Most of this analysis originally lived behind the ADOTAT+ velvet rope about a month ago. We’re letting some of it out now not out of generosity, but because the industry has finally stumbled into the same conclusions and is pretending it got there on its own.

This isn’t about demos.
This isn’t about buzzwords.
This is about incentives. Always was.

AI in adtech has become the industry’s most dependable stage act.

Same machinery.
Same pipes.
Same toll booths.

But now everything wears better clothes. Add a few adjectives like “agentic,” “predictive,” and “autonomous,” dim the lights, cue the breathless keynote music, and hope nobody asks the rude question: what actually changed?

We’re told these systems learn.
That they anticipate.
That they self-optimize.

The implication floats just beneath the surface: intelligence has arrived, humans are optional, and the spreadsheet has finally become sentient.

Peel off the gloss and the truth underneath is neither exciting nor futuristic. It’s old. It’s boring. It’s merciless.

AI does not optimize for advertiser performance.
AI optimizes for the business model that pays its salary.

Nobody needs to say this out loud. Everyone involved already knows the rules of the game.

No villains required. No secret cabals. Just incentives doing exactly what incentives always do: pointing downhill.

If you want to understand what AI is really doing in advertising, stop watching product launches and start reading earnings transcripts.

Amazon’s advertising business didn’t gently grow. It exploded, jumping from roughly $26B to over $61B in four years, a 23.7% CAGR, precisely as it layered in automated bidding, AI-driven optimization, and “assistive” buying systems.

The Trade Desk posted 18% YoY revenue growth, with Kokai positioned as the intelligent engine quietly humming underneath.

These companies never publish the one metric that would actually settle the debate:

Revenue per advertiser.

Because once you see that number, the question of who the AI is optimizing for stops being philosophical.

Here’s the industry’s least shocking open secret:

Most of what gets marketed as “AI” is the same machine-learning skeleton from 2012 wearing trendier shoes.

Nothing wrong with that tech. It works. It always has. The deception is pretending it suddenly woke up one morning and developed opinions.

Generative tools? Agentic systems?
Those are the stage props. Incredible in demos. Unstable in production. Too unpredictable to be trusted with real budgets at scale.

Which is why, when platforms brag about AI-driven improvements like 94% higher CTRs or 26% lower CPAs, they never show causality. No baselines. No counterfactuals. No advertiser-level truth.

Because the revolution didn’t happen in the math.
It happened in the vocabulary.

If you want to know what an AI system is actually designed to do, don’t read the brochure. Watch its behavior in the wild.

And every quarter, like clockwork, the same numbers get celebrated:

Missing from the slide deck every single time?

Verified advertiser ROI.

Industry-wide, AI-driven programmatic spend is growing 13–15% annually, yet no one publishes incremental conversion data. Not because it’s hard. Because it’s inconvenient.

At some point, calling this intelligence becomes dishonest.
It’s revenue engineering with a glossy dashboard.

AI’s obsession with precision has quietly become one of its biggest liabilities.

Platforms love to talk about CPAs, in-market segments, and intent signals. Meanwhile, the actual experience looks like this:

Campaigns hammer the same users over and over, like a telemarketer convinced the next call will finally work.

Marketers won’t say this onstage. They whisper it privately:

AI doesn’t underperform because it’s wrong.
It underperforms because it’s too certain about bad signals.

The fallout is predictable:

Lumen shows template-driven AI creative loses attention faster.
Amplified Intelligence shows sharper drop-offs.
Realeyes shows weaker emotional response curves.

This is how “precision” becomes asymmetry: a strategy so narrowly optimized it forgets most of the market exists.

He’s not talking about AI as technology.
He’s talking about AI as capital gravity.

His view is uncomfortably simple:

This isn’t a vision of what AI should become.
It’s a map of what investors will reward.

And history is clear: consolidation always grows platform revenue faster than advertiser value.

AI didn’t invent adtech’s incentive problems.
It industrialized them.

More automation.
More opacity.
More budget movement away from visibility.
More “performance” that collapses under causal scrutiny.
More consolidation among the same platforms already holding the keys.

Independent verification remains nonexistent.
Lift is modeled, not measured.
Attribution is correlation wearing a lab coat.
AI creative is a fatigue machine.

And still, no platform publishes the only number that matters:

Incremental conversions.

Until someone demands real experiments, real baselines, and real accountability, the honest conclusion is the least exciting one:

AI in advertising is working exactly as designed.
Just not for the people paying the bills.

Stay Bold, Stay Curious, and Know More than You Did Yesterday