“My AI Skeptic Friends Are All Nuts” — The Fault Line Running Through 2,826 Comments

In June 2025, a legend of the security industry lobbed a grenade onto Hacker News: “My AI skeptic friends are all nuts.” 2,356 points, 2,826 comments. The developer community split in two. And the fault line is still widening.

1. The Essay That Shook Hacker News

On June 2, 2025, a piece appeared on the Fly.io blog. The title: “My AI Skeptic Friends Are All Nuts.”

What made this essay fundamentally different from the usual “AI praise” was the author.

Thomas Ptacek (@tqbf). Co-founder of Matasano Security, one of America’s largest software security firms. He later founded Latacora, and now works at Fly.io. A thirty-year veteran who has been building software in C, C++, Ruby, Python, Go, and Rust since the mid-1990s. Black Hat speaker, and one of the most respected members of the Hacker News community.

Here’s the key: he is not an “AI bro.” He doesn’t run an AI startup, doesn’t sell AI products. By the nature of his profession as a security expert, he is instinctively skeptical. That is the person who declared “AI skeptics are nuts.”

Within two hours of posting, the essay had 695 comments. Final tally: 2,356 points and 2,826 comments. One of the hottest technical debates in Hacker News history.

2. What the Essay Argues

Ptacek’s piece is provocative but systematic. He develops more than ten distinct lines of argument, dismantling AI skeptic positions one by one, expletives included. The core claims, distilled:

“LLMs Are the Second Most Important Thing in My Career”

“All progress on LLMs could halt today, and LLMs would remain the 2nd most important thing to happen over the course of my career.”

Coming from someone with a thirty-year career, this declaration carries weight. He doesn’t say what the first was, but the prevailing guess is the arrival of the internet.

”What You’re Criticizing Is 2023 AI”

“If you’re making requests on a ChatGPT page…you’re not doing what the AI boosters are doing.”

Ptacek’s sharpest jab. Many skeptics are criticizing the practice of copy-pasting questions into the ChatGPT webpage — a 2023 mode of use. AI coding in 2025 is agents autonomously navigating a codebase, executing tools, compiling, testing, and fixing errors. It is, he argues, an entirely different experience.

”Hallucination Is a Solved Problem”

“If hallucination matters to you, your programming language has let you down.”

“If hallucination is a problem for you, that’s your programming language failing.” LLM agents run linters, compilers, and tests themselves. When they generate broken code, they receive the error feedback and self-correct. It works especially well in languages with strong type systems, such as Go. (He concedes that Rust is still difficult.)

”A $20-a-Month Intern”

“Does an intern cost $20/month? Because that’s what Cursor.ai costs.”

Cursor is $20 a month. Compare that to the cost of hiring one intern. If a conversation goes off the rails, open a fresh session. The intern doesn’t take offense when you give feedback, doesn’t get tired, doesn’t complain about repetitive work.

”We Are Problem Solvers, Not Artisans”

“Professional software developers are in the business of solving practical problems for people with code. We are not…artisans.”

Ptacek says he has a woodworking shop in his basement. But he doesn’t build his office desk by hand — he buys it. The hobby of woodworking and the practicality of buying furniture do not conflict. The same applies to coding. You can acknowledge the aesthetic value of coding while accepting that in a professional context, using tools is the rational move.

”We’re in the Business of Automating Other People’s Jobs”

“We’re a field premised on automating other people’s jobs away.”

His most uncomfortable punch. Software developers are, by definition, in the business of automating other people’s work. Accountants, bank tellers, switchboard operators, logistics dispatchers — the software we built has displaced countless jobs. For us, of all people, to be outraged at the possibility that “AI might replace developers” is, he argues, hypocrisy.

3. Comment War — 2,826 Voices

The Hacker News comment thread became a battlefield in its own right. The sides were evenly matched, and both produced compelling arguments.

The Pro Camp — “I Tried It and It Actually Works”

The most upvoted comment was from matthewsinclair. He confessed that he had initially been skeptical of AI, but his mind changed once he started using Claude Code. His metaphor stuck:

LLM coding is like a “mech suit.” Powerful, but it needs decisive piloting.

Not self-driving, but a power suit. Give an agent total autonomy and it fails; pair it with clear guidance, tests, and documentation, and productivity explodes.

kentonv argued that the learning curve for Claude Code is “essentially zero,” and criticized detractors for judging the tool theoretically without actually using it.

Many comments cited historical precedent. When calculators arrived, people said math education was over. When search engines arrived, people said memory would atrophy. When high-level programming languages arrived, there was resistance: “real programming is assembly.” They were wrong every time.

The Con Camp — “Have You Forgotten the Hype Cycle?”

wpietri offered the most systematic rebuttal. The rational move, he argued, is to wait until Gartner’s hype cycle reaches the “plateau of productivity.” During crypto we had the same social pressure — “you’ll be left behind if you don’t use it” — and we all know how that ended.

raxxorraxor shared on-the-ground experience. The code LLMs produce, especially early on, is “shockingly bad.” Useful for boilerplate, documentation, and learning a new language, but inadequate for the core problem-solving work.

chinchilla2020 had a sharp observation. LLMs are excellent at replacing Google search, but the code they generate is often “convincing but wrong.” In an engineering context, “convincingly wrong” code is more dangerous than straightforwardly wrong code.

The Most Uncomfortable Moment — The Author’s Confession

The most uncomfortable moment in the debate was a confession Ptacek himself made in a follow-up HN discussion.

“I forgot how to write table tests because I generate all that. And it’s frightening.”

The most fervent AI advocate had admitted that AI was degrading his own skills. That single sentence was more powerful than any counterargument.

Systemic Concerns

soraminazuki raised a systemic problem beyond individual choice. Management will push AI for cost reduction, regardless of the quality of the code AI produces. Even if individual developers conclude “AI isn’t there yet,” organizational decisions move by a different logic.

4. The Counterattack — “Contra That Terrible Article”

Ptacek’s piece spilled beyond the HN thread into a series of rebuttal essays.

Ludicity — “Contra Ptacek’s Terrible Article On AI”

The most direct rebuttal came from the tech blog Ludicity. The title alone is aggressive: “Contra Ptacek’s Terrible Article On AI.”

The core criticism: people had been making the same arguments as Ptacek months earlier in June 2025. Nobody paid attention then. The reason Ptacek’s piece detonated, the author argues, was not the quality of the argument but the author’s reputation and the provocative title. The “your friends are nuts” framing, the critic argued, did not promote debate — it made debate hostile.

Skarlso — The Death of the Junior Pipeline

Developer Skarlso’s rebuttal took dead aim at the biggest problem Ptacek didn’t address: the cultivation of junior developers.

LLMs “reason with absolute conviction.” A senior developer can tell whether that conviction is hallucination. A beginner cannot. And when a beginner leans on AI to skip the learning process, they will never become a senior developer. If you don’t grow juniors, you will eventually have no seniors.

This was the biggest blind spot in Ptacek’s essay, and the most repeated concern in the comments.

Niko Heikkila — “My AI Agents Are All Nuts”

The Finnish developer Niko Heikkila responded with parody. The title: “My AI Agents Are All Nuts.” An elegant inversion of Ptacek’s love letter to agents.

The Earlier Debate — Casey Newton vs Ed Zitron

This debate had a precursor. In December 2024, Platformer’s Casey Newton published “The Phony Comforts of AI Skepticism,” arguing that AI skeptics had become immune to evidence. Ed Zitron flipped exactly the same frame back with “The Phony Comforts of AI Optimism.” Ptacek’s essay poured gasoline on this preexisting fire.

5. What the Numbers Say — Inconvenient for Both Sides

When you look at this debate through numbers, both sides get uncomfortable.

Numbers That Trouble the Optimists

IndicatorFigureSource
AI project failure rate70-85%MIT / RAND
Enterprises reporting “zero value” from AI95%MIT Media Lab / NANDA
Agentic AI projects to be scrapped by 202740%+Gartner forecast
Employment impact three years after AI adoption90%+ of firms “no change”NBER (Feb 2026)

AI investment in 2025 hit $202.3B (~280 trillion won). And 95% of enterprises reported “no value.” If those numbers are accurate, this is one of the largest corporate investment failures in history.

Numbers That Trouble the Skeptics

IndicatorFigureSource
Enterprises that have adopted AI78%2025
Developers using AI tools regularly85%End of 2025
Productivity ROI from AI$3.70 per dollar investedFirms reporting success
2025 AI investment growth75% YoYYear over year

It is hard to dismiss as a “bubble” a technology that 78% of enterprises have adopted, that 85% of developers use, and that is attracting 75% YoY investment growth. Even if most projects fail, the value created by the small minority that succeed could justify the aggregate investment.

Gary Marcus called 2025 “peak bubble,” while Marc Andreessen called AI “the most powerful technology in history.” MIT Technology Review dubbed 2025 “the great AI hype correction.”

The truth is probably somewhere in between.

6. Things to Think About — So What Do We Do?

The most striking thing about this debate was Ptacek’s real intent, as revealed by Simon Willison (Django co-creator).

Ptacek’s goal was not to persuade the skeptics. It was to break “a static, unproductive equilibrium — the equilibrium of uninformed argument about how this technology actually works.”

That is the real value of this essay. Not the binary of “AI is good / AI is bad,” but a demand for informed argument.

On Criticizing Without Trying

Ptacek’s most valid criticism is that “people who have never used an agent are criticizing agents.” Judging Claude Code in 2025 by your 2023 ChatGPT experience is like judging the 2010 smartphone by your 2007 feature phone. The technology made a qualitative leap in between.

The Question No One Can Answer

But there is a question the AI optimists cannot answer either. The junior developer pipeline.

If AI replaces boilerplate coding, what do junior developers learn from? One HN comment described a Danish school where teachers required LLM use while grading with LLMs — the learning process itself was bypassed. Even if a surgeon delegates a procedure to AI, the process of learning to operate is still required. Isn’t coding the same?

The Conclusion: Skepticism Isn’t the Problem; Uninformed Skepticism Is

Here is the conclusion I reach after reading this.

Skepticism about AI is itself healthy. A critical posture toward technology prevents bubbles, surfaces risks early, and promotes ethical use. The problem is uninformed skepticism. Judging 2025 agents through the 2021 frame of the “stochastic parrot.” Evaluating Claude Code’s autonomous coding through copy-paste ChatGPT experience. Claiming to know without ever trying.

The same applies to optimism. The declaration that “AI will change everything” must be humbled by a 70-85% project failure rate. Even Ptacek confessed that he had forgotten how to write table tests.

What the 2,826 comments show is that the developer community has not yet found an answer to this question. And that is okay. What matters in the process of finding an answer is, at minimum, honestly trying, and honestly evaluating. That, beneath the provocative title, is what Ptacek was really asking for.

References