The Last 15 Years Were Programmatic. What Comes Next Will Be Agentic.
Somewhere around 2009, the ad tech industry made a collective decision that would define the next decade and a half. The decision was to build pipes. Real-time bidding, supply-side platforms, demand-side platforms, data management platforms, header bidding, identity graphs, clean rooms — all of it was pipe. Infrastructure for moving impressions from publishers to advertisers at machine speed, at scale, with targeting that got progressively more sophisticated over time.
That infrastructure project is now largely complete. The pipes exist. OpenRTB is a mature standard. The major DSPs and SSPs have been built, consolidated, and in some cases acquired by holding companies. The programmatic supply chain — brand to agency to DSP to SSP to publisher — is a known architecture with known participants, known margin structures, and known failure modes. It is not going away. But it is no longer where the defining value in advertising is being created.
The next era of ad tech is not about better pipes. It is about who controls the intelligence layer sitting on top of those pipes — and that layer is being rebuilt from scratch around AI agents.
The shift from hands-on-keyboard to prompting agents is the most significant rearchitecting of ad tech since programmatic displaced direct IO buying.
What Programmatic Actually Built — and What It Left Unfinished
To understand what is changing, it helps to be precise about what programmatic accomplished and what it deliberately did not attempt.
The automation of the transaction
Programmatic's core achievement was automating the transaction layer of advertising. Before real-time bidding, buying a display impression involved insertion orders, rate cards, account managers, and manual trafficking. It was slow, expensive, and opaque. RTB compressed that entire process into a 100-millisecond auction. That is a genuine technological achievement — one that unlocked scale that would have been impossible any other way.
But programmatic automated the transaction. It did not automate the strategy. Human traders still sat behind DSP keyboards setting bid strategies, defining audience segments, allocating budgets across channels, and interpreting performance reports. The machine moved fast. The human still decided where to aim.
The intelligence gap it left open
Every major programmatic platform — The Trade Desk, DV360, Amazon DSP — built AI and machine learning into their bid optimization over time. Smart bidding, predictive audiences, automated budget pacing. These were genuine advances. But they operated within a human-defined frame: the human set the campaign objective, the human defined the audience, the human approved the creative, the human evaluated the results. The AI optimized within those parameters. It did not set them.
The intelligence gap that programmatic left open is the strategic layer: the planning, the decision-making, the interpretation, the adaptation. That gap is exactly what agentic AI is now filling.
The Shift: From Hands-on-Keyboard to Prompting Agents
The operational model of programmatic advertising for the past 15 years has been the trading desk: a team of humans with keyboards, dashboards, and spreadsheets, executing media buys across DSP interfaces. The mental model is one of control — the human trader as the primary intelligence in the system, using software to execute decisions faster than manual buying would allow.
That mental model is being replaced.
What the new model looks like
In the emerging agentic advertising model, a media buyer does not log into a DSP interface and pull levers. They prompt an agent. They describe the objective — reach in-market SUV buyers in Q2 with a $2M budget, optimize for cost-per-test-drive-scheduled — and the agent plans, executes, monitors, and optimizes the campaign. The human reviews outputs, provides direction, and makes judgment calls on strategy. The agent handles execution.
This is not science fiction. Viant's Outcomes product, launched in January 2026, already operates this way for programmatic campaigns — advertisers define a business outcome, and Lattice Brain manages execution autonomously. The Trade Desk's Kokai platform positions itself as a "copilot" for human traders. FreeWheel and NBCUniversal executed the first agent-to-agent media buy in January 2026, with AI agents on the buy side negotiating directly with AI agents on the sell side using Model Context Protocol — no human in the transaction loop.
The direction is clear. The question is not whether the shift from hands-on-keyboard to prompting agents will happen. It is how fast, and who builds the infrastructure for the world on the other side.
Why this is structurally different from previous automation waves
Ad tech has seen multiple automation waves: programmatic replaced direct IO, real-time bidding replaced waterfall, header bidding replaced the waterfall again. Each wave automated a previously manual process while leaving the strategic layer intact. Humans adapted their roles — from insertion order buyers to programmatic traders — but the fundamental model of human-as-primary-intelligence remained.
The agentic shift is different because it automates the strategic layer itself. When an AI agent can take a campaign brief, develop a media plan, execute buys across channels, optimize in real time against outcome metrics, and report results — the role of the human changes not just in execution but in kind. The human becomes the agent's director, not its executor. That is a more fundamental shift than any previous automation wave in advertising.
The Infrastructure Problem: Why Programmatic Rails Are Not Enough
Here is the critical point that most of the ad tech industry has not yet fully reckoned with: the programmatic infrastructure that exists today was built for human operators, not AI agents.
DSP interfaces are built for humans
Every major DSP — The Trade Desk, DV360, Amazon DSP, Viant — was designed around a human user interface. Campaign setup flows, audience builder tools, creative management dashboards, reporting visualizations. These interfaces assume a human is reading them, interpreting them, and making decisions based on what they see. An AI agent does not read a dashboard. It needs an API, a structured data feed, and a protocol for sending and receiving instructions.
The industry is adapting. Most major platforms now offer APIs that agents can call. But the API was designed to replicate what a human does in the UI — it was not designed for agent-to-agent communication from the ground up. The difference matters architecturally and will matter commercially as agentic buying becomes the norm rather than the exception.
The signal mismatch
Programmatic targeting was built around two signal types: behavioral (what the user has done) and contextual (what page the user is on). Both are proxies for intent — imperfect approximations of what a person might want, inferred from indirect evidence. They are the best signals the programmatic architecture could access, because programmatic was built for web pages, where direct intent expression is rare.
Agentic AI surfaces change this entirely. When a user tells an AI assistant exactly what they are looking for — the product category, the price range, the feature priorities, the timeline — they are expressing intent directly in natural language. That signal is not a proxy. It is the thing itself. And it requires a fundamentally different infrastructure to capture, classify, and activate — one that reads conversational context, classifies intent semantically, extracts entities, and matches to demand in real time. The programmatic stack was not built for this. Adgentek's Agentic Ad Server was.
The format mismatch
Programmatic delivers rectangles. Banner ads, interstitials, native tiles, pre-roll video — every format in the programmatic creative taxonomy was designed for a visual, scroll-based environment where an ad occupies a fixed slot in a page layout. None of these formats belong inside an AI agent interaction.
When a user is mid-conversation with an AI assistant — asking about family vehicles, comparing mortgage options, researching travel destinations — a banner ad is not just irrelevant. It is architecturally wrong. The conversation has no fixed slots. The interface has no rectangles. The user is engaged in a dialogue, not scanning a page. The ad format must be native to that dialogue — interactive, conversational, brand-authored — or it will be ignored, rejected by publishers, or actively harmful to the user experience that makes the AI surface valuable in the first place.
Where the Value Accrues in the Agentic Era
Every major shift in ad tech has produced a redistribution of value. The programmatic shift moved value from direct sales teams to trading desks and DSPs. The mobile shift moved value to mobile-first DSPs and attribution vendors. The streaming shift is moving value toward CTV-native platforms and identity resolution companies.
The agentic shift will be no different — and the value redistribution it produces will be larger than any previous shift, because it affects the entire stack simultaneously.
The intelligence layer
The companies that own the intelligence layer — the systems that understand what an AI agent or a user inside an AI surface is trying to accomplish, and can match that intent to relevant brand demand — will capture the margin that currently goes to human trading operations and the platforms that power them. This is not a small margin. Trading desk fees, DSP take rates, and managed service premiums represent a substantial portion of the dollars flowing through the programmatic ecosystem today.
Adgentek's Agentic Ad Server owns this intelligence layer for conversational and agentic AI surfaces — the intent classification engine, the entity extraction system, the semantic context matching, the nine-bucket intent taxonomy that determines which demand tier fires, which format renders, and what outcome is measured.
The protocol layer
OpenRTB created value by standardizing the language that buyers and sellers use to transact. Before OpenRTB, every exchange had proprietary APIs and data formats. OpenRTB made the ecosystem interoperable and unlocked the scale that programmatic achieved. The companies that shaped OpenRTB's early versions had disproportionate influence over how the programmatic ecosystem developed.
The Ad Commerce Protocol (AdCP), stewarded by Adgentek via AgenticAdvertising.org, is the OpenRTB of agentic advertising. It defines how AI buying agents and AI selling agents communicate — the message structures, governance sequences, signal schemas, and transaction protocols that enable agent-to-agent advertising at scale. AdCP is currently on version 3.0. The parallel to IAB's OpenRTB in 2010 is direct: the protocol layer is being written now, by the companies participating in writing it.
The format layer
Every major channel shift has produced a new dominant ad format. Search produced the text ad. Social produced the sponsored post. CTV produced the non-skippable pre-roll. Each format became a multi-billion dollar revenue category, and the companies that owned the format owned a toll booth on the channel.
Agentic advertising's dominant format is not yet determined — but the direction is clear. Static sponsored links, the format OpenAI launched with ChatGPT ads in February 2026, are the banner ad of the agentic era: a legacy format adapted to a new context, capturing none of the channel's unique value. The formats that will define agentic advertising are interactive, conversational, and intent-generating. Adgentek's Spark format — interactive Q&A units where users self-qualify through branded conversations — is the first purpose-built agentic ad format at scale. Early data shows 50% higher ROAS versus static sponsored links and 3.2 average interactions per session, generating intent signals no impression-based format can replicate.
What This Means for Everyone in the Ecosystem
For brands and agencies
The planning and buying model your teams have used for the past 15 years — campaign briefs, audience definitions, insertion orders, trafficking, reporting — is going to look fundamentally different in five years. The agencies and brand teams that are experimenting with agentic buying and agentic ad formats today are building the institutional knowledge and the vendor relationships that will matter when the shift accelerates. The ones waiting for the market to mature before engaging will find the competitive disadvantage already baked in.
The immediate practical question for brand advertisers is where to place early bets. Conversational AI surfaces — the AI assistants and chatbots where hundreds of millions of people are already spending meaningful time — are the accessible entry point. The audiences are real, the intent signals are unprecedented, and the early-mover CPMs are substantially below what those audiences will command once the category reaches scale.
For publishers and AI surface operators
The programmatic ecosystem took years to develop meaningful publisher revenue because the demand side had to be built alongside the supply side. The agentic advertising ecosystem is developing faster because programmatic demand already exists — it just needs new integration paths into new surfaces. AI app publishers who integrate agentic ad infrastructure now are connecting to real brand budgets, not waiting for a demand pool to materialize.
The publishers who will define the agentic advertising supply side are the ones making integration decisions today. The AI apps and assistants that reach meaningful scale in the next 18 months — and connect that scale to premium brand demand through native, conversation-aware ad infrastructure — will be the premium publishers of the agentic era, the same way top-tier web publishers commanded premium programmatic CPMs in the display era.
For the existing ad tech stack
The companies most exposed to the agentic shift are those whose value proposition is primarily intermediation — moving money and data between buyers and sellers without adding intelligence, format, or context. The programmatic supply chain has always had more layers than it needed, and agentic agent-to-agent buying removes the architectural requirement for many of them. FreeWheel's January 2026 proof-of-concept demonstrated that an AI buying agent can transact directly with an AI selling agent, compressing a multi-step human-mediated workflow into a direct machine-to-machine transaction. The AdExchanger analysis of that proof-of-concept noted explicitly that "an agentic approach theoretically removes the need for DSPs and SSPs."
That does not mean DSPs and SSPs disappear. It means the value they provide must come from something other than occupying a position in the transaction chain. The companies in the existing stack that will survive the agentic shift are those that own genuine intelligence — audience data, contextual understanding, measurement infrastructure, creative optimization — not those that own a toll position.
The Parallel to 2009
In 2009, a small number of companies understood that real-time bidding would fundamentally restructure the economics of display advertising. Most of the industry was skeptical — the existing direct sales model was profitable, relationships were entrenched, and the new infrastructure was unproven. The companies that moved early — building the exchanges, the DSPs, the data platforms — defined the programmatic era and captured the majority of its value. The companies that waited for certainty found the positions already occupied.
The agentic advertising shift is at a similar inflection point. The infrastructure is being built. The first real transactions have occurred. The major platforms are making moves — some defensive, some aggressive. The category is real, the timing is clear, and the early-mover advantage is substantial.
The last 15 years were about building the pipes. The next 15 are about who controls the intelligence, the formats, and the protocols that determine how those pipes are used — and by whom. That work is happening right now, by the companies willing to build before the outcome is certain.
For a deeper look at what agentic advertising is and how it works, or to explore what it means specifically for brand advertisers and AI surface publishers, those resources cover the specifics in detail.
