AI is having a rough week
In a Nutshell
A YouTuber blames AI for his 9-month jobless streak after losing to a candidate who used 100% AI-generated code with superior features, urging devs to embrace "slop" production as software postings rise 11% and skills like code readability become obsolete. AI tools like Copilot shift to costly token-based billing amid 25% error rates and document corruption in workflows, while a dad freaks out over his 9-year-old's Gemini use, valuing human depth in kids' creativity over AI's superficial width. Humans win with story-driven depth; AI excels in breadth but robs learning, process, and genuine skill-building.
These notes were generated by AI and may contain inaccuracies.
A YouTuber makes a video about the miserable job market for software developers. Thumbnail shows his left eye super red. Video starts unhinged: stares deadpan at camera while sipping tea without breaking eye contact for 5 seconds, then leans in, points at eye, says, "Yeah, this this is from AI. This is literally from the stress of job hunting."
He's been laid off for nine months, applying non-stop, going far into interview processes. One company: nine rounds, 100% sure job is his, doesn't get it. Heart drops. Asks why, they share winning job entry. Winner used 100% AI, had 30 more features than Garrett who used 50% AI and 50% manual.
Garrett like guy in 1849 who heard about gold in California and walked to Nevada instead. Winner wasn't even present when application submitted—maybe taking a walk while Claude was formulating. Winner now has equity and 401k. Garrett at home writing unit tests, making code dry and ship. His eye filling with blood. One guy hemorrhaging. Market chose the AI user.
Software engineering postings up 11%. Market wants more stuff, faster, cheaper. Market responds to quality/maintainability concerns: "Did I stutter, bitch?" Thing that got you hired in 2019 gets you fired in 2026. Nobody reads your code anymore—dead skill like reading hieroglyphics. Adapt and face the slop. Don't argue. Imagine more slop. Do the slop. Be the slop. Slop is peace. Slop is strength. You will eat zbugs and you will make zlop. Learn to enjoy it.
For over two years, AI tools greatest sale in software history: $20/month for half-brilliant, half-senior engineer in a box. CEO almost had heart attack. Called it unlimited—meant without limit. Now unlimited means until we find out how much you want, then real number. Was a survey. In 2026, results: you'll pay anything. Now price is anything.
Microsoft leaked Copilot moving to token-based billing because weekly cost nearly doubled since January. Bad answer costs as much as good. Tokens include model's wasted tokens thinking out loud: argues with itself, writes draft, second-guesses, starts over, "Actually, let me reconsider." Paying for model's nervous breakdown to millionth of syllable. Answer wrong 25% of time per new Microsoft paper.
New paper on arXiv last week: "LLMs corrupt your documents when you delegate." Authors from Microsoft Research. Across 52 domains, frontier models corrupt 25% of document content by end of long workflow. Errors sparse but severe—like a sniper. Gave model tools: made it 6% worse. First time tool becomes less useful with more capability.
Experiment: edit document then undo. Best AI couldn't without insane corruption. We had Control Z—it was free.
Dad walks in on 9-year-old daughter on laptop, she slams it shut, cries, "Dad, it's not what you think." He knew: using Google AI (Gemini). Worried for her future. She asked: how to get along with sisters, swim faster after meet, fanfiction plot lines for favorite book series.
Dad posts on Reddit like discovered her selling feet pics. "She's not in trouble, but long conversation, she's devastated." Now understands how sycophantic and insidious it is. Taught kid two four-syllable insults for bot. Poor kid's spelling test: teacher asks "Use sycophantic in a sentence," she says, "The large language model was sycophantic. It is also insidious."
Dad struggles to explain tech he doesn't understand: no precedent for omniscient always-on ghost in kid's pocket. Want kids good at something, earn skills. But friend's parents let her use AI: at 9 shipping AAA games to Steam, by 13 friend shipping enterprise SaaS with 3% uptime (Anthropic would kill for that).
Speaker conflicted: 4-year-old's house drawing (square, triangle roof, sun) makes him cry. AI-generated oil on canvas, stunningly beautiful, unimpressed, bothered. Why ugly thing moves, pretty thing cringes?
AI has whopping width—sheer breathness impossible for humans. Humans have depth impossible for AI. Cry at sloppy drawing for infinite depth: manifestation of love, 4 years nurturing—from not knowing "dada," crawling, laughing, walking, eating with spoon, to drawing family home. That crayon painting more Mona Lisa than actual.
AI art: no story. Without story, humans aimless floating molecules. Art is depth, is story.
AI usage no benefit except commercial (limited even then). Hobby project: for process enjoyment or learning/skill climb. AI robs both—not enjoyable nor educational.
Dad did right thing: gave daughter chance to struggle vs. gambling on OpenAI/Anthropic oversold promise. Chose own values over Sam Altman and Dario Amodei. If wrong, Opus 20 by college requires no mind (arguably true today). She gains voice, style, perspective, suffering, experience, depth, longing, understanding, story about life/beauty/lack.
AI kid learned to delegate. Nothing else.
Truth somewhere in middle. Curious what you think.
Keep Mo Bitar in your library
Save the videos and channels worth coming back to, and find them again in one place.
