Between November 2025 and March 2026, virtually every major advertising platform launched an AI agent. The term dominated CES 2026, took over IAB ALM, and even prompted AdExchanger to launch a dedicated new conference called Programmatic AI in 2026, alongside its flagship Programmatic I/O. If you only read the press releases, you would think the entire advertising ecosystem has been reinvented overnight.

The reality is more nuanced and more interesting. A lot of what is being called “agentic” today is not. But something genuinely important is happening underneath the hype. And the companies that understand the difference between relabeling automation and building real intelligence will be the ones that define what comes next.

Key Takeaways

• Not all AI agents are truly agentic. Much of what is being called agentic today is not, and understanding the difference between relabeling automation and building real intelligence is critical.
• The industry is entering an “agentwashing” phase. Many vendors are rebranding existing capabilities as agentic, even though fewer than 130 companies today demonstrate real agentic capabilities. 
• A true agent goes beyond interface and automation. A genuine agent can take a campaign brief, reason about context and audience signals, build a strategy, identify what is missing, and prepare it for activation within a single workflow. 
• The real shift is happening at the intelligence layer. The programmatic ecosystem is being restructured around AI agents, but long-term value will come from understanding context, content, and what drives attention, not just infrastructure.

The biggest gap is understanding the moment, not automating tasks. Most AI systems can optimize and execute, but they struggle to explain why something works, especially how content and context shape human attention and response.

The Agentic Gold Rush

To understand the current moment, it helps to look back at the timeline.

In the span of five months, the industry saw a wave of product launches all claiming the agentic label. Major DSPs introduced AI-powered campaign creation tools. Verification companies added agent-based planning features. Commerce platforms launched recommendation services designed to power AI shopping assistants. Microsoft shut down its Xandr DSP, replacing it with a Copilot-powered buying interface, consolidated under the Microsoft Advertising Platform. This was an explicit bet that the traditional DSP model is becoming obsolete.

The pace is striking. But pace alone does not equal progress.

Gartner has already warned the industry about what it calls agentwashing, the rebranding of existing products without substantial agentic capabilities. Their prediction is sobering. Over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear value, or inadequate risk controls. Forrester calls 2026 a year of reckoning, noting that buyers are now seeking proof over promises.

This is the context in which any honest conversation about AI agents in digital advertising needs to take place.

What "agentic" actually means - and what it does not

The word agent is being used to describe at least three different things, and conflating them is making it hard to have an honest conversation about where the industry actually stands.

If you are asking, “What is Agentic AI”, the answer starts here.

The first is genuine agentic AI systems, which can receive an objective, break it into steps, make decisions, use tools, and adapt based on what they find, without a human directing each action. These systems don't just respond; they pursue a goal across a workflow.

The second is LLM-enhanced tooling. These are interfaces powered by large language models that make existing platforms more accessible and easier to use. You describe what you want, and the system translates it into actions. Faster than manual setup, genuinely useful, but the intelligence is mostly in the interface, not in any autonomous reasoning.

The third is rebranded machine learning. Bid optimization, predictive targeting, and budget pacing have existed for years. Today, many of these AI systems are simply being relabeled as agentic because the market rewards them.

A genuine agent can take a campaign brief, reason about context and audience signals, build a strategy, identify what’s missing, and prepare it for activation, all within a single workflow, with minimal human intervention at each step. An LLM wrapper helps you navigate a platform faster. 

Both have value. Treating them as equivalent is how “agentwashing” happens, and why Gartner estimates that of the thousands of vendors claiming agentic capabilities today, fewer than 130 are the real thing.

The Infrastructure Fight that Matters More Than the Product Launches

While the industry debates product features, a quieter but more consequential battle is playing out at the protocol layer.

Two camps are working to define how AI agents should communicate in advertising. On one side, the Ad Context Protocol (AdCP), backed by Yahoo, PubMatic, Scope3, Magnite, and roughly twenty other companies, builds new, purpose-built schemas on top of emerging agent-to-agent standards, with interoperability designed from the start. On the other hand, IAB Tech Lab's Agentic RTB Framework (ARTF) extends existing programmatic standards like OpenRTB and OpenDirect, arguing that a decade of infrastructure should not be discarded. ARTF's containerized architecture is designed to reduce bid latency by up to 80%, not by replacing the stack, but by making it dramatically faster and more agent-friendly.

Both approaches have merit, and they are not mutually exclusive. What they share is a common bet: that the programmatic ecosystem will not be replaced by agentic AI, but restructured around it. The pipes will get smarter. The question is what flows through them.

This is where the economic tension becomes real. Infrastructure that simply routes transactions is under pressure. Intelligence that makes those transactions better, by understanding context, content, and what actually drives attention, becomes more valuable. An agent still needs to know where to buy. The hard problem is knowing why a placement is right and how to solve problems in real time.

At Seedtag, we are building Liz Agent to be protocol-agnostic, because the transport layer will converge regardless of which standard wins. What will not converge is a proprietary understanding of how content environments shape human attention and response. That is the layer that compounds over time. And it is the layer that makes an agent's recommendations worth acting on.

Why Context is Becoming the Foundation, Not the Alternative

Alongside the agentic wave, a deeper structural shift is reshaping what advertising intelligence actually means.

The global contextual advertising market has surged past $225 billion and is projected to reach between $380 and $468 billion by 2030, according to industry estimates from sources such as Statista and Fortune Business Insights. This is not a niche category. It is becoming the foundation of how advertising works. We have seen this firsthand. Seedtag was built without cookies from day one, so the shift everyone else is preparing for is one we have been operating in for over a decade.

The infrastructure that identity-based advertising relies on is disappearing from multiple directions simultaneously. Google killed Privacy Sandbox in October 2025 after the industry invested an estimated $2.3 billion preparing for cookie alternatives that never materialized, according to industry reports. 

Oracle abruptly shut down its entire advertising division, including Grapeshot, one of the largest contextual targeting platforms, redistributing hundreds of millions in annual spending. Third-party cookies survived in Chrome, but roughly 47% of the open internet is already unaddressable by traditional trackers, and Apple continues escalating with iOS 26, stripping platform-specific click identifiers from all browsing sessions.

But the bigger story is that technology has fundamentally changed. Contextual advertising is no longer keyword matching. Modern contextual systems use transformer-based models, computer vision, sentiment analysis, and deep learning to understand content at a semantic, emotional, and intent level. Many of these advancements are powered by generative AI and advanced artificial intelligence systems. The best systems classify content across thousands of categories, identify hundreds of objects and situational contexts in visual content, and process tens of millions of articles daily in real time.

The performance data increasingly support this shift. Research shows contextual targeting delivers significantly lower cost-per-click and cost-per-impression than behavioral approaches, with meaningfully better ad recall and engagement. Nearly 80% of consumers report being more comfortable with contextual ads than behavioral ones, according to multiple industry studies.

CTV is accelerating this further, with the industry moving from genre-level to program-level and even scene-level contextual analysis, a domain where identity-based approaches cover only a fraction of available inventory. It is why we have expanded our capabilities in CTV and partnered with platforms like IRIS.TV for content-level signals, building contextual intelligence natively into streaming, not bolting it on after the fact.

The Gap That Most AI Agents Still Cannot Close

Here is the honest assessment of where the industry stands.

Most AI systems in advertising, even the good ones, are optimized for efficiency. They process data faster, automate setup, and streamline execution. These systems perform tasks efficiently and can automate complex workflows. These are real gains. But they tend to operate on the surface of what makes advertising effective.

Relevance in modern advertising is not just about reaching the right person or placing an ad next to the right keyword. It depends on understanding the conditions of the moment, the cognitive and emotional context in which a message appears. How does the content around an ad shape the way that ad is perceived? What is the reader's mindset? Are they in a mode of exploration, comparison, or decision-making?

These are questions that standard optimization models, even sophisticated ones, are not designed to answer. 

They can tell you what is happening. They struggle with why it matters.

A concrete example: An optimization engine can tell you that a running shoe ad performed 40% better on a wellness article. It cannot tell you that the article was about training for a first marathon at 45, and readers were in a mindset of personal reinvention, which is why the message resonated.

This is where the combination of AI and contextual intelligence becomes genuinely differentiated. 

Not AI applied to the same data everyone else has, but AI applied to a proprietary understanding of how content environments shape human attention and response.

What We are Building with Liz Agent

So, where does Seedtag fit in all of this?

Seedtag was built from inception without cookies or personal identifiers. For over a decade, our engineering team has been developing Liz, a proprietary AI engine that processes over 60 million articles daily across 30,000+ publishers. 

Liz combines NLP, deep learning, computer vision, and sentiment analysis to understand content at a level that goes far beyond keyword matching or standard IAB taxonomy, thousands of contextual categories, hundreds of visual objects and situational contexts, across 10+ languages in real time.

Our Neuro-Contextual methodology takes this further. Working with Professor Moran Cerf at Columbia University, we used EEG measurements to study how context shapes cognitive processing. The results were concrete: 3.5x higher neural engagement versus non-contextual ads. A 30% lift versus standard contextual. A 26% increase in positive, approach-oriented emotional response.

Liz Agent, which we launched in February 2026, is our entry into the agentic AI system space. 

It is built on a multi-agent orchestration engine that combines LLMs with a Retrieval-Augmented Generation (RAG) framework grounded in Seedtag's proprietary data. This allows the system to operate with strong human oversight while still enabling automation where it matters.

Let me explain why the RAG architecture matters. A lot of AI agents in advertising are built on generic LLMs. They are smart, but they do not know anything specific about contextual intelligence, content environments, or how audiences respond to different contexts. They can hallucinate. They generalize. They produce outputs that sound right but are not grounded in real data.

Liz Agent's RAG framework means every response, every insight, every recommendation, every strategic direction is grounded in our proprietary Neuro-Contextual data, not generic knowledge. When Liz Agent analyzes a campaign brief, it draws on a decade of contextual intelligence infrastructure and neuroscience research, not a pretrained model's best guess.

Through a conversational interface, we can move from insights to campaign activation within a single workflow. Liz Agent integrates directly with our proprietary data for real-time contextual and audience intelligence, leverages our models to turn content from our publisher network into real-time embeddings, extracting audience signals and competitive insights that are native to our ecosystem, not scraped from generic sources, and connects strategy directly to activation across our global inventory.

Our intelligence, our proprietary contextual data, our Neuro-Contextual methodology, and our decade of models are what make Liz Agent's outputs actually useful, not just fluent.

The Real Question for 2026

The industry is asking whether AI agents will transform advertising. I think that is the wrong question.

AI agents will compress workflows, make platforms more accessible, and automate tasks that currently consume disproportionate time. That much is certain. 

The better question is: what intelligence are these agents built on?

An agent built on generic data will produce generic outputs. An agent built on a deep, proprietary understanding of how content environments shape human attention and response will produce something qualitatively different. Not just faster answers, but better ones.

Forrester now includes agentic AI as a formal scoring criterion for advertising platforms. Gartner predicts 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% today. Meanwhile, the IAB's 2026 State of Data report found AI-improved measurement alone could unlock $26.3B in media investment.

The opportunity is real. But in a market flooded with agentwashing, the differentiation will not come from who has the best chatbot interface. It will come from those who have the deepest understanding of what actually drives relevance, and the AI architecture to act on it.

In a fragmented, privacy-first digital environment where nearly half of all inventory is already cookieless, relevance is no longer built on identity data. It is built on understanding the moment, the content, the context, and the cognitive and emotional conditions in which a message appears.

The question I would ask any vendor pitching you an AI agent: Show me the data it is grounded in.  If they cannot answer that, you are probably looking at a chatbot with a marketing budget.