The AI Dark Age: How the Industry That's Reshaping the World Can't Seem to Read the Room

Let me start with a confession that probably won't surprise you by now: I keep finding the Dunning-Kruger Dark Age everywhere I look. It's like learning a new word and then hearing it six times before lunch. Once you name a pattern, it refuses to stay quiet.

But I genuinely did not expect the cleanest example of it to be the entire AI industry — the very people building the future, the ones who keep telling us they're the smartest folks in any room they walk into.

So here's the post. Buckle up. This one's going to sting a little, and I say that as someone who uses these tools every single day and broadly likes them.

The Room Nobody Is Reading

Let's start with the audience, because the audience is screaming.

A March 2026 Gallup survey found that 71% of Americans oppose building an AI data center in their local area — and nearly half, 48%, are strongly opposed. For perspective, that's more opposition than Americans have to building a nuclear power plant nearby. People would rather host fission than your GPU farm. Sit with that for a second.

And it's not just the data centers. YouGov data shows most Americans use AI but still don't trust it, with skepticism running especially high in sensitive areas like finance and healthcare — exactly the places the industry keeps insisting AI will revolutionize first.

So the public has spoken, fairly clearly: We're nervous, we don't fully trust this, and please not in our backyard.

The industry's response? Build faster. Ship harder. Put "AI" in the name of everything, including, I assume, the breakroom microwave.

Now, skepticism alone doesn't make something a bad idea. The public has been wrong before. But here's where it gets interesting — because the industry's own data is starting to agree with the skeptics.

The ROI That Keeps Not Showing Up

If the public's distrust were the only signal, you could wave it off as fear of the new. Plenty of good technology arrived to a chorus of boos.

The harder problem is that the money isn't materializing either.

A Gartner study covered by Fortune surveyed 350 global executives at billion-dollar companies. The finding: while 80% of those piloting AI reported workforce reductions, there was no correlation between those layoffs and higher ROI. Let me translate that out of consultant-speak. Companies cut people, blamed AI, and got... roughly nothing extra for it. The layoffs happened regardless of whether the technology actually delivered.

As the Gartner analyst put it, "Chasing value only through headcount reduction is likely to lead most organizations down a path of limited returns."

Then there's the quiet part, which Sam Altman said out loud: a chunk of these cuts are just "AI washing." Companies blaming AI for layoffs they wanted to do anyway, because "the robots made us do it" sounds better on an earnings call than "we over-hired and panicked." It's a fig leaf with a neural network printed on it.

So we have a public that's deeply skeptical and an ROI story that, by the industry's own measurement, mostly isn't there. You'd think that combination might prompt a pause.

It hasn't. And the reason why is where this gets genuinely strange.

The Tech Debt Paradox

Here's the justification I hear most often inside companies, usually in a hushed, slightly embarrassed tone: "Yeah, the public numbers are rough, but AI is going to let us pay down our tech debt."

It's a seductive story. We've all got mountains of crusty legacy code nobody wants to touch. The dream is that an army of tireless agents finally cleans up the mess we've been ignoring since 2014.

Except the evidence is pointing the other way. AI isn't buying down tech debt. It's opening a new line of credit.

JetBrains has a name for the new variant: Shadow Tech Debt — low-quality, architecture-blind code generated by agents that have no structural understanding of the projects they're modifying. The agent completes the task and moves on. It doesn't read the architecture decision records. It doesn't know why a weird pattern was chosen three years ago. It just... ships. And quietly undermines the coherence of the whole codebase while everyone celebrates the velocity.

Meanwhile, the team at Port mapped out seven blocks of hidden infrastructure debt surrounding AI agents in enterprise systems — integrations, context management, agent registries, measurement, human-in-the-loop, governance, orchestration. Their core point is brutal and funny at once: building an agent is easy, but the agent is the smallest part of the system. Everything around it is where the complexity — and the debt — actually lives. They estimate that at a certain scale, half a team's capacity goes to building the surrounding infrastructure nobody planned for.

So the tool we adopted to reduce our maintenance burden is generating code we can't fully trust, wrapped in infrastructure we haven't built, accumulating debt we can't yet see. We didn't pay off the credit card. We opened three new ones and lost the statements.

Here's the question that should keep leaders up at night: the people running these companies can see all of this. The surveys, the ROI gaps, the debt research — it's not hidden. It's in Fortune and Gallup and on the front page of every engineering publication. So why does the machine keep accelerating?

Running the Dark Age Diagnostic

You know where I'm going with this.

In my earlier posts, I argued that the Dunning-Kruger Dark Age shows up at the organizational level as a very specific failure: high ability, low motivation. A company perfectly capable of seeing the truth, profoundly unmotivated to act on it.

Bring in BJ Fogg, who tells us behavior requires motivation, ability, and a prompt all arriving together. Then point that lens at the AI industry and watch it light up like a server rack.

Ability? Off the charts. These are the most data-rich organizations on the planet. They have the surveys. They commissioned the ROI studies. They employ the engineers writing the Shadow Tech Debt warnings. If anyone on Earth can see the gap between the AI narrative and the AI reality, it's them. Ability is emphatically not the problem.

Motivation to act on what they see? Functionally zero — or worse, pointed backward.

Because the prompt in Fogg's model isn't "what's good for users." The prompt is investor pressure. It's the competitor who just announced their own agent. It's the terror of being the one CEO who didn't mention AI on the call and watched the stock dip. The motivation that should be aimed at "build the right thing" got hijacked by "protect the narrative."

Look at OpenAI, the supposed crown jewel — reportedly spending far more than it earns, revenue and costs upside down, and yet treated as the inevitable winner. That's not demonstrated success. That's declared success. And declaring success instead of demonstrating it is the single loudest alarm bell in any healthy operating model.

This is the Dark Age in its purest organizational form. The truth is sitting right there on the table, fully legible. The industry has simply decided it doesn't get a vote.

This Is Textbook Drift

In the Health-Driven Operating Model I keep coming back to, organizations move along a trajectory — from Aligned (signals shape decisions) to Drift (signals are gathered but selectively applied) to Detached (signals are overridden entirely, and success is declared rather than demonstrated).

The AI industry is somewhere between Drift and Detached, and it speedran the journey.

Look at the symptoms of Drift and tell me they don't fit. Signals are gathered yet selectively applied. We have the public opinion data — we just don't let it touch the roadmap. The roadmap feels busy but not directional. Has there ever been a busier roadmap than "ship AI into every product simultaneously"? It's frantic. It's also not pointed anywhere in particular except away from the question of whether any of it works.

That's the tell. Not silence — noise. A Dark Age isn't a lack of activity. It's a surplus of activity disconnected from the signals that should be steering it. Everyone's sprinting. Nobody's checking the compass, because the compass keeps saying things the quarterly narrative can't afford to hear.

What I Actually Want From You

I'm not here to tell you AI is a fraud. I use it constantly. There's real value buried in here, and the "people amplification" use cases — making humans better rather than replacing them — are where the actual ROI keeps showing up in the data. The technology isn't the villain. The relationship to evidence is.

So here's the move, and it's the same one I keep coming back to because it's the only one that works.

Name the pattern. Out loud, in the room, when someone proposes shipping AI into something because "we have to." Call it what it is: a motivation problem wearing an innovation costume.

Run the Fogg diagnostic on your own org. Ask the uncomfortable question: Are we incapable of seeing whether this AI investment helps anyone — or are we just unwilling to find out, because the answer might be "no" and the board doesn't want to hear "no"? Ability problems you fix with resources. Motivation problems you fix with courage, incentives, and occasionally telling a powerful person something they don't want to hear.

Demand demonstrated success over declared success. If your AI initiative can't show its work — real ROI, real adoption, real reduction in toil rather than a fresh pile of Shadow Tech Debt — then you don't have a strategy. You have a press release with a budget.

And then the only question that matters, the one I'll leave you with:

Are you building this because the evidence says so — or because the narrative does?

Be honest. You already know which one is driving most of what's getting shipped right now. The first step out of any Dark Age is admitting you could read the room the whole time. You just decided the room didn't get a vote.

The magic was never in pretending the skepticism, the missing ROI, and the mounting debt aren't real. The magic is building something that works anyway — and being brave enough to check.