There's a piece making the rounds right now by Matt Shumer, an AI startup founder who essentially wrote a secular book of Lamentations about how AI already replaced him at his own job and the rest of you are next. He compares the current moment to February 2020. Everyone's at brunch. Nobody's stockpiling toilet paper yet. Three weeks later your office is your couch and your kids won't stop touching your monitor during Zoom calls.

He's right. About almost everything. The capabilities are real. The acceleration is real. The compounding thing where AI builds the next AI which builds the next AI? Real and genuinely unsettling.

But Shumer writes for tech. I write for adtech. And in adtech we have a very specific, very expensive, very deliberate problem that his piece doesn't touch.

The people selling you AI don't want you to understand it. Because the moment you do, their pricing falls apart.

Let me explain. And then, because I'm feeling generous and maybe a little reckless, I'm going to hand you the actual playbook. For free. Right here in Part 1. No $5,000 conference ticket. No "AI Masterclass" run by someone whose LinkedIn said "affiliate marketing ninja" eighteen months ago. Just the thing itself.

You're welcome.

Here's what nobody on the conference circuit wants you to internalize: AI in adtech is not hard. The technology is not mysterious. The applications are not arcane. What's happening across our industry right now is one of the most straightforward use cases for large language models that exists, and it's being deliberately, systematically made to sound complicated because confusion is where the margin lives.

There are five gears in this machine. They all turn together.

Pricing power. If AI sounds mystical, vendors can justify higher take rates, SaaS fees, and "AI tax" line items that you can't benchmark against anything. When it's framed as rare, specialized wizardry instead of what it usually is, which is pattern matching on ad logs, it gets real easy to defend headcount, professional services fees, and "black box optimization" margins in every RFP and M&A deck that crosses your desk. You can't negotiate against magic. That's the whole point.

M&A theater. AI is table stakes in pitch decks now. It's a screening criterion in adtech M&A. Everyone needs an "advanced" story to look fundable or acquirable. Here's the trick though: if your AI is understandable, it looks like something a bigger platform could clone next quarter or a client could bring in-house by July. If it's framed as deep, proprietary, almost spiritual complexity? Strategic moat. The vaguer the better. That's not a bug. That's the business model.

Old tech in a new trench coat. A lot of what currently ships as "AI" in programmatic stacks is glorified bid rules, lookalike audiences, or regression analysis that got a fresh landing page and a generative AI logo. The complexity narrative hides how incremental the actual improvement is. It also makes it harder for buyers to notice that the same outcomes could come from clean supply paths, better measurement, or simpler tooling. Not a shiny new AI layer with a six figure annual contract.

The priesthood. This is the one that makes people squirm. If AI is simple and productized, you don't need as many "strategists," "solutions engineers," or "platform whisperers." If it's arcane and mysterious, you still need specialists to interpret and "tune" it. Keeping AI hard preserves the priesthood. The people who translate knobs and dashboards for CMOs and CFOs get to keep control of the narrative and the budgets. It's job protection dressed up as expertise. Not all of it, obviously. Some of these people are brilliant. But the incentive structure is real and nobody talks about it.

The narrative trap. Boards, CEOs, and now even Super Bowl audiences have been sold AI as epoch defining, civilization scale technology. So vendors feel pressure to posture accordingly. The reality is often workflow shortcuts and prediction. Useful stuff! Genuinely valuable. But the story has to sound grander than that, so it leans into jargon, complexity, and perceived difficulty. The gap between what AI actually does in your stack and what the press release says it does is wide enough to park a fleet of ad servers in.

The net effect is simple: the harder AI sounds, the easier it is to sell, mark up, and defend. That's true until someone forces the conversation back to concrete questions like "what auction levers did this actually pull?" and "how much incremental lift did we really get?" and "can you show me that without your own attribution model grading your own homework?"

All of that above? It's the diagnosis. Here's the medicine. And I'm putting it right here in the free article because proving that this is accessible is the whole point.

You don't need a $5,000 conference. You don't need a certification from someone who discovered the letters A and I fourteen months ago. You need about ten minutes and a willingness to type in plain English.

Here's how you actually use AI for your adtech work, right now, today.

Step one: stop asking it for facts. AI is not a search engine. It's not an encyclopedia. If you type "what is SPO?" you're using a Ferrari to go get the mail. The value is not in what it knows. The value is in how it thinks. Ask it to argue with you. Ask it to stress test a vendor pitch. Ask it to poke holes in your own strategy. Ask it to play the role of a skeptical CFO reviewing a DSP's claimed "AI optimization" and identifying where the margin, risk, or unverifiable uplift could be hiding. That's a real prompt. That works. Right now. Today.

Step two: give it context like you're onboarding an analyst. The number one mistake people make with AI is being vague. Don't say "analyze this campaign." Say "you are an adtech industry analyst, here is the company, here is their positioning, here is their product, here is who they sell to, here is my hypothesis that their agentic AI pitch is lipstick on workflow RPA. Now: infer their business model and hidden incentives, give me ten questions I should ask in an interview that would falsify their claims, and tell me what data I'd need to verify each answer." That's not a course. That's a paragraph. And the output will be better than most of what you'd get from a consultant billing you four hundred dollars an hour.

Step three: use it as a peer reviewer, not a ghostwriter. Write your analysis. Form your opinion. Then feed it back to the AI and say "play the role of in-house counsel at this vendor. Where would you attack this? What corrections would you push for?" Or: "act as a skeptical CFO. Where are my claims not well supported by the data I've shown?" This is the move that separates the people who use AI from the people who let AI use them. You do the thinking. AI pressure tests the thinking. That's the workflow.

Step four: build repeatable systems. The real power isn't one clever prompt. It's turning your research process into something you can run again and again. Upload earnings transcripts and ask AI to build a timeline of every mention of "AI" or "agentic," categorized by claim type, evidence provided, and measurable predictions you can track over time. Feed it supply paths, domains, and sellers.json entries and ask it to cluster by MFA risk signals. Give it a retailer's measurement PDF and ask it to identify where the methodology could systematically over-attribute lift to their own media. These aren't hypotheticals. These are things you can do this afternoon in a chat window.

Step five: make it argue both sides. First pass: let it generate a neutral take. Second pass: "rewrite as if you believe 60% of AI in adtech M&A is theater. Identify which deals are most vulnerable to that critique and why." Third pass: "now defend the buyers' logic. What defensibility could be real?" Then for each draft: "highlight logical leaps, missing baselines, and places where I'm asserting without falsifiable evidence." This is how you actually get smarter. Not by having AI agree with you. By making it fight you.

Shumer is right that AI is transforming everything and that people need to wake up. Where he and I diverge is on who benefits from the confusion.

In adtech, the confusion is the product. The complexity is manufactured. The mystique is monetized. And the $5,000 conferences, the "AI Expert" badges, the breathless panel discussions about "agentic intelligence" from people who couldn't explain a supply path optimization if you spotted them the first three words? That's not education. That's a toll booth on a road that's actually free.

The tools cost twenty bucks a month. The prompts are written in English. The concepts are not harder than what you already know about programmatic, measurement, or media buying. If you've ever read an earnings call and thought "that's not what the data says," you already have the instinct. AI is just the thing that lets you run that instinct at scale, at speed, without needing to hire three analysts.

So here it is, the big secret that is supposedly worth five grand: talk to the AI like a smart colleague, give it real context, make it argue with you, and never let it or the vendor selling it speak in a language you can't pressure test.

That's it. That's the whole thing. The emperor has no algorithm. He just has very good tailors and a conference budget.

Parts 2 through 4 will go deeper: actual prompt templates for fraud detection and campaign audits, a map of what's genuinely hard versus what's just dressed up as hard, and why the "AI priesthood" forming in adtech is the single biggest threat to independent publishers and honest buyers in the market right now.

But Part 1 was free because you deserved to know the shape of the thing before anyone else charged you to see it.

Now go open a chat window and start typing. It's not that complicated. That's literally the point.