For years, the advertising industry was promised that some form of workflow automation would remove operational complexity and improve efficiency.

We were told that manually booking campaigns would disappear. Reporting would become seamless and fully integrated. Optimisation would become self-driving. Campaign measurement, pacing, billing, reconciliation, targeting, and yield management would all become cleaner, faster, and less directly dependent on human intervention.

And to a degree, that happened, programmatic advertising did automate parts of media buying and selling. But anyone who has worked inside ad operations, programmatic, publisher monetisation, or media trading knows what really happened. A myriad of disconnected systems emerged: wrapper tags, header bidding, clean rooms, identity, consent signals, floors, deals, taxonomies, brand-safety rules, reporting joins, waterfall dynamics.

From an engineering perspective, a problem got simply reduced to latency, executing tasks faster and faster, but with no guarantee that any two systems describing the same impression will ever agree on what actually happened. Events, schemas, signals, platforms, partners, policies, and measurement definitions, with no single source of truth.

And this is exactly where classical automation reaches a ceiling, focused on only forward-execution, a fixed sequence or steps, predefined rules, authored against a known interface, that assumes the world holds still. Human operated, 9-5 monitored, gut-feel tweaked, and best-effort delivered.

That is where agentic differentiates.

Not because it is another buzzword. Not because it replaces every human operator. And not because it magically solves every problem in the media supply chain.

It matters because agentic workflows are fundamentally different from the process automation models the industry has tried before.

Agentic has agency.

An agent takes a goal, an outcome, maintains context, transaction history, uses a powerful frontier LLM reasoning layer, has specialised skills, and tools it can utilise, 24h a day, 365 days a year. Instead of blindly executing a predefined sequence, it can inspect the situation, decide what information it needs, call the right tools, evaluate the results, and determine the next steps based on that.

Automation follows inputs. Agents reason against outcomes.

Read side by side, it reveals the most important architectural lesson in this space: Agentic and Automation are not competing for the same job.

Agentic is deliberately asynchronous, its negotiations take minutes or days, they accommodate embedded human judgment for complex terms, and pre-approval for known paths. An OpenRTB auction evaluates, and settles in under a hundred milliseconds. These are different layers.

Agent is an application layer, where intent and strategy are expressed and where an agent’s reasoning loop earns its keep. Automation is a transaction layer, where individual decisions execute under hard performance constraints and where you want deterministic rules, not a language model.

And most importantly this is no longer theoretical. The Ad Context Protocol (AdCP), governed by AgenticAdvertising.org lets agents reason, discover inventory, buy media, build creatives, and activate audiences across platforms. IAB’s Agentic Real Time Framework (ARTF), runs as a co-located container that handles the programmatic bidstream during the auction to cut execution latency even further and smarter.

As a founding member of AgenticAdvertising.org, and an active observer of the IAB Tech Lab, at pubX we are bringing all this together into the next generation of agentic advertising infrastructure designed for speed, accuracy, transparency and safety.

Agentic Advertising. Built on trust.