Scene 1: Why

Let's kick this off with one of my favorite genres: horror.

In The Exorcist, there's a scene that has nothing to do with fear and everything to do with knowledge. Father Karras, the young Jesuit priest played by Jason Miller, has brought his veteran colleague Father Merrin, played by Max von Sydow, to a house on Prospect Street in Georgetown where 12-year-old Regan is, well, going through it. The priests are downstairs gearing up when Karras asks Merrin:

"Do you want to hear the background of the case first, Father?"

Merrin replies:

"Why?"

I've always remembered that exchange. Lately I've been thinking about it a lot.

Merrin's "Why?" is not a dismissal. It's the most succinct expression of expertise from lived experience I can think of. He's not incurious, he just doesn't need to be briefed. His expertise is internalized. He knows.

That kind of knowing is becoming harder to find, and not because experts are disappearing. The conditions that recognized, rewarded, and produced them are changing. Quietly, and from two directions at once.

If you're looking for a scholarly article on AI or American politics, this is not that. But stick around.

Scene 2: What's the deal?

I came across the philosopher Michael Polanyi who had a useful answer. "We can know more than we can tell," he wrote, a phenomenon he called "tacit knowledge": understanding that comes from accumulated experience and can't be fully captured in words. The doctor who intuits the cause before seeing the CT scan. The sommelier who identifies the grape by smell alone.

(Side note: Polanyi made an educational film in 1940 called Unemployment and Money: The Principles Involved. Too bad he's not around for AI.)

Tacit knowledge is what Merrin was expressing. He had seen so many versions of the room upstairs, he already knew what he was dealing with. The background of the case wouldn't change his approach, because his approach isn't assembled on the fly from available information. It's the result of everything he's already been through.

There's something else operating in the background of genuine expertise. A true expert is more likely to say "I don't know" than someone who is performing expertise. Not as an admission of failure, but as an understanding of their own range. The expert who says "I don't know" has mapped what they know and where it ends. That's what distinguishes true expertise from its impersonations.

Scene 3: Confidently wrong

An impersonation of expertise usually starts with data, adds a convincing facade, and presents itself as the real thing.

Marketing has been doing this for years in the form of fictional characters called personas:

"Jennifer, 38, is a small business owner who values efficiency and work-life balance. She knows what she needs, but not how to get there. She has done some initial research and…"

(You get the point.)

Jennifer was built from research data. Jennifer does not exist, she's a composite. But organizations make product decisions based on what Jennifer wants, because she is the best approximation they have, and the idiosyncrasies of real customers are very difficult to capture on one slide.

AI creates a similar illusion. It is the aggregation of past patterns presented as present understanding. It looks like the real thing because it was built from examples of the real thing, the same way Jennifer looks like a customer because she was built from actual customer data.

There's something related worth naming. In 2003, two psychologists, Daniel Kahneman and Gary Klein, spent six years trying to determine the reliability of expertise-based intuition. Kahneman was skeptical of it, Klein was a believer. But rather than argue from a theoretical place, they sat down together in what they called an "adversarial collaboration" to put their beliefs into practice.

They figured out that trusting expert intuition depends entirely on the environment. They described "kind" and "wicked" environments (so good). In kind environments, which are stable, predictable, and precedented, intuition based on expertise works. The past reliably predicts the present, and trusting your judgment is a safe bet. In wicked environments, reliance on intuition can lead to something Kahneman and his colleagues had identified years earlier: the illusion of validity. The expert feels certain, but the certainty isn't justified. In their words:

"[People] express great confidence in the prediction that a person is a librarian when given a description of his personality which matches the stereotype of librarians, even if the description is scanty, unreliable, or outdated. The unwarranted confidence which is produced by a good fit between the predicted outcome and the input information may be called the illusion of validity."

That's also a surprisingly accurate description of how AI can look from the outside. It doesn't always know when its patterns are no longer reliable. It keeps being Jennifer even when the forces that shaped Jennifer have totally changed.

Scene 4: Imposter syndrome

Here's where impersonation starts having consequences.

A recruiter once told me about a client who accidentally left the name of the company's internal app in a job posting. It was built in-house and used only internally, no one outside the company could have experience with it. But every AI-generated resume they received listed proficiency with the app. Every single one.

AI didn't get anything wrong exactly, it used available data to connect the dots. It did what it was built to do, and inferred knowledge it didn't possess. Like faking a book report from the table of contents.

Before AI, the gap between expertise and non-expertise was more visible. A resume contained only its author's experience, not the demands of a future role. A bad piece of writing looked like a bad piece of writing and got caught. AI hasn't removed the need for expertise. It just makes it difficult to know when expertise wasn't involved.

Here's the math that most people don't do: the upfront work required to get AI output right often takes longer than putting a competent expert in the room with a clear directive. Unlike a human expert, AI on its own doesn't accumulate judgment — the tacit knowledge from earlier. It can reference previous work, but it can't build on it the way a person does.

Scene 5: Good enough

My experience as a freelancer working inside different companies gives me a front-row seat to what's actually happening.

A writer on a client team recently said this on a shared Slack channel:

"My marketing partners consistently use AI before sharing with working teams. I keep asking folks to involve us early, I've shared our process documentation, I've set up meetings to talk about process. I keep asking to get involved early but nothing seems to work."

Other writers responded immediately, picking apart the AI-generated work, noting all the usual red flags and brand violations. The verdict at the end of the thread: "Speed seems to be the constant issue."

The people who knew what could go wrong could see exactly what went wrong. But the work was already approved. When the cost of a shortcut is invisible, speed wins every time.

This is what the disappearance of expertise looks like from the inside. Not a dramatic announcement. Not a policy change. Just a Slack thread where the people who know what right looks like are the last to find out.

There's something else worth mentioning here. AI doesn't just sideline expertise, it changes who gets assigned the work in the first place. That decision is made based on the assumption that AI in anyone's hands can create something credible enough to pass. And when mostly right becomes the standard, it tends to stay the standard. Nobody planned it that way. It arrived as a byproduct of the efficiency frenzy.

Scene 6: Deletions

I bet you've been waiting impatiently (or with dread) for the political tie-in. So far this has been a story about AI. But there's another force at work simultaneously, doing something similar by different means.

In 2025, the Trump administration removed or modified more than 8,000 web pages and approximately 3,000 datasets across federal agencies. More than 3,000 pages were altered or taken down from the CDC's website. The phrase "climate change" was scrubbed from the EPA's homepage. Dozens of national park signs and exhibits were changed or removed.

Instead of arguing with expertise, the Trump administration just deletes it. Remember what I said earlier about AI? It gives no overt signal that expertise is missing, unless you're the expert scanning for it. Add those two phenomena together, and the result is a single disappearing act performed from opposite ends of the stage.

The erasure isn't identical in both cases. One is structural and indifferent, the other is deliberate and malicious. But the outcome is roughly the same: whoever or whatever held the truth is gone, left posting on Slack, made irrelevant, or missing entirely. The narrative that remains, whether incomplete or an approximation, becomes the source of truth.

And here's where it gets darker. Future AI systems will increasingly learn from what's left after the deletions. They can't distinguish a manipulated source from an accurate one. It reproduces what it finds with equal confidence regardless of origin. Political agenda removes the truth. AI learns from what's left.

Scene 7: Ghosts in the machine

Entry-level roles are already disappearing. Without the traditional first step in acquiring job mastery, it's hard to know where future expertise will come from. A friend asked me recently: who will be the senior people in ten or twenty years? I didn't have an answer. 

The funnel that produces senior-level judgment no longer has a clear entry point. Junior work is being handled by AI, and more senior people are inheriting a pile of AI tasks with no change to their job titles. The conditions that once produced someone like Merrin — the years of work, the early failures, the slow accumulation of judgment — are being quietly removed. 

Already, job postings ask for candidates who can write, prompt, strategize, and validate AI output simultaneously, at a mid-level salary. Maybe in ten or twenty years "senior" will mean something different entirely. It will describe the person with the most sophisticated relationship with AI, rather than the most wisdom. Expertise won't disappear overnight. It will just get harder to spot.

AI is a remarkable tool. (It proofread this essay.) It can synthesize, summarize, generate, and accelerate in ways that are genuinely useful. But a tool is what it is. It doesn't develop judgment. It doesn't accumulate the kind of knowledge that can tell you when to stop trusting it. The people best positioned to use it are the ones who were doing the work before it showed up, because they know what good looks like — and when AI is getting it wrong.

Cut back to Prospect Street, where Karras is creating a prompt that extracts what Merrin knew about exorcisms. The guidance sounds credible, but his mentor has disappeared, along with the wisdom that made AI worthwhile in the first place.

——————

Sources:

https://en.wikipedia.org/wiki/2025_United_States_government_online_resource_removals

https://www.levernews.com/trumps-epa-just-deleted-climate-change/

https://www.cnn.com/2026/06/13/politics/judge-ruling-national-park-sign-change