You’ve probably felt it. That sinking feeling when your team rolls out another “AI-powered” chatbot and it… doesn’t work. The answers are wrong. The workflow breaks. The customer gets frustrated. And someone says, “We just need better data.”
You’re not crazy. The problem isn’t you. It’s the system.
The real crisis in enterprise AI isn’t technical—it’s organizational. A chatbot cannot fix a company that can’t define how it works.
This is the uncomfortable truth that Silicon Valley’s hype machine will never tell you. The people selling AI have every reason to oversell it. The people buying it have every reason to believe the hype. And the ones who actually have to use it? They’re the ones who see the emperor has no clothes.
In a recent deep dive, Cory Doctorow and Nikhil Suresh laid out exactly why this happens. The incentives to lie about AI are built into the system. Vendors need to sell subscriptions. Executives need to show innovation. And the media needs clicks. Everyone profits from the myth that AI is magic—except the poor schmuck trying to get a refund from a chatbot that can’t even understand the request.
The only AI that works is the one that makes a skilled person more skilled. Everything else is theater.
Think about the real success stories. They’re not about replacing humans. They’re about a developer using an AI autocomplete on steroids, a radiologist using a model to highlight anomalies, a writer using a tool to rewrite a clumsy sentence. These are “centaur” workflows—human and machine together, each doing what they do best. The AI doesn’t take over the job. It amplifies the person.
But most enterprise AI projects don’t start with that mindset. They start with a press release and a mandate to “cut costs.” And they fail—spectacularly. According to the research, the success rate for enterprise AI initiatives is close to zero. Not because the technology is bad, but because the organizations are broken.
Here’s the twist: Before you buy a chatbot, try writing down how your company actually works. If you can’t do that, no AI will save you.
Most companies don’t have good internal documentation. They don’t have clear, consistent processes. They don’t have a single source of truth for how things get done. So when they bolt a chatbot on top of that chaos, the chatbot just reflects the chaos. It doesn’t fix it. It just makes the chaos faster and more visible.
The most honest thing anyone in AI can say is: “This is a really good autocomplete.” But that doesn’t sell keynote tickets. That doesn’t justify a billion-dollar valuation. So instead, we get promises of revolution, transformation, and the end of work as we know it.
Don’t buy it. The companies that will win with AI are the ones that already have their shit together. They have clear processes, good documentation, and skilled people who know how to use new tools. The rest will blame the technology and keep buying the next shiny thing.
Stop chasing magic. Start fixing your processes. Then, and only then, let AI make you faster.
FAQ
Q: Isn't AI actually transforming some industries?
A: Yes, but in narrow, specific ways. AI is a powerful tool, not a magic wand. The transformative claims are real only when paired with skilled operators and solid organizational processes. The hype is dangerous because it creates unrealistic expectations that lead to expensive failures.
Q: What should I do instead of buying an AI chatbot?
A: First, document your internal processes. Define clear workflows, responsibilities, and decision trees. Train your people to use AI as an augmentation tool, not a replacement. Then, pilot a narrow use case with a specific, measurable goal. The companies that succeed with AI are the ones that already have their operational house in order.
Q: Isn't AI just a bubble?
A: No, the core technology is real and valuable. The bubble is in the hype. The same forces that drive legitimate adoption also fuel unrealistic expectations. The crash will come for overhyped vendors and poorly executed projects, not for the underlying capability. The challenge is separating signal from noise.