Agents Aren’t the Point. Signals Are. Start Here for the latest AI Insights for Business

Separating the wheat from the chaff on agentic advertising.

Every year is the year of something.

It was the year of mobile for a few years. Then it was the year of CTV. And if CES and ALM are any indication, we’ve now entered the year of AI theatre.

Not because AI isn’t real. It is.

There are meaningful use cases already driving real money and real outcomes. But the jockeying for relevancy—trying to land a logo in the right corner of the Lumascape and a buzzword in the first paragraph of the deck—has muddied the water.

So let’s separate what’s real from what’s bullshit. And more importantly: what matters from what doesn’t.

The year of AI theatre

“Agentic advertising” is the hottest phrase in the room because it sounds like the next inevitable chapter: software that doesn’t just automate tasks, but actually decides and acts—planning, buying, optimizing, adapting creative, negotiating, measuring, iterating.

In theory, it’s the end of dashboards and the beginning of outcomes.

In practice, most of what’s being sold as “agentic” right now is one of three things:

Helpful? Sure. Revolutionary? Not yet.

The agent demos are seductive because they’re easy to show. What’s hard to show—and much harder to build—is the underlying prerequisite that makes agents useful instead of dangerous:

clean, permissioned, standardized signals that a system can trust.

Agents aren’t the point. Signals are.

The chasm is widening (and everyone can feel it)

At the same time, the gap between the walled gardens and the open ecosystem has widened into a canyon.

Investment in traditional digital publishers keeps getting diluted relative to Meta, Google, Amazon, and a shrinking list of “must buy” CTV platforms. Not because people stopped consuming premium content on the open web—consumer engagement didn’t evaporate—but because the buying experience and outcome confidence are asymmetrical.

Walled gardens offer:

The open ecosystem offers:

That doesn’t mean the open web is doomed. But it does mean the open web’s advantage can’t just be “reach.” It has to be signal strength—and signal portability—at a level that lets buyers trust outcomes.

AI hasn’t changed media yet (except where it has)

Here’s a statement that’s going to annoy a lot of people: AI hasn’t had a broad impact on media yet.

Meta has. AppLovin has. A few others have. But for most of the ecosystem, the impact has been incremental: a little efficiency here, a little automation there, a few nicer insights sprinkled on top of the same old pipes.

Why? Because AI doesn’t create leverage by existing. It creates leverage when it has:

That combination is exactly what walled gardens have by design. And it’s exactly what the open ecosystem is still trying to assemble—often while arguing about standards in one room and shipping point solutions in another.

The irony: programmatic is already the right substrate for “agentic”… if it’s not being gamed

Here’s the funny part: the programmatic bidstream is actually a pretty solid foundation for semi-autonomous execution.

It’s real-time. It’s machine-readable. It’s already an environment where software is making micro-decisions at scale.

So yes—if you squint, you can absolutely see how agentic systems could sit on top of programmatic pipes and do meaningful work: supply selection, bid policies, pricing strategies, creative eligibility, frequency controls, attention and sustainability constraints—the whole thing.

But there’s one big catch: agents don’t fix a broken market. they scale it.

If the inputs are manipulated—if participants are spoofing signals, misrepresenting supply, hiding fees, washing paths, or gaming measurement—then an agent doesn’t create intelligence. It creates acceleration. It just makes bad decisions faster and more confidently.

So the bidstream is the substrate, sure. But the future hinges on whether we can make the signals trustworthy enough for semi-autonomous decisioning.

Post-ATT: the blueprint is modeling + creative, not magic identity

If you want a clean blueprint for what works post-ATT, you don’t need to read a thousand thought leadership pieces. You just need to look at who adapted fastest.

ATT didn’t just hurt “targeting.” It constrained visibility into cross-app behavior—the thing the open ecosystem leaned on for years.

That forced a survival event:

Meta’s advantage isn’t simply “they have AI.” It’s that they have the right kind of signals (engagement, consumption, device, time, behavior patterns) and the ability to constantly test and tune creative—offers, images, copy, CTAs—at a scale humans can’t touch.

AI is unbelievably good at exploring creative permutations when it has signal and feedback. It’s not “creative.” It’s combinatorial optimization with taste constraints.

And that’s the uncomfortable part for the open ecosystem: the biggest performance gains are coming from systems that can connect signal → decision → creative → outcome in a closed loop.

If we squint: the open ecosystem survives by becoming signal-rich and decisionable

If we squint through the looking glass, you can see a future where the platforms that slow (or save) traditional digital are the ones that bring enough signal to the party to build a sophisticated model that delivers the right ad, to the right person, at the right time.

The promise of advertising that has been marketed for decades… and rarely delivered consistently outside of closed systems.

But the open ecosystem doesn’t need to become a walled garden to compete. It needs to become:

This is where “starting at the source” becomes more than a tagline. The closer you are to the actual consumer experience—the player, the content, the engagement—the more defensible your signals become. (And yes, that’s also why platforms embedded in the viewing stack talk so much about being signal generators, not just delivery pipes.)

What agents actually change: strategy at scale, not just execution at scale

Here’s where agentic gets real.

Humans can do strategy. Humans can do relationships. Humans can do taste and brand intent.

Humans cannot continuously design, run, and refine thousands of micro-strategies across:

Agents can. That’s the promise.

Not “an AI that buys media for you.”

An AI that can test and optimize combinations that are impractical for a person to manage—as long as the system has clean signals and guardrails.

This is also why the most credible “agentic” roadmaps look less like chatbots and more like curation + forecasting + portability + measurement: the unsexy infrastructure that turns signal into repeatable outcomes.

The part everyone wants to skip: autonomy is proportional to signal quality

Now the punchline.

The only way agentic advertising works—at any meaningful scale—is if the data is clean.

Standards matter because they scaffold consistency. But standards don’t create differentiation.

Because in the next chapter of media, the survivors won’t be the companies who merely adopt standards.

They’ll be the companies who have unique and exclusive signals—signals that improve decisioning in a way other participants can’t easily replicate.

Signals and context aren’t just the secret sauce in media. They’re the secret sauce in anything agentic.

And that’s why this whole conversation gets inverted:

People are arguing about agents like they’re the product.

They’re not. Agents are a capability layer — a way to scale decisioning across complexity.

The real battle is over who owns the signals, who can standardize them, who can prove them, and who can convert them into outcomes without poisoning the consumer experience.

Because in an agentic world, the supply chain doesn’t just execute.

It decides.

Agents Aren’t the Point. Signals Are. Start Here for the latest AI Insights for Business

Separating the wheat from the chaff on agentic advertising.

Every year is the year of something.

It was the year of mobile for a few years. Then it was the year of CTV. And if CES and ALM are any indication, we’ve now entered the year of AI theatre.

Not because AI isn’t real. It is. There are meaningful use cases already driving real money and real outcomes. But the jockeying for relevancy—trying to land a logo in the right corner of the Lumascape and a buzzword in the first paragraph of the deck—has muddied the water.

So let’s separate what’s real from what’s bullshit. And more importantly: what matters from what doesn’t.

The year of AI theatre

“Agentic advertising” is the hottest phrase in the room because it sounds like the next inevitable chapter: software that doesn’t just automate tasks, but actually decides and acts—planning, buying, optimizing, adapting creative, negotiating, measuring, iterating.

In theory, it’s the end of dashboards and the beginning of outcomes.

In practice, most of what’s being sold as “agentic” right now is one of three things:

Workflow automation (faster trafficking, reporting, QA),

Recommendations (a human clicks “approve”), or

Creative generation (a firehose of variants with no real decisioning loop).

Helpful? Sure. Revolutionary? Not yet.

The agent demos are seductive because they’re easy to show. What’s hard to show—and much harder to build—is the underlying prerequisite that makes agents useful instead of dangerous:

clean, permissioned, standardized signals that a system can trust.

Agents aren’t the point. Signals are.

The chasm is widening (and everyone can feel it)

At the same time, the gap between the walled gardens and the open ecosystem has widened into a canyon.

Walled gardens offer:

dense first-party signals,

closed-loop measurement,

native creative optimization,

fewer moving parts,

less ambiguity.

The open ecosystem offers:

fragmentation,

inconsistent taxonomies,

inconsistent measurement,

and still way too much spreadsheet-and-PDF commerce pretending to be modern infrastructure.

That doesn’t mean the open web is doomed. But it does mean the open web’s advantage can’t just be “reach.” It has to be signal strength—and signal portability—at a level that lets buyers trust outcomes.

AI hasn’t changed media yet (except where it has)

Here’s a statement that’s going to annoy a lot of people: AI hasn’t had a broad impact on media yet.

Meta has. AppLovin has. A few others have. But for most of the ecosystem, the impact has been incremental: a little efficiency here, a little automation there, a few nicer insights sprinkled on top of the same old pipes.

Why? Because AI doesn’t create leverage by existing. It creates leverage when it has:

enough signal to build a real model,

the ability to act inside the system, and

a feedback loop to learn continuously.

That combination is exactly what walled gardens have by design. And it’s exactly what the open ecosystem is still trying to assemble—often while arguing about standards in one room and shipping point solutions in another.

The irony: programmatic is already the right substrate for “agentic”… if it’s not being gamed

Here’s the funny part: the programmatic bidstream is actually a pretty solid foundation for semi-autonomous execution.

It’s real-time. It’s machine-readable. It’s already an environment where software is making micro-decisions at scale.

So yes—if you squint, you can absolutely see how agentic systems could sit on top of programmatic pipes and do meaningful work: supply selection, bid policies, pricing strategies, creative eligibility, frequency controls, attention and sustainability constraints—the whole thing.

But there’s one big catch: agents don’t fix a broken market. they scale it.

If the inputs are manipulated—if participants are spoofing signals, misrepresenting supply, hiding fees, washing paths, or gaming measurement—then an agent doesn’t create intelligence. It creates acceleration. It just makes bad decisions faster and more confidently.

So the bidstream is the substrate, sure. But the future hinges on whether we can make the signals trustworthy enough for semi-autonomous decisioning.

Post-ATT: the blueprint is modeling + creative, not magic identity

If you want a clean blueprint for what works post-ATT, you don’t need to read a thousand thought leadership pieces. You just need to look at who adapted fastest.

ATT didn’t just hurt “targeting.” It constrained visibility into cross-app behavior—the thing the open ecosystem leaned on for years. That forced a survival event:

model what you can see,

infer what you can’t,

and tighten the creative loop until performance recovers.

Meta’s advantage isn’t simply “they have AI.” It’s that they have the right kind of signals (engagement, consumption, device, time, behavior patterns) and the ability to constantly test and tune creative—offers, images, copy, CTAs—at a scale humans can’t touch.

AI is unbelievably good at exploring creative permutations when it has signal and feedback. It’s not “creative.” It’s combinatorial optimization with taste constraints.

And that’s the uncomfortable part for the open ecosystem: the biggest performance gains are coming from systems that can connect signal → decision → creative → outcome in a closed loop.

If we squint: the open ecosystem survives by becoming signal-rich and decisionable

If we squint through the looking glass, you can see a future where the platforms that slow (or save) traditional digital are the ones that bring enough signal to the party to build a sophisticated model that delivers the right ad, to the right person, at the right time.

The promise of advertising that has been marketed for decades… and rarely delivered consistently outside of closed systems.

But the open ecosystem doesn’t need to become a walled garden to compete. It needs to become:

signal-rich (real, high-fidelity signals),

permissioned (usable without crossing privacy lines),

standardized (so the signals mean the same thing across pipes),

and auditable (so buyers can trust the “why”).

What agents actually change: strategy at scale, not just execution at scale

Here’s where agentic gets real.

Humans can do strategy. Humans can do relationships. Humans can do taste and brand intent.

Humans cannot continuously design, run, and refine thousands of micro-strategies across:

media decisioning (pricing, supply path selection, pacing, frequency, format mix, auction policies),

creative decisioning (offer, image, CTA, tone, duration, placement rules),

and optimization (not once a week, but continuously, with constraints).

Agents can. That’s the promise.

Not “an AI that buys media for you.” An AI that can test and optimize combinations that are impractical for a person to manage—as long as the system has clean signals and guardrails.

This is also why the most credible “agentic” roadmaps look less like chatbots and more like curation + forecasting + portability + measurement: the unsexy infrastructure that turns signal into repeatable outcomes.

The part everyone wants to skip: autonomy is proportional to signal quality

Now the punchline.

The only way agentic advertising works—at any meaningful scale—is if the data is clean.

consistent definitions,

clear permissions,

traceable provenance,

and standards that reduce ambiguity.

Not because AI isn’t real. It is.

There are meaningful use cases already driving real money and real outcomes. But the jockeying for relevancy—trying to land a logo in the right corner of the Lumascape and a buzzword in the first paragraph of the deck—has muddied the water.

ATT didn’t just hurt “targeting.” It constrained visibility into cross-app behavior—the thing the open ecosystem leaned on for years.

That forced a survival event:

Not “an AI that buys media for you.”

An AI that can test and optimize combinations that are impractical for a person to manage—as long as the system has clean signals and guardrails.

Because in the next chapter of media, the survivors won’t be the companies who merely adopt standards.

They’ll be the companies who have unique and exclusive signals—signals that improve decisioning in a way other participants can’t easily replicate.

Because in an agentic world, the supply chain doesn’t just execute.

It decides.