AI Has a Snake Oil Problem

Let me get one thing out of the way before the pitchforks come out, because I know how the headline reads. I am not an AI hater. Just the opposite.

As a leverage machine, AI is absurd, in the best way. It has compressed the software development cycle down to something that would've sounded like science fiction three years ago. It lets me move at obscene speed. I genuinely believe it's going to be one of the most powerful technologies our species has ever invented, maybe the most powerful, full stop. So no, this isn't a "AI is garbage" article.

This is a "the people selling it to you are, a lot of the time, outright charlatans" article. And that's a very different problem.

The problem isn't the tech. It's the snake oil salesmen.

Here's the uncomfortable truth. A huge share of the loudest voices promoting AI online have no earthly idea how it works or where it breaks. They are not builders. They are performers who found a hot keyword and a captive audience, and they are monetizing your confusion.

So let me hand you the one thing about AI that every single one of these gurus conveniently forgets to mention.

At the end of the day, a large language model is fancy autocomplete. That's it. It predicts the next word in a sentence, over and over, extremely well. That's the entire trick. And once that clicks, everything else does too, including why it will look you dead in the eye and tell you something completely false without blinking. It's a prediction machine. It has no built-in ability to say "I don't know," because "I don't know" is rarely the most probable next word. So it just keeps confidently predicting. That isn't a bug you can smooth-talk away in a webinar. That is what the thing fundamentally is. (I wrote about this same reality in AI Sucks on a Blank Page.)

Enter the charlatans

Now picture the archetype. The internet-marketing funnel crowd. The guys filming "my swarm of 300 AI agents runs my entire business while I sip espresso on a yacht" videos with a rented Lambo in the driveway.

These people are genuinely elite at exactly one thing: optimizing for views. I'll give them that, it's a real skill. But the overwhelming majority of them have never shipped a real SaaS product in their lives. They have never watched a system faceplant at 2am. They have never been on the hook when the thing they promised silently corrupts a client's data. They don't understand the technology's limitations because they have never once smacked into them. And from that throne of confident ignorance, they sell "solutions" that range from useless to borderline fraud.

And look, I sell. I'm a salesman. I learned it knocking on doors, I've written about why sales is the skill that beats systems thinking, and these days I genuinely enjoy it. Selling and marketing yourself is a legitimate, necessary part of business. But there is a canyon between selling and lying, and these snake oil peddlers pitched a tent right at the bottom of it. Their entire model preys on people who don't understand the underlying technology, people who've been trained to treat AI like the Oracle of Delphi, some all-knowing god you kneel before for answers. Guys. It predicts words. That's the whole magic act.

This is why 94% of AI initiatives fail

There's a reason the overwhelming majority of corporate AI initiatives are crashing and burning. It is not the technology.

It's that an executive got pitched on the golf course. Or by someone charming in an expensive vest. Or he watched a "300 agents" video at midnight and strolled into Monday convinced magic was on sale. These are decision-makers who, respectfully, never set foot in the day-to-day operations of their own company. They don't understand how the tech works, they have no systems thinking to catch the con, and they expect miracles. Reality does not care what you expect.

This ties straight into what I laid out in On Automation. You cannot automate, or "agentify," a process you haven't first done yourself and documented. You need a real, understood process before you hand it to a machine. And half the time an "agent" is flat-out the wrong tool, a single well-placed API call would do the job cheaper, faster, and without falling over. (Whole separate piece coming on when you actually want an agent versus when you're just being sold a buzzword.)

The example that actually stung

Here's the one that bothered me most, because it wasn't some faceless grifter. It was an entrepreneur I genuinely look up to. A guy worth somewhere north of a hundred million dollars, someone whose business acumen I have real respect for.

I watched him push a tool called OpenClaw to his enormous audience like it was Jarvis in a box, the second coming, the thing that changes everything.

The reality? Touch half the plugins in that ecosystem, and I'm being generous with "half," and you'd get your machine hacked. That was never in the pitch, obviously. And the second it stopped being the shiny new toy, he quietly buried it and sprinted to the next miracle cure. No correction, no "hey, careful with that," nothing.

That's the whole rot in one story. When even the people I respect will wave around a genuine security hazard because it's trending, you understand how deep the snake oil runs. Without actually understanding the underlying technology and its limits, you have no business promising anyone anything. And when you promise results you cannot deliver, let's stop being polite about it: that is a lie. It's just a lie with a nicer haircut and a better funnel.

What honest selling actually looks like

Here's the difference, and it's the entire ballgame. The reason I can sell effectively is that I never leave the ground. I say: here is precisely what this technology can do, here is exactly where it stops, and here are the concrete results real clients have actually seen. That's it. No smoke, no oracle, no yacht. When your pitch is built on what's real instead of hype and noise, you don't need the snake oil, and you sleep fine.

And no, before anyone asks: you are not getting the human out of the loop. Not now, and honestly I don't believe within my lifetime. That's the hill I die on, and it's the same one I planted a flag on in Judgment Is the Real Moat Now. Here's the short version, and maybe I'm going out on a limb. A prediction machine that only improves by predicting its own next word, with no outside observer to sanity-check whether any of it is true, does not quietly bootstrap itself into a god. It drifts. It's the same reason it hallucinates. A system with no external ground truth doesn't get wiser, it goes cancerous. You need a human holding the wheel. You always will.

The takeaway

So here's my blunt ask for this entire industry. We need to start saying, out loud and without flinching, which AI use cases are brilliant and which ones are just plain stupid. Not everything deserves an agent. Not every demo is real. Not every miracle is anything but a magic trick with your wallet as the volunteer. And not every confident guy with a ring light and a course to sell has the faintest clue what he's talking about, frankly, a few of them would make the Count of St. Germain look like an honest broker.

AI is the most exciting tool I have ever put my hands on. It deserves a hell of a lot better than being the headline act in a traveling medicine show. Learn how it actually works. Demand that anyone selling it to you has actually built something real. And treat everyone promising you magic exactly like what they are: a salesman betting his commission on you not knowing better.

Want AI from someone who actually builds it?

No magic, no oracle, no swarm of 300 imaginary agents. Real systems, grounded in what the technology can genuinely do. That's what we do at HyppoAI.

Build it right