I’ve always loved sci-fi movies and the way they portray futuristic worlds shaped by advanced technological innovation. One of my favorite movies is the ’80s classic “The Terminator.” And the more I hear about “agentic AI” in ad tech, the more I think the film is a perfect analogy for what’s happening right now.
The Terminator had one job: to find Sarah Connor and “terminate” her. Simple enough, except the machine didn’t know which Sarah Connor. So it did what any advanced cybernetic organism without real intelligence does: It went through the phonebook one name at a time, working the list sequentially until it found the right target.
That’s not intelligence; that’s automation with sunglasses and a leather jacket.
It eventually gets the job done. But it doesn’t know anything. It has no real training data. No context. No ability to prioritize. It’s executing a workflow, not reasoning through one.
The agentic AI hype is real. The intelligence behind it, less so.
Over the past year, practically every TV and digital advertising platform has announced some version of “agentic AI.”
The Trade Desk launched Koa Agents, its most consequential product announcement in years, described as a system where marketers articulate campaign goals and AI handles the rest. Vibe.co explicitly named __“agentic AI capabilities for media buying” __as a core pillar of its $50 million Series B. And MNTN’s CEO has talked openly about agentic AI’s potential to “shake up media buying by allowing buyer and seller agents to handle sales from end to end.”
What they’re describing is true. Most “agentic AI” tools eliminate manual busy work by monitoring budget pacing, flagging anomalies and automating IO workflows. They move dollars between line items based on preset rules and reform reports. Is this efficient? Yes. Is it intelligent? Not so much.
There’s even a technical term for what most of these systems are actually running on: deterministic, rules-based logic. If cost-per-click exceeds X, lower the bid. If the budget hits Y, the line item. It’s reactive, executing mindlessly.
That’s the “A” without the “I.”
The difference between agentic AI and real AI is architecture and data
Traditional AI is built on proprietary training data. A model trained on enough of the right data doesn’t just execute tasks; it learns patterns humans can’t see across a data set no human could process, surfacing decisions that are faster, more comprehensive and better than what manual analysis could produce.
At Tatari, we’ve spent a decade building the Performance TV data set, a proprietary record of TV media investments and outcomes across hundreds of brands, billions of dollars in spend and years of campaign data.
When we apply traditional AI to that data set, two things happen that aren’t possible with agentic automation:
Planning at real scale: Instead of choosing from a curated shortlist of ~100 inventory options, our models evaluate hundreds of thousands of purchasable inventory items and find the combinations that will actually perform for a given brand. No human planner can hold that much context. No rules-based agent can reason across it.Impression-level routing: We can use that outcomes data set to route the most relevant impression to the best-fitting brand based on what has actually driven outcomes for similar advertisers across similar inventory over several years.
That’s a data set no platform can wrap an agent around and replicate. You can automate a workflow in days or weeks. You can’t manufacture 10 years of outcome data.
Why this distinction matters for advertisers.
The industry is making a category error that will cost advertisers real money.
When platforms claim “agentic AI” and mean “automated budget pacing,” buyers start believing the intelligence problem is solved. They stop asking what the underlying model was trained on. They stop asking whether the recommendations are based on historical outcomes or just heuristics. They accept efficiency gains as a proxy for intelligence.
The Terminator became more dangerous in the sequels, not because it got a better workflow, but because it finally had proper training data. That’s the version of AI that TV advertising should be building toward – the difference between agentic AI and real AI.
Let’s start saying “hasta la vista” to workflows dressed up as intelligence and start demanding the real thing.
For more articles featuring Philip Inghelbrecht, click here.