed to worry about right now.
That changed recently. Earlier this month, Digiday reported that Omnicom is actively testing AI agents that negotiate directly with media platforms on behalf of buyers. "We've actually executed real media buys for several clients using our agent framework, doing agent-to-agent buying, which is all in service to shortening the media supply chain," explained Paolo Yuvienco, Head of AI. Then came another milestone: Vox Media and Boostr announced what they described as the industry's first landmark agentic media buy executed using the AdCP protocol. In the span of just a few weeks, agentic trading moved from concept to public implementation. It is no longer a thought experiment. It is no longer a prediction. It's happening.
The significance of these developments extends well beyond Omnicom, Vox Media, or Boostr. For the past fifteen years, digital advertising has been organized around a simple assumption: if you want to buy media at scale, you do it through auctions. Real-time bidding, or RTB, became the default mechanism for buying media because it solved a difficult coordination problem. Millions of advertisers, billions of impressions, thousands of publishers, and decisions that had to be made in milliseconds. Auctions were fast, scalable, and largely neutral, and as such became the beating heart of modern programmatic advertising.
But agentic systems introduce a fundamentally different dynamic. Instead of submitting bids into an auction, buyer agents can evaluate objectives, budgets, constraints, historical performance, and available supply, then negotiate directly with seller agents representing publishers, streaming platforms, retail media networks, and other inventory owners. In this model, media buying starts to resemble continuous automated deal-making rather than split-second bidding. This possibility sits at the center of both AdCP and the broader agentic media movement.
The implications are profound. If buyer agents and seller agents can negotiate directly, what happens to the layers that sit between them today? More precisely, what happens to the economics of DSPs (Demand-side Platforms) and SSPs (Supply-side Platforms)? What happens to a market that has spent the last decade optimizing for auctions when machines become capable of negotiating outcomes instead?
None of this means real-time bidding disappears—nor am I suggesting it. Nor does it mean DSPs suddenly become obsolete. The more interesting question is how much of the market ultimately moves in this direction. Will agentic negotiation account for five percent of media buying? Twenty percent? Fifty percent? Will it remain concentrated among large platforms and premium inventory, or will it spread throughout the broader ecosystem?
We don't know the answer yet, and anyone who says they do is not being truthful. What we do know is that the industry has crossed an important threshold. Agentic trading is no longer theoretical. The technology exists, the incentives are obvious, and some of the largest buyers and publishers in the world are already putting it into production. The debate now is not whether agentic trading arrives, but how much of the advertising industry it ultimately consumes.
RTB Solved a Problem That May No Longer Exist
For most of the history of programmatic advertising, the auction wasn’t just the best solution. It was the only solution. When real-time bidding emerged around 2010, digital advertising faced a massive coordination problem. Millions of advertisers needed access to billions of impressions spread across thousands of publishers. Decisions had to be made in milliseconds, and there was no practical way for buyers and sellers to negotiate directly at that scale.
The auction solved this problem. It was fast, scalable, and—perhaps most importantly—created a common marketplace where buyers could access fragmented supply through a single interface, the Demand Side Platform, or DSP. Over time, this became the foundation of modern programmatic advertising. Today, the vast majority of digital display advertising is purchased programmatically, with real-time bidding accounting for most open-web transactions.
But what if the assumptions that created the auction no longer hold? This question came up recently during a Signal & Noise conversation with ad verification pioneer Dr. Augustine Fu. Fu argued that advertisers should increasingly buy media directly from platforms rather than routing every transaction through a DSP. His concern wasn’t simply efficiency, but economics. Every layer introduced between buyer and seller creates opportunities for fees, markups, arbitrage, and misaligned incentives.
Whether one agrees with Fu’s ultimate conclusion or not, his argument highlights an uncomfortable reality for the industry. Much of what we think of AdTech today—DSPs, SSPs, Ad Exchanges, Brand Safety—emerged because buyers needed a centralized way to access fragmented supply. But agentic systems introduce another possibility. What happens when machines can interact directly with media platforms, streaming services, retail media networks, and publishers through standardized protocols and APIs?
Suddenly, the auction is no longer the only way to coordinate the market. Why can't a buyer agent evaluate objectives, budgets, performance history, frequency requirements, and business constraints?Or why can't a seller agent evaluate inventory availability, pricing, audience composition, and yield objectives? Rather than submitting blind bids into an auction, the two systems can negotiate directly and continuously.
This is fundamentally different from traditional programmatic buying. In the auction model, allocation emerges from millions of individual bids. In an agentic model, by contrast, allocation is determined upfront, with autonomous systems negotiating toward a desired outcome before a single impression is purchased. As we argued previously, decision-making begins moving upstream from execution into allocation itself.
To be clear, this won't make auctions disappear. Auctions remain extraordinarily effective at clearing large amounts of inventory scattered across thousands of publishers quickly. They will likely continue to play a central role across much of the open web for years to come. But for the first time in the history of digital advertising, auctions may no longer be the only scalable coordination mechanism available. And if that’s true, then every participant in the ecosystem has to start asking a different question: Not whether agentic trading can happen, but where it makes more sense than an auction.
Agentic Trading Doesn't Have to Replace RTB to Matter
One of the mistakes people make when thinking about agentic trading is assuming it needs to replace real-time bidding entirely to become important. It doesn't. Programmatic auctions remain remarkably efficient mechanisms for clearing vast amounts of inventory by connecting buyers and sellers in an automated marketplace. If an advertiser wants to purchase millions of impressions spread across thousands of long-tail websites, an auction is still probably the best tool available. It is difficult to imagine autonomous agents negotiating individual transactions for every impression scattered across the open web.
But let's face it, not all media behaves like remnant inventory. Many forms of advertising already rely on negotiation, packaging, and human judgment. Premium homepage takeovers, sponsorships, custom audience segments, connected television inventory, retail media placements, live sports, and high-value publisher relationships often involve discussions about availability, pricing, frequency guarantees, creative requirements, and measurement methodologies. These are precisely the kinds of decisions autonomous agents may eventually handle exceptionally well.
Imagine a buyer agent acting on behalf of a CPG brand. Rather than issuing an RFP or creating a campaign brief and working with the AOR to set up and execute buys across multiple platforms, the buyer agent uses its deep knowledge of the brand's business to flesh the advertiser's objectives, target audience, historical performance, budget constraints, brand safety requirements, and desired outcomes. It then reaches out to seller agents representing publishers, streaming platforms, and retail media networks.
The seller agents respond with proposals. One publisher can offer a premium homepage takeover in several major metropolitan markets. Another bundles contextual inventory around travel and lifestyle content. A streaming platform proposes a guaranteed audience package against households likely to purchase a specific product category. A retailer might suggest an activation using loyalty-derived audiences measured against in-store sales.
The buyer agent evaluates each proposal, asks follow-up questions, negotiates pricing, modifies terms, and assembles a media plan optimized around outcomes rather than inventory silos. No human trafficking team is required, nor are insertion orders are exchanged—at least between people—and no spreadsheets are emailed back and forth. The entire process unfolds continuously and autonomously.
Importantly, auctions do not disappear in this world. Agents may still decide that the best way to acquire certain types of inventory is simply to participate in existing RTB markets. In fact, many DSPs and SSPs may evolve into execution utilities (orchestration layers) that autonomous systems invoke when auctions remain the most efficient clearing mechanism.
This is why the future of agentic trading probably looks less like a replacement for programmatic advertising and more like an additional coordination layer sitting above it. Some impressions will still be bought through auctions, while others will be negotiated directly. Many campaigns will likely combine both approaches.
In other words, the question is not whether agentic systems replace RTB. The question is whether advertisers and publishers prefer negotiation when negotiation becomes effectively free.
Negotiation Becomes Free
For most of human history, negotiation has been expensive. People schedule meetings, buyers issue RFPs, and publishers package inventory. Agencies compare proposals, account teams revise terms, lawyers review insertion orders, and procurement pushes back on pricing. Days or weeks pass before a deal is finalized.
Programmatic advertising arose essentially to solve this problem by replacing negotiation with auctions.
Instead of discussing opportunities with thousands of publishers, advertisers gained access to massive amounts of inventory through DSPs connected to exchanges and SSPs. Decisions could be made in milliseconds and campaigns could scale almost infinitely.
But scale was never truly free. The modern programmatic ecosystem is supported by an extraordinary number of intermediaries. DSPs facilitate bidding, SSPs manage publisher inventory, and exchanges clear transactions. Identity providers enrich audiences, while brand safety vendors evaluate environments, and verification companies measure viewability and fraud. Data providers contribute additional signals, and each participant takes a small percentage of the transaction.
Individually, these fees appear manageable. Collectively, they are enormous. Studies over the years have suggested that more than half of every advertising dollar flowing through the open exchange may never reach a publisher. Some publishers report retaining as little as 38 to 40 percent of advertiser spend after accounting for intermediaries, data costs, and operational overhead, I reported last year. Programmatic may have delivered unprecedented scale, but it also created one of the most elaborate toll systems in modern commerce.
Agentic trading introduces a very different possibility. If buyer agents and seller agents can discover each other, understand objectives, evaluate inventory, negotiate terms, and execute agreements directly, many of today's intermediaries become optional. In this paradigm, there is no exchange collecting a clearing fee, no SSP charging a take rate, and there may be no DSP submitting bids on behalf of advertisers. Furthermore, there may be fewer verification vendors, audience providers, and workflow tools sitting between buyer and seller. Instead, more of the advertiser's dollars theoretically allocated to working media, and publishers receive a larger share of the spend—a rare win-win in advertising. Brands gain access to more inventory for the same budget, and publishers see higher yield.
Agencies benefit as well. Holding companies are, in many respects, financial institutions. They oversee hundreds of millions or billions of dollars on behalf of advertisers and are constantly seeking ways to maximize the efficiency of those funds. If fewer dollars are diverted toward transaction costs and intermediaries, more capital remains available for media investment itself. That creates additional flexibility, stronger client outcomes, and potentially greater economic leverage for the organizations managing those budgets.
This is what people often miss when they hear the phrase free negotiation. It does not simply mean that autonomous agents can negotiate without human involvement—no one is really suggesting that now. Rather, it means the market may no longer need to pay fifty cents on the dollar to coordinate buyers and sellers at scale. If that thesis proves true, agentic trading may not merely change how media is bought, but it could also fundamentally alter where value accrues throughout the advertising ecosystem.
Who Gets Disrupted First?
Whenever a new coordination mechanism emerges, the question isn’t whether value disappears—because value rarely ever disappears—but rather where it moves. The digital advertising ecosystem generates hundreds of billions of dollars annually, but that value is distributed (some would say siphoned off) across a complex web of intermediaries.
DSPs charge platform fees for execution, SSPs monetize access to publisher inventory, exchanges facilitate clearing, and identity providers sell match rates and enrichment. Verification companies assess fraud and viewability, brand safety vendors score content, and workflow vendors automate operations. These many businesses exist because programmatic advertising needs them to function.
Agentic trading raises a more uncomfortable proposition: how many of these functions remain necessary when autonomous systems can coordinate directly? DSPs may be the first to feel pressure.
For years, DSPs have differentiated themselves through bidder technology, supply integrations, optimization algorithms, and workflow tooling. But if buyer agents increasingly make allocation decisions upstream, DSPs risk becoming execution utilities rather than decisioning platforms. They may continue to provide access to auctions and scaled inventory, but they may no longer be the primary interface through which budgets are planned and allocated. This is a huge risk for them.
SSPs face a similar challenge. Today, SSPs aggregate publisher inventory and expose it to buyers through exchanges. But if publishers begin exposing inventory through seller agents capable of negotiating pricing, packaging sponsorships, guaranteeing outcomes, and optimizing yield dynamically, portions of SSP functionality may become redundant.
Identity providers occupy a more nuanced position, and are probably well positioned to weather the coming storm. This is because in an agentic world, identity arguably becomes more important, not less. Buyer agents still need to understand audiences, and seller agents still need rich contextual and behavioral signals. The difference is that identity providers may increasingly function as data utilities rather than transaction toll collectors. Thus their overall take may come down, but their role will still be an important one.
Verification and brand safety companies may also survive, but their roles could (and probably should) evolve substantially. Agents still need trusted signals about fraud, content quality, and suitability, but instead of inserting these services into every impression-level transaction, agents may consume them as reference datasets that inform negotiation and decision-making.
Publishers, meanwhile, may emerge as some of the largest beneficiaries—if they seize the opportunity. For years, publishers have complained that too much advertiser spend is siphoned off away before it reaches their properties. Agentic trading offers the possibility of reclaiming yield by exposing inventory directly to buyers, packaging differentiated products, and negotiating against business outcomes rather than CPMs alone.
Agencies and holding companies may benefit as well. Media agencies are often described as service businesses, but operationally they behave much more like financial institutions when they operate at a HoldCo level. They manage enormous pools of capital on behalf of advertisers and seek to maximize the productive deployment of those funds. If fewer dollars are lost to intermediaries, agencies can potentially direct more capital toward working media, improve campaign performance, and strengthen their economics at the same time. What's not to like?
Of course, disruption rarely happens overnight. And the advertising industry has a long and distinguished history of layering new systems on top of old ones rather than replacing them entirely. As such, agentic trading may follow the same pattern. DSPs may become agent execution layers, while SSPs evolve into publisher operating systems, and verification companies simply expose APIs that autonomous systems consult continuously.
If you ask me, the biggest risk may not be becoming obsolete. Rather, it may be continuing to charge tolls in a market that no longer needs toll roads.
The Future Looks Messier Than Anyone Expected
One thing becomes clear after talking with publishers, agencies, AdTech vendors, and media owners experimenting with agentic systems: almost everyone agrees something is changing, but nobody agrees on how much changes. This uncertainty is understandable, considering that digital advertising has a long history of layering new technologies on top of old ones rather than replacing them outright. Header bidding didn’t eliminate ad exchanges, and connected television hasn't (yet) killed linear television. And retail media didn’t displace search. Instead of wholesale displacement, each innovation expanded the market while reshaping where value accrued.
If you ask me, agentic trading will likely follow a similar path. If you ask around, the most common criticism of agentic trading is that it assumes way too much. For example, buyers want control, publishers care about transparency, and bands require robust governance. I'm not suggesting these assumptions are incorrect, but rather they assume a zero-sum game of one model winning and another losing, when in all actuality this industry rarely operates that way.
Brands still rely on agencies for strategic advice and judgment, which I do not believe will change anytime soon. And let's face it, many campaigns simply won't justify negotiation. A million remnant impressions spread across a long tail of thousands of websites may always be better suited for auctions. These objections are reasonable, but I still think they miss the point. Agentic trading does not need to consume the entire market to become meaningful. It only needs to capture the transactions where negotiation creates more value than bidding.
Consider the following scenarios that might swap to agentic trading first:
Premium homepage takeovers
Sponsorships.
Retail media activations
Live sports inventory
CTV guarantees
Custom audience packages
Cross-publisher deals
High-value contextual opportunities
These are all transactions that already involve conversations, packaging, and commercial discussions. They are also transactions where publishers have historically struggled to expose differentiated inventory through commodity auction mechanisms. Perhaps agentic trading ultimately accounts for only five percent of media spend? Maybe it reaches 20 percent? Or maybe it never moves beyond guaranteed inventory and premium publishers. The fact is, no one knows.
But even five percent of global advertising spend represents tens of billions of dollars—more than enough to reshape incentives, change product roadmaps, and force every intermediary to ask whether their business model depends on friction that machines are increasingly capable of removing.
The irony here is that auctions themselves may survive and even thrive. A buyer agent negotiating a sponsorship package with a publisher may simultaneously instruct a DSP to acquire long-tail inventory through RTB. Seller agents may reserve premium inventory for direct negotiation while continuing to expose remnant impressions through existing exchanges. DSPs may evolve into execution engines. SSPs may become publisher operating systems. Verification providers may expose APIs that agents consult before making decisions.
The future is thus unlikely to be purely auction-based. Yet it is equally unlikely to be purely agentic.
More likely, advertising enters a hybrid era where auctions clear commodities, agents negotiate differentiated opportunities, and humans increasingly focus on strategy, governance, and relationship management. So perhaps we have been asking the wrong question all along. The question isn’t whether agentic trading replaces RTB, but whether whether advertisers and publishers will continue paying fifty cents on the dollar to coordinate transactions once they discover they no longer have to.
Those Who Win May Not Have the Best AI
There is a tendency in our industry to assume that technological shifts are primarily technology problems. They aren’t. They are instead organizational problems. Agentic trading does not require publishers to become AI companies, nor does it require agencies to suddenly hire armies of machine learning engineers. The challenge is something else entirely.
Most organizations are simply not structured to expose what their agents need to know:
For publishers, that means understanding what inventory is genuinely differentiated, what data assets are available, what yield objectives exist, and how commercial constraints should be represented. A seller agent cannot negotiate effectively if the publisher itself does not understand the products it is trying to sell.
For agencies, the challenge is similar. Buyer agents will need access to objectives, audience definitions, budget constraints, historical performance, measurement frameworks, and governance policies. Much of this information lives today in disconnected systems, buried inside spreadsheets, PowerPoint decks, emails, and the minds of experienced traders.
Advertisers face an equally difficult problem. What are the actual business outcomes they care about? What tradeoffs are acceptable? How should an agent prioritize reach versus efficiency? Brand safety versus scale? Guaranteed outcomes versus flexibility?
In many organizations, these questions have never been answered explicitly because humans compensated for ambiguity through conversations and judgment. As we shift to an agentic model, agents force those decisions into the open. This may ultimately be the most significant implication of agentic trading.
In summary, the companies that benefit most may not be those with the most sophisticated AI models.
They may simply be the organizations that know themselves best, are open to change, and are comfortable with experimentation. In other words, a publisher that understands its inventory, an agency that understands and listens to its clients, or an advertiser that understands its objectives.
And perhaps most importantly, the companies willing to experiment and take risks before everyone else.
Because the first generation of agentic trading systems will almost certainly be imperfect, and
the winners won’t be the organizations that wait for standards to mature, protocols to stabilize, and vendors to declare victory. They will be the companies that run pilots, discover where agents created value, and learn how to operate in a world where machines negotiate alongside humans.
Agentic Trading Has Already Started. We Just Don’t Know How Big It Gets.
It is tempting to think about agentic trading as another overhyped technology trend. Advertising has certainly seen its share of buzzwords over the years. Blockchain (LOL) was going to reinvent media buying. The metaverse was going to transform consumer engagement and how we work. Third-party cookie deprecation was supposed to upend the entire ecosystem. Reality, as always, proved more complicated and less black and white.
Agentic trading may prove no different. Auctions are not disappearing, DSPs are not going away, and SSPs will continue to provide important capabilities. Throughout it all, humans will remain responsible for strategy, governance, creativity, and relationships. Most media transactions may still occur through familiar channels 10 years from now.
But that may not matter. Agentic trading does not need to replace real-time bidding to become significant. It only needs to become the preferred mechanism for a relatively small portion of high-value transactions—premium publisher relationships, retail media activations, sponsorships, guaranteed inventory, custom audiences, live events, and outcome-based packages.
Even if agentic systems ultimately capture only five or 10 percent of media spend, that represents tens of billions of dollars moving through a fundamentally different market structure. That is enough to reshape incentives, alter product roadmaps, compress margins, and force nearly every participant in the ecosystem to reconsider where they create value.
And perhaps that is the most important observation of all. The future of advertising may not be determined by who has the best bidder, the fastest exchange, or the most sophisticated optimization algorithm. It may instead belong to organizations that understand their inventory, their audiences, their objectives, and their economics well enough to allow autonomous systems to negotiate on their behalf.
No one knows how large agentic trading becomes. And no one knows whether it settles at five percent of media spend or 50 percent. No one knows for sure whether DSPs evolve into execution engines and dumb pipes, publishers become autonomous marketplaces, or entirely new intermediaries emerge to coordinate the agents themselves.
What we do know is that the industry has crossed an important threshold. The technology exists, the incentives are obvious, and the experiments have begun. Agentic trading is no longer a thought experiment. It is already happening. We just don’t know how big it gets.
--------------------------------------------
Rio is an executive with 20+ years at the intersection of strategy consulting, AdTech, data, and media. He's a trusted advisor on customer experience, digital strategy, and marketing transformation. He's a partner at Credera, Omnicom's consulting arm. He's also a podcast host, writer, and public speaker focused on the future of advertising and AI-driven infrastructure.
Comments