The conference room walls are plastered with posters about AI transformation. The deck says "autonomous optimization." The demo runs on ChatGPT with a fresh coat of paint. And somewhere in Silicon Valley, a founder is practicing the line: "We're disrupting advertising with proprietary machine learning."

Jason White has heard this song before. Hell, he's heard it in four different keys across two decades.

"The biggest lie," he tells me, leaning back with the weary confidence of someone who's actually built things that work, "is that it's gonna uproot everything, automate everything." He pauses, and you can almost see him scrolling through the mental Rolodex of bullshit he's witnessed. "I've been working in some form of internet technology company for over two decades. And I've never seen an absolute, right? It just doesn't exist. I immediately discount it when somebody says everything."

This isn't cynicism. It's pattern recognition.

White has survived Viacom, OpenX, CBS Interactive, Fox MySpace, TrueCar, Arena Group, and multiple startups with acronyms that sound like pharmaceutical side effects. He invented real-time bidding structures when "programmatic" was still a word people had to explain at cocktail parties. He built actual exchanges, actual platforms, actual systems that moved actual money. In an industry where "building tech" often means "hiring an agency to make a deck," White is one of the rare ones who can read the code, architect the system, and tell you why it's probably going to break at 3 AM.

So when everyone is running around screaming that AI is either the messiah or a caffeinated raccoon in a data center, his voice matters. Because unlike the LinkedIn prophets and the keynote warriors, he knows what the inside of a machine actually looks like.

"It's the soup du jour," White says. "And that's why there's value in it."

He's talking about the hype cycle, that beautiful con that happens whenever capital floods into a sector. The story is always the same: This time it's different. This time it's real. This time we've cracked the code on something that will fundamentally transform everything about how humans interact with commerce, information, and each other.

White has seen this movie. He lived through the dot-com bubble, when every company with a URL was valued like they'd discovered cold fusion. He watched ad networks convince everyone they had magical behavioral targeting that was somehow superior to every other magical behavioral targeting. He survived programmatic's promise of perfect transparency (spoiler: there is none). He endured the great mobile migration when everyone suddenly needed an app for their app.

"The mantras and the BS is spewed because it's capital," he explains. "Everybody's got so much interest invested in it. And I think some people are gonna get disintermediated by it. And so, you know, they feel like they have to slap that jargon on."

Here's what they're not telling you in the pitch decks: According to industry data, somewhere between 60 and 70 percent of all "AI features" are just rule-based logic—the same if-then statements that have been around since the Carter administration—with maybe a large language model on top to make the interface nicer. It's like putting a Tesla body on a golf cart and calling it innovation.

White experienced this firsthand at Jiffy AI, his ad operations startup. "There's a form of automation called RPA that does have if-thens in it," he says. "But there's other components like OCR technologies that enable you to read a PDF and extract information from that and put that into platforms. So that's where RPA comes in."

That's real automation: optical character recognition, robotic process automation, actual technologies solving actual problems. But here's the thing—it's not sexy. It doesn't make for a good TED talk. You can't claim you're "disrupting the paradigm" when you're just making computers read PDFs faster.

So companies rename it. They rebrand it. They slap "AI" on the label and raise another round.

"I think that some people just misname it, you know," White says, and there's something almost generous in that phrasing. "Make no mistake, there is value in that. But just be transparent about it. You didn't build your own LLM. You're leveraging another LLM." And you built basically the equivalent of a PRD, a thoughtful PRD, that I don't have to do, which is great. You're providing a solution for me that's gonna do this thing that's gonna make my life easier in some form or fashion. "That's great. Just kind of be transparent on that."

Transparency. In advertising. That's funny.

Let me tell you what nobody mentions in the AI sales pitch: ChatGPT is only right about 50 percent of the time.

And it will argue with you about it.

"It will argue for five minutes that you are wrong and you will provide proof," I tell White. "No, that's wrong. I see how you could think that way. But two plus two really is five, and this is why."

"Yeah, that's right," he laughs. "And it glazes you. It tells you that you're right, right?"

This is the dirty secret of large language models. They're exceptionally confident and frequently wrong. They hallucinate facts, invent citations, and construct elaborate justifications for errors. They're basically every mediocre manager you've ever had, except they type faster.

"A lot of it centers on prompting," White explains. "Prompting is so critically important. And the human in the loop is vital." And that's the way that we look at it. Human architect is the creator and operators that are the human in the loop. Proofing, checking the paper, so to speak. "You're almost kind of like an English teacher or a math teacher that's looking over work and correcting it."

At his current company, they have over five agents working together with Model Context Protocols. "They cannot be fully autonomous," White says flatly. "This I can tell you. Prompts are extremely important and to your point, hallucinations are real things."

So when a vendor tells you their system is "fully autonomous" or "requires no human oversight," what they're really telling you is they haven't tested it enough to know where it fails. Or they have, and they're hoping you won't notice until after the contract is signed.

The economics make this funnier. A year ago, using AI cost maybe $50 a month. Now? Try $1,000 for the same compute. Token costs are rising. The cheap party powered by venture capital subsidies is ending. The LLM companies are burning billions monthly—ChatGPT reportedly loses a billion dollars a month—and eventually, someone has to pay.

"Not for long," White says when I mention this. Because of course. The business model is the same one Netflix and Amazon Prime used: Get you hooked at $5.99, then quietly raise it to $10.99, then $15.99, then suddenly your "affordable" AI solution costs more than your cloud infrastructure.

"These LLM companies aren't profitable," White notes. "So, you know, the revenue is gonna have to come from somewhere. I do believe they'll play in ads."

Of course they will. Everything plays in ads eventually.

Here's the part that keeps White up at night, and should probably keep you up too: What happens when ChatGPT or Claude or whatever LLM you're building on top of just... builds your feature themselves?

"I've seen some people post where they talk about those that are using the most tokens on ChatGPT," White says. "And you look at some of those companies and you go, ChatGPT is absolutely going to build that. And if you don't think that they're looking at it and going, hmm..."

Token usage is essentially a roadmap. Every API call your company makes is telling OpenAI or Anthropic exactly what use case has demand, what workflows are valuable, what features people actually pay for. It's market research delivered directly to your potential competitor.

This isn't speculation. We've seen this movie too. White remembers the games built on Facebook's platform—FarmVille, Mafia Wars, all those viral hits that made Zynga worth $10 billion. "And then, you know, boom," he says. "Unless you're the platform, you control the distribution, which the LLMs do, then you really have control."

Facebook didn't kill those games out of malice. They just changed the algorithm, adjusted the viral mechanics, decided to prioritize different content. When you build on someone else's platform, you're building on land they own. They can change the zoning laws whenever they want.

The LLM platforms will do the same thing. They're watching which wrappers get traction, which features drive usage, which business models actually work. And then they'll integrate them directly. Why would a user pay for your AI writing assistant when ChatGPT can do it natively? Why would they use your research tool when Claude can browse the web?

"A lot of these are natural progressions," I say.

"Well, we saw Facebook do it," White confirms.

The only defense is to build something that can't be easily replicated, something that requires deep domain expertise or proprietary data or genuine innovation. But that's hard. It's much easier to wrap an API and call it a product.

"Look, there's no question that we're in a bubble," White says. "I mean, we've seen these things before, right? But coming out of bubbles, 10 percent or so changed the world."

He's right about the math. Most of the companies claiming to revolutionize advertising with AI will be museum pieces within three years. You'll find them in the technology graveyard next to Flooz and Pets.com and every company that ever put "blockchain" in their pitch deck.

But some won't. Some will actually solve real problems. Some will build lasting businesses. The question is how to tell the difference before you've invested time, money, or your career in the wrong horse.

White offers a simple test: "Are we solving an expensive problem that will save money, give more bandwidth, make people more revenue? The more those concentric circles align, and you're really thoughtful about that, the more you have a winner. Maybe even a unicorn if you hit all three."

Save money. Give bandwidth. Make revenue. Three circles. If a technology hits all three, you might have something. If it only hits one, or worse, if it's just promising to hit them eventually, pending the next round of funding, well...

"Real tech should solve real problems," White says. It sounds obvious, almost trite. But in an industry where companies regularly spend millions on solutions searching for problems, it's radical.

The survivors won't be the ones with the best demos or the slickest decks. They won't be the ones using the word "autonomous" most enthusiastically. They'll be the ones who are honest about what they've built, transparent about what it can't do, and focused on solving actual pain points instead of chasing whatever's trending on LinkedIn.

"I think you're gonna see some solid companies," White says, and for a moment he sounds almost optimistic. He mentions AdCP, agents talking to each other, new protocols being developed in Slack channels. "It reminds me of the OpenRTB spec 1.0 conversations back in the day."

Maybe this time is different. Maybe 10 percent of the current AI frenzy really will change how advertising works. Maybe we'll look back and see this as the moment when automation finally delivered on its decades-old promise.

But if you're betting your company, your budget, or your career on it, White has one piece of advice, passed down from his father, a madman from the industry's earlier era:

"Never believe your own shit."

Coming in Part 2: Revenue discipline, shiny object syndrome, and why the metaverse ate everyone's R&D budget. Plus: the middle managers who are destroying your culture and don't even know it.

The bill comes due eventually. Always does.

Jason White's AI expenses: $50 a month last year. $1,000 a month now. Same work. Token costs climbing. LLM companies hemorrhaging billions monthly. The cheap party is ending.

"These LLM companies aren't profitable," White says. "So, you know, the revenue is gonna have to come from somewhere."

This is the hidden cost of AI theater: opportunity cost. Lost quarters chasing something that was never going to work because it was never supposed to work. It was supposed to impress. The deck. The keynote. The board presentation about transformation.

Meanwhile, the thing that actually makes you money sits there, waiting for someone to remember it exists.

White has watched this movie from the inside of more companies than most people work for in a lifetime. He knows what revenue discipline looks like. And more importantly, he knows what it doesn't look like.

Arena Group: Three CEOs in a year. Private equity stepping in. Résumés updating.

"That business was being run, like most businesses were, over the past five, 10 years as a revenue game, right?" White explains. "We're playing the top line, you know, acquire, acquire, acquire, and not really focused on profitability."

Growth at all costs. Revenue hockey sticks. Never mind that you're burning cash. The next round will fix it. Profitability will just... happen.

Except it doesn't.

Arena Group was "wildly unprofitable." Now they're profitable. What changed?

"You laser focus on getting that thing that is making you the money and scale it and repeat it. Like it's got to be scalable and repeatable and you keep focusing on profitability. You only carve out R&D for new products and business lines that went through that same rigor."

Scalable. Repeatable. Rigorous. The exact opposite of every pitch deck promising to revolutionize everything with AI.

But it works.

"Why do companies making $100 million profit suddenly spend $100 million on the metaverse?"

White pauses. "My experience is that they usually don't" think it through.

He mentions an agency—Sir Martin Sorrell's, though he's too polite to name it—that spent their entire R&D budget on the metaverse. All of it. Now they have nothing. Because now they want to spend it all on AI.

This is FOMO as business strategy. When the narrative becomes more important than the numbers.

White's father taught him something: "Never believe your own shit."

But they do.

CBS. Someone pitched Programmatic TV. The future! The convergence!

"If you really understand how linear is done, it's impossible," White says.

He kicked the tires. Popped the hood. "Sure enough, there's little people behind the green curtain running data back and forth."

No AI. No automation. Just people manually moving data, making the demo work long enough to close the deal.

Then Google came with programmatic guaranteed pricing. Take what works in ad exchange—15-20% of a $2 CPM (about $0.40)—and apply it to premium deals.

White did the math. "So you want the same model on a $30 CPM and your effective CPM is $6?"

The Google rep's response: "Well, you have a great point. But what if I just told you to go pound sand?"

At least he was honest.

Is culture the posters or how people behave when things go wrong?

"I think it's how people behave when they go wrong, when things go wrong," White says. "I think it's all about learning from it."

He's worked places with cat posters. Hang in there, baby. All the laminated lies HR calls culture.

Real culture is what happens when someone fails. Can you test? Can you learn? Can you fail without being executed?

"That's an internet mantra. Like that's science mantra. It's like, you know, test, failure's good. We applaud that now. Don't do the same thing wrong over and over again. That's different. But I see a lot companies get that wrong. People get fired for making like one mistake. That's not healthy. That's not a good culture. It's toxic."

He credits Jim Lanzone—"the best in the business"—for showing him good culture. CBS as a "Federation starship." Different brands, different cultures, but connective tissue holding it together.

"It's not just the fun activities that you do. It's like how you treat people."

What destroys culture fastest?

"I think middle management does personally," White says.

"They're not doers anymore. They're just parroting what leadership is basically telling them to do."

Office Space. Taking paper from one side of the building to the other. Calling it leadership.

They kill culture one interaction at a time.

Twenty years into programmatic. Billions invested. Thousands of companies promising transformation.

"None of the stuff I see actually solves an issue I have when buying ads."

White's test: "Are we solving an expensive problem that will save money, give more bandwidth, make people more revenue? The more those concentric circles align, and you're really thoughtful about that, the more you have a winner. Maybe even a unicorn if you hit all three."

Three circles:

Save money.

Give bandwidth.

Make revenue.

Hit all three, you survive the hype cycle.

Here's what we didn't talk about: The intermediaries taking 30-50% margins. The SSPs that aren't actually integrated into DSPs. Why Trade Desk hasn't added a new SSP in eight years. The ad networks that actually worked before programmatic promised transparency and delivered opacity.

Want the technical deep-dive? The part where White explains exactly how much money you're losing and who's taking it?

That's ADOTAT+.

Part 3: The Programmatic Paradox—Where White walks through Arena Group cutting out intermediaries, why curation is just ad networks with better marketing, and the specific mechanics of how you're being charged $6 for work that used to cost $0.40.

Part 4: The Future of Advertising Technology—AdCP protocols. Agents talking to each other. What actually survives. And White's 10-year vision for an industry that's spent the last 20 years solving the wrong problems.

The free parts told you what's broken. The paid parts tell you how much it's costing you and what to do about it.

Your CFO will thank you. Or fire you for not knowing sooner.

[Subscribe to ADOTAT+]

You’ve seen the problem.
You haven’t seen who profits from it.

The missing incentives, the quiet failure points, the math nobody puts in the deck.
Only in ADOTAT+.

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