The Agentic Future of Advertising: Built on First Principles, Accelerated by AI

The advertising industry stands at an inflection point. Large language models, transformer architectures, and breakthroughs in GPU compute are fundamentally changing how we connect advertisers with audiences through autonomous or semi-autonomous agents that facilitate media discovery, planning, buying, and other functions.

I discussed this evolution in my opening keynote at the IAB Tech Lab Summit last June and have been thinking about it ever since. It’s exciting. But amid the excitement, we risk losing sight of what matters most: the constituents this ecosystem exists to serve.

Before we augment our infrastructure for an AI-native future, we need to ask the right questions. Not “how do we make advertising more automated?” but rather: “What do the people in this ecosystem actually need, and how can AI help us deliver it better?”

From New York to California to London and Berlin, and from Sydney to Tokyo, I’ve spent the better part of the last four months asking this question of the key stakeholders in our industry.

The Four Stakeholders, The Four Truths

Every ad tech innovation should be measured against how well it serves four fundamental needs:

Advertisers want to grow their business. They want to acquire new customers, deepen relationships with existing ones, sell more products, and attract more profitable customers. That’s why they advertise. Not because they love impressions or click-through rates, but because advertising is a tool for business growth.
Agencies want to help advertisers achieve that growth. They bring expertise in audience understanding, media strategy, and creative execution. Their value lies in making complexity manageable and turning marketing budgets into business outcomes.
Publishers want to fund the content their users love. Whether it’s streaming entertainment, quality journalism, mobile games, or productivity apps—publishers create value for users, and advertising helps make that content accessible. The best ads don’t interrupt this value exchange; they enhance it by connecting users with relevant products and services.
Users and Shoppers want help finding what they’re already looking for. They want to compare options in categories they’re considering. And occasionally, they want to be surprised and delighted by discovering something they didn’t know they needed. Good advertising serves these purposes. Bad advertising is just noise.

These aren’t abstract principles. They’re the foundation on which everything else should be built.

The Real Opportunity AI Presents

With this foundation clear, the question becomes: how can AI and agentic systems help us better serve these needs?

The answer isn’t “automate everything and remove humans from the loop.” The answer is: lubricate the friction points that prevent value from flowing between these stakeholders.

Consider what happens today when a media planner wants to run a CTV campaign. They have deep knowledge of their brand’s customers, built over years of direct mail, loyalty programs, and first-party data. They’ve segmented this database carefully. They’ve crafted a thoughtful brief about how they want to reach new customers.

Now they need to match this intent with publisher inventory. They’ll use DSPs to access demographic data. They’ll select shows, apps, and channels they believe will reach their target audience. They’ll run the campaign, analyze tabular dashboards, re-segment based on what’s working, and iterate.

Every step involves manual effort, data translation, and information loss. The brief in the planner’s head doesn’t fully translate into the targeting parameters available in the DSP. The rich context about a publisher’s content doesn’t fully surface in standardized inventory feeds. The feedback loop from campaign performance to planning insights is slow and lossy.

This is where AI creates real value.

An agent with a deep understanding of publisher inventory, audience taxonomies, and content context can help match advertiser intent with opportunity far more precisely. Natural language interfaces can capture nuance that drop-down menus can’t. LLMs can surface connections between a brand’s customer segments and a publisher’s audience composition that would take humans weeks to discover.

But here’s the critical insight: this value creation depends entirely on precise, deterministic standards beneath it.

Why Existing Standards Matter More in an Agentic World, Not Less

When humans are in the loop at every step, ambiguity can be resolved through judgment, conversation, and intuition. When agents are orchestrating complex workflows across multiple systems, ambiguity becomes catastrophic.

Imagine an agent negotiating a programmatic deal across three different SSPs. If each SSP represents “video inventory” differently—with subtly incompatible definitions of playback methods, viewability measurement, or content categories—the agent will either fail silently or produce results no one intended.

We’ve already seen what happens when AI systems operate without deterministic grounding: they hallucinate. They confuse concepts. They generate confident-sounding nonsense. In advertising, this means fuzzifying audiences, misrepresenting placements, misclassifying content, and creating openings for fraud at scale. Shared definitions, transparent interfaces, and enforceable governance allow trust and accountability. And trust needs to be embedded in the agentic systems.

The solution isn’t to slow down AI adoption. The solution is to ensure that agentic systems are built on object models and taxonomies that provide semantic precision. When an agent says “video impression with autoplay sound-off on a news site reaching adults 25-54 interested in cooking,” every term in that phrase needs to resolve to a specific, industry-agreed definition.

Building on What Works

This is why IAB Tech Lab’s approach to the agentic future starts with existing industry standards. It’s not because we’re resistant to change, but because these standards represent compressed industry knowledge refined through billions of transactions.

AdCOM provides canonical domain objects: What is a placement? What is a video impression? What are the attributes of a device or user?
OpenRTB handles real-time bidding at massive scale with battle-tested semantics. It overlaps with AdCOM, supporting “rails” into OpenRTB.
OpenDirect manages programmatic guaranteed workflows for direct media buying.
The Ad Management API standardizes creative submission and approval workflows between buyers and sellers.
The Deals API standardizes the synchronization of dealID metadata.

Tying it all together are the standardized Tech Lab taxonomies, including our Audience, Content, and Privacy taxonomies that provide a common language to describe what audience is being targeted, against what context (your targeting parameters), while ensuring privacy compliance as agents trade and match consumer data.

Critically, all of these share largely the same underlying object models. A video impression means the same thing whether you’re executing a real-time bid or setting up a programmatic guaranteed deal. This semantic consistency is precisely what agents need to operate reliably across myriad workflows.

This is the “language” of our industry.

The innovation layer, where discovery, negotiation, and orchestration occur, can be built on protocols such as A2A (Agent2Agent), which provides agentic coordination of primitives, or MCP (Model Context Protocol), which provides a standardized interface for AI to synchronously access tools and data sources within a single session or application context. The term application context is critical here, as these coordination layers should reference the widely adopted, agreed-upon schemas and domain objects the industry has already defined and is actively using across thousands of advertising applications today, rather than reinventing them.

Think of it like the financial markets. High-frequency trading didn’t require redefining what a stock is or what a trade means. It required a faster execution infrastructure built on top of established definitions. The most sophisticated algorithmic traders in the world rely on standard object models. They just operate on them faster and with more intelligence. And deep learning in high-frequency trading isn’t just for day traders; the largest institutional investors in the financial services market leverage oceans of transaction data and massive-scale AI computing to better understand their long-term investments and make long-term investment decisions.

Don’t Take Our Word For It

The Tech Lab isn’t advocating the use of existing battle-tested schemas, object models, and taxonomies because we want to be the crufty dad in the room, wagging our finger at the industry and saying, “Do it this way!” Both MCP and A2A are fundamentally schema-driven. Just ask Claude:

“Are MCP and A2A protocols fundamentally schema-driven?”

“Without IAB Tech Lab’s OpenDirect, AdCOM, OpenRTB, related schemas and taxonomies, agent-to-agent programmatic negotiation would be impossible. The schemas are the shared protocol that enables automation. This is why a focus on Tech Lab standards and agent protocols is so powerful. Building on established schemas rather than inventing new ones means instant interoperability with existing ad tech infrastructure.”

What We’re Building

IAB Tech Lab has developed an Agentic Roadmap grounded in these first principles and our existing standards. Our approach is phased, starting with foundational capabilities and expanding as the industry builds confidence in agentic workflows.

The near-term focus is on the highest-value, lowest-risk applications: helping agencies and advertisers discover publisher inventory more efficiently, matching briefs to opportunities through natural language, and accelerating the setup of direct deals that currently require manual coordination across multiple systems.

As these patterns prove out, we’ll expand to more semi-autonomous workflows, always with appropriate human oversight, supported by deterministic standards that provide the semantic foundation to prevent hallucination and fraud.

We’re building sample agents and MCP server implementations not as products, but as reference architectures that demonstrate how agentic systems should interact with existing ad tech infrastructure. The goal is to help the industry build interoperable, standards-compliant agents that work together rather than further fragment the ecosystem.

The Path Forward

The agentic future of advertising is genuinely exciting. AI can help advertisers find the right audiences more precisely. It can help agencies deliver better outcomes more efficiently. It can help publishers monetize their content more effectively. It can help users see ads that actually serve their needs rather than interrupting their experience.

But realizing this potential requires building on solid foundations. The industry has spent fifteen years developing semantic precision around what advertising objects mean. That precision is exactly what AI systems need to operate reliably.

The opportunity isn’t to rebuild advertising infrastructure from scratch. It’s to accelerate and lubricate the value exchange between advertisers, agencies, publishers, and users, using the language they already speak, enhanced by intelligence they’ve never had access to before.

We’re not just excited about the agentic future. We’re building it, systematically, on first principles that put the needs of advertisers, agencies, publishers, and users at the center of everything we do.

The agentic future of advertising is genuinely exciting, but it is also important to acknowledge reality. We are all learning this as we go. Core protocols such as MCP and A2A are barely a year old, and the patterns for applying agentic systems to media buying are still emerging. This is new territory for everyone, and there are no shortcuts to getting it right.

AI can and should accelerate value creation across the ecosystem, but only if we remain disciplined, humble, and grounded in existing industry-agreed deterministic standards, shared object models, and semantics. The opportunity ahead is not reinvention for its own sake, but thoughtful evolution that preserves interoperability, trust, and scale as these technologies mature.

We see agentic approaches as an augmentation and evolution of the media industry, not a replacement. We are remodeling the kitchen and adding a new garage, not bulldozing the house. The goal is to build on what already works by using AI to reduce friction, improve outcomes, and strengthen the value exchange among advertisers, agencies, publishers, and consumers.

The purpose of standards is changing. The advertising ecosystem needed object models, schemas, and standardized APIs to support interoperability. Now, these same standards are needed for context so that agents are grounded in them and trained to execute actions asked in natural language with repeatable accuracy millions of times. That is our focus at the IAB Tech Lab.

What do you think? I welcome your perspectives on where agentic AI can create the most meaningful value, and where we should proceed carefully as an industry.

Join the conversation on LinkedIn and join the webinar on January 28.