Ed Zitron is the Mainstream Media’s Perfect Skeptic

April 2025. On Hacker News, a developer shared a simple observation: 75% of engineers in his survey said they actively use AI coding tools. Cursor. GitHub Copilot. Claude. Something.

He had receipts. 216 developers, self-selected but honest, mostly shipping code for a living. This wasn’t a think tank survey or a VC-funded report. It was engineers talking about their actual day.

Ed Zitron—the AI skeptic with a popular subreddit and Substack, the guy the mainstream press calls when they need a confident voice saying AI is oversold—did not engage with the data. He mocked the sample size. Blocked the guy. Moved on.

This moment is not about Zitron. It’s about what Zitron represents: the mainstream press doesn’t need someone who understands AI. They need someone who sounds credible while dismissing it. And for that job, Zitron is perfect. He has numbers. He has followers. He has the confidence of someone who has never had to ship anything that actually works.

The press outsourced their skepticism to a performer, and now they get to feel like they’re covering both sides of the story without actually learning anything.

The New York Times does not use AI in production. The Washington Post does not have a team of engineers shipping AI features daily. The Wall Street Journal runs it through PRs and legal, not through actual product development.

So when AI became a story they had to cover, they faced a problem: how do we write about this in a way that doesn’t require us to actually understand how it works?

Enter the confident British finance guy with numbered arguments.

Zitron came pre-loaded with exactly what legacy media needed. He spoke their language. He cited “data.” He quoted numbers—user acquisition, revenue multiples, market cap—as if those numbers meant something they didn’t actually mean. He sounded like the kind of person who reads 30-year-old business school frameworks and mistakes them for insight.

Most crucially: he said AI is not working. It’s a bubble. The companies building it are dysfunctional. This was comforting to an industry that cannot compete with the speed at which AI is reshaping what readers want and how readers consume information.

Here’s what happened inside newsrooms: an editor needs a story saying “AI skepticism is serious.” They search for someone who sounds authoritative. They find Zitron. He has a subreddit. He has thousands of followers. He cites numbers.

Nobody checked whether those numbers actually said what he claimed. Nobody asked: “Does this guy actually build products? Does he use these tools daily? Has he shipped anything that had to work at scale?”

The answer is no. He’s a consultant and a columnist. His incentive is engagement. His business model is Twitter followers and Substack subscribers. Every piece he writes that says “AI is doomed” performs better than a nuanced take. Every unhinged claim about Google or Meta “desperately inserting AI everywhere” gets more retweets than “here’s what’s actually happening.”

This is not a conspiracy. This is just how algorithmic engagement works. And the mainstream press, completely divorced from product reality, had no frame of reference to recognize it.

I’ve been shipping products for 25 years. Buzzmedia, back in 2004, reached 102 million people a month across 45 properties. We had to actually understand how people used things. We didn’t have a subreddit of followers; we had data. We had to be right or we went broke.

When you’ve shipped at scale, you stop treating numbers like predictions. You stop citing data as if it proves something it doesn’t. You get precise about what you actually observe.

When I look at AI adoption right now, not the hype cycle but the actual daily adoption, here’s what I see:

Among engineers, adoption is approaching 80%+ for AI coding tools. This is not a survey finding. This is what I observe talking to founders weekly. Cursor, Copilot, Claude Code. These are not novelty tools anymore. They’re baseline infrastructure.

Among operators, AI is being integrated into everything. Payroll, customer service, content systems, data pipelines. The tools are imperfect. Some fail. Many save 10-20 hours a week per person. That math is powerful.

Among companies, the ones that figured out how to actually use LLMs in their product roadmap are shipping features that work. The ones that pretended it was a fad or forced it into places it didn’t belong are the ones losing ground.

This is not worldwide adoption. It’s not “AI solved everything.” It’s messy, uneven, and plenty of things people said would work with AI don’t. But it’s real. It’s happening. And it’s happening fastest in the places where people actually have to ship or die.

Zitron doesn’t see this because he’s not in those rooms. He quotes user acquisition curves and revenue guidance. He compares quarterly earnings to predictions he made. He’s running a financial analysis on data that doesn’t support a coherent argument—which is exactly what mainstream business journalism taught him to do.

But product adoption isn’t finance. It’s ethnography. It’s “what are 100 engineering teams actually doing on Monday?” And that developer had the receipts. Zitron had the block button.

If you’re reading Zitron and thinking he’s credible because he quotes numbers, you’re falling for the same mistake the mainstream press made: confusing confidence with knowledge, and data with understanding.

Here’s the test: Does the skeptic actually use the thing he’s skeptical about?

Zitron uses generative AI the way a food critic uses drive-thru burgers—occasionally, to confirm his priors, not to actually understand what’s possible. He’s an analyst commenting from the sidelines, not a builder.

The skepticism worth listening to comes from people shipping. People who tried something, it failed, and they learned why. People who use these tools every day and can tell you exactly what breaks. People who have skin in the game and are still skeptical.

Those people exist. They’re just not the ones getting 50K likes on X by saying “Google’s doomed” or “Meta is faking their AI gains.”

The mainstream press chose Zitron because he required them to not actually learn anything. They could quote him, feel like they’d covered both sides of the debate, and go back to writing about Facebook’s user numbers as if that data meant something about product reality.

But that developer and 216 engineers had moved on. They were already using this stuff. They didn’t need permission. They didn’t need a journalist to tell them whether it was real.

What makes this infuriating is not that Zitron is wrong. It’s that his wrongness is profitable. His business model is built on being confidently, elaborately wrong in a way that makes newsrooms feel smart.

And that means smarter skeptics—the kind who actually know what they’re talking about—get crowded out. The signal degrades. Founders stop trusting press coverage of AI because it’s either blind cheerleading or performative doom, with almost nothing in between that reflects actual product reality.

The people who should be reading serious skepticism about AI—founders, operators, people building this stuff—end up ignoring it all because they can’t afford to spend an hour parsing whether a number means what someone claims it means.

That’s the real damage. Not that Zitron made bad predictions. It’s that he gave the mainstream press permission to stop trying.

Stop looking for one person to tell you whether AI is real. Talk to five engineers actually shipping features. Ask them what breaks. Ask them what they couldn’t do six months ago that they’re doing now. Ask them what they still can’t do.

Read the people building this stuff. Not the people commenting on it.

Read the subreddits where engineers are solving problems, not the Substacks where someone’s polishing a narrative.

Read the companies shipping. Not the analysts analyzing.

Most importantly: if someone’s telling you a complex technology is either definitely going to reshape the world or definitely won’t—and they’ve never actually had to build with it—you already know more than they do. You just need the confidence to act like it.

Zitron will keep getting quoted. The mainstream press will keep outsourcing their AI skepticism to performers.

But here’s what actually happens: that developer and those 216 engineers already moved on. They’re not waiting for a think piece that tells them whether they’re allowed to use these tools. They’re in the shipping phase. By the time the debate over whether AI works reaches peak volume in the mainstream press, the people who matter have already answered the question through code.

The only people left arguing are the ones who never had to build anything in the first place.

---

This essay was inspired by Dan Luu’s analysis of Ed Zitron’s prediction accuracy. The April 2025 anecdote about AI tool adoption came from the Hacker News discussion of that article.