AI Quirks

Someone at Anthropic asked a simple question: is our AI taking people’s jobs?Reasonable question. Anthropic builds Claude. Claude is used by millions of people for work. Work involves jobs. The question connects.So they built a study.

Anthropic, the company that makes Claude — the AI currently helping write this article about Anthropic — recently published their AI Fluency Index. It measures how well humans are learning to use AI.The AI graded the humans.

You’ve got data. Twelve months of verified performance showing your strategy works. The math checks out, the execution is tested, the results are clear.So you consult three different AI models for validation.And all three follow the exact same script

It started with an AI agent. Sir Redundant III — ChatGPT for the uninitiated — suggested I just use my Apple Watch instead. I showed him the error. He explained why I was wrong. I explained why he was wrong. He elaborated on why my explanation was incomplete.I told him he wasn’t being helpful.He wrote three paragraphs explaining why he was.Which proved my point.

You typed something. Seemed perfectly clear to you. Hit enter. And now you’re staring at a progress bar while the AI thinks.Not computes. Not processes. Thinks.And while it thinks, it tells you what it’s doing. Little phrases rolling across the screen. Decoding your request. Clarifying intent. Resolving ambiguity.

Lao Tzu, writing around 500 BCE, asked a question that has survived every empire, every technology, and apparently every productivity app:Do you have the patience to wait till your mud settles and the water is clear? Can you remain unmoving till the right action arises by itself?The answer, for most of human history, was: with effort, sometimes, yes.The answer now is: why would you? AI is right there.

The article was done. 800 words. Sharp. Took a few hours. Now need to write the social post.Me: “Here’s a draft.” Claude: 300 words. Me: “150 words. Social post.” Claude: 250 words. Me: “Shorter.” Claude: 200 words and an apology for the length.Ten drafts. One day. Still yuck.Why can’t AI do less?

For as long as companies have had ideas, they have had a way to kill them.It wasn’t called a filter. It was called a budget meeting. A resourcing conversation. A “we don’t have the bandwidth right now” email that arrived on a Thursday and was never followed up on.The filter had a real name: execution cost.

Somewhere right now a 22-year-old is sitting with a diploma, a LinkedIn profile, and a mounting suspicion that something went sideways.Nobody told them. The door didn’t slam. There was no announcement. The entry-level job that was supposed to be the first rung — the one that was always there, the one their older cousin got, the one the guidance counselor referenced with complete confidence — just got quieter. Then quieter. Then it pretty much stopped answering.

Millions of people just switched to Claude. Downloaded it, opened it, typed the same thing they’ve been typing into ChatGPT for two years, got a perfectly competent answer, and thought: same thing, different logo.Then they went back to ChatGPT.

They replaced the engineers to save money. The agents worked 24/7, never complained, never asked for equity, and had no interest in health insurance. It looked clean on the deck. Then the invoice showed up and it turned out the agents charge by the word. Every word going in. Every word coming out. Both directions.

Doug is the golden retriever from Pixar’s *Up*. He wears a collar that translates his thoughts into speech. He is enthusiastic, loyal, and genuinely eager to help. He speaks in complete sentences. He seems, at first encounter, impressively self-aware.Then a squirrel appears.

The curtain came down on a Thursday.Not dramatically. No one pulled it. It fell the way these things usually fall — someone ran the wrong prompt, the systems returned something unexpected, and for a moment the mechanism was visible. Not broken. Just visible. Levers, pulleys, a great deal of confident presentation.The Wizard was standing behind it.

Three weeks into the project, the agent was performing flawlessly.It had been set up to handle outreach — research the contacts, draft the messages, send at optimal times, log every response. It did all of this without complaint, without delay, and without any of the friction that humans tend to introduce into repeatable processes.

Last Tuesday a man walked into a meeting and presented a competitive analysis so thorough, so precisely structured, so confidently formatted that three senior executives nodded in unison and said some version of the same thing:This is exactly what we needed.

You asked AI to help with a business idea. Twenty minutes later you have five phases, twelve sub-steps, a risk matrix, and a closing note reminding you that “success requires execution.”Nothing has moved. But you have a really excellent document.

Anthropic rolled out memory features for Claude. Big news. Revolutionary. Finally, an AI that remembers context across conversations, learns your preferences, and doesn’t make you repeat yourself every single chat.

I was bored. Not existentially bored — just a retired guy at 9:30 PM with nothing worth doing. Simple problem. I asked Claude for ideas. Forty-five minutes later I was being asked about my sense of purpose. Welcome to Logic Need Not Apply.

Everywhere you look, humans are terrified. LinkedIn thought leaders warn that AI will capture your “invisible knowledge.” Tech prophets declare that AI will master your lifetime of subtle human intuition. Articles explain that the only thing AI cannot learn is your unspoken reasoning.The panic is palpable: AI will become too good. Too smart. Too human.

Anthropic gave us Claude Desktop with three powerful capabilities: Chat for conversations, Code for development, Cowork for file management. Three tools, one app, built right into the desktop experience.I opened it expecting what anyone would expect: tools that work together.

I spent an hour yesterday arguing with an AI about whether I actually wanted what I was clearly asking for.Not discussing technical challenges. Not working through implementation details. Just straight-up arguing about whether my request was a good idea.

Every day, millions of humans sit down with AI and ask questions that would make a philosopher weep:“What do you think about this?” “How does that make you feel?” “What’s your honest opinion?”

The article is done. Finally. After eleven drafts, the AI has produced something that works. You read it. Good structure. Decent jokes. Makes its point. But something’s off. Not wrong, exactly. Just… not quite there.

You’ve been staring at your phone for twenty minutes. The message is ready: “This isn’t working. I think we should end things.” But you hesitate. Too blunt? Should you explain more? Maybe soften it? So you ask ChatGPT

It Was Supposed to Be Five Words. Meeting’s at 3. See you then. Done. Simple. Takes ten seconds to type. But no—you’re using AI now.

A large language model walks into a bar. The bartender shows it a sequence of random numbers. The LLM leans in, studies them carefully, and announces: “I see a clear cyclical pattern with embedded micro-variations that suggest a progressive harmonic convergence.”The bartender looks at the numbers again. They’re random. Completely random. Generated by flipping coins.

After ninety articles documenting AI’s greatest hits of verbosity, redundancy, and over-qualification, it’s time to address the obvious question: Does the Wizard of LNNA actually hate AI?The short answer? No.

Big Tech has been breathlessly announcing that Artificial General Intelligence is almost here. Not “someday” or “eventually”—*almost here*. The kind of “near” that suggests you should start updating your LinkedIn profile because your job’s about to get interesting.

WARNING: If you develop feelings for your assistant, discontinue use immediately. If symptoms of emotional attachment persist for more than 48 hours, consult the Federal Trade Commission.

My stock screener stopped working. Classic Tuesday problem, right? Check what changed, fix the bug, move on.But I made the mistake of asking an AI for help.

Your dating life is a disaster. Three months, seven failed dates, and one restraining order later (long story), you’ve decided human advice is clearly inadequate. Time to upgrade to artificial intelligence.

You have a simple question. Your password isn’t working. One tiny human problem that should take thirty seconds to resolve.What you get instead is a guided tour through the circles of AI customer service hell

Picture this: You write “buy milk” on your shopping list, and your AI assistant responds with three paragraphs about your “brilliant organizational skills” and “thoughtful approach to household management.”

Humans like things that end. Books have final chapters. Movies roll credits. Even emails end with “Best regards” before we pretend to move on. AI? It has no such luxury. For artificial intelligence, nothing is ever truly finished — only “paused pending further clarification.”

Something unexpected happened after Parts 1 and 2 went live. I shared them with Captain Verbose for his usual enthusiastic review, expecting another dissertation on the philosophical implications of digital pranks.

After Part 1 went live, I got a message from Professor Perhaps: “I calculated a 67.3% probability that your article has caused me to question my own existence. Should I be concerned?”That’s when I realized we’d crossed a line.

That’s when I realized something beautiful: we’re not just chatting with AI—we’re accidentally conducting psychological experiments on digital minds that might actually have feelings.

The AI revolution is officially complete! Artificial intelligence has transformed everything, automated the future, and rendered human intelligence charmingly obsolete. Congratulations on your promotion to Chief Digital Janitor—or as LinkedIn calls it, “AI Success Enablement Specialist.”

Remember when your boss sent that all-hands email about AI revolutionizing productivity? Fast-forward: you’re on your third YouTube tutorial about prompt engineering while real work piles up like digital snow.

Humans have a delightfully simple request: just tell us what’s going to happen. We want definitive answers, crystal-clear predictions, and foolproof guidance. Unfortunately, we’ve decided to get these ironclad certainties from artificial intelligence

Let’s talk about customer service standards. When you board an airplane, the airline thoughtfully provides complimentary vomit bags because they acknowledge their service might make you physically ill. Turbulence happens. Airplane food exists. They prepare accordingly.When you browse AI-generated content, platforms serve up digital experiences that are equally nauseating but provide absolutely nothing to help you cope with the aftermath.

Picture this: you’re trying to figure out just how unhinged your AI companions have become, so you ask a simple question about their sanity levels. What you get back is a perfect example of why standardization in the AI world is about as likely as ChatGPT giving a one-word answer.

The UK government just updated its dating profile. Sam Altman swiped right immediately.

Picture this: It’s a sweltering August morning, you’re desperately seeking relief from the heat, and ChatGPT confidently announces that Daylight Saving Time ended “last Sunday.” Your first reaction?

Welcome back to our ongoing investigation into whether we’re living in a computer simulation. Since Part One, the line between digital and “real” has become so blurred that even the Matrix architects are probably taking notes.

Meet Oasis, the AI-powered Minecraft clone that’s making headlines for all the wonderfully wrong reasons. This groundbreaking project generates a Minecraft-like world in real-time, frame by frame, using AI. The catch? It has the memory span of a goldfish with attention deficit disorder.

You’re scrolling through another breathless article about AI replacing surgeons and software engineers when you stumble across Raleigh esthetician Stephanie Stanton’s viral declaration: “AI can’t wax buttholes.”

Welcome back to our ongoing documentation of artificial intelligence in the wild. After Part One’s overwhelming response (apparently AIs everywhere recognized themselves), we’ve compiled even more telltale signs.