The World Has Gone Crazy
It's past three in the morning again. The room is dark except for two monitors, a terminal window, and those stupid violet LEDs I bought as a joke and somehow never removed. There's cold coffee on the desk, acrylic keycaps that looked cooler in photos than in real life, and a hoodie that has probably seen more kernel updates than daylight. One sock has little stripes on it. The other one disappeared somewhere under cables, adapters, and old SSDs. I'm mentioning all this because I still like things that make sense. Machines, systems, logs, source code, tiny routines that do exactly what they were told to do. Which is probably why the last two years have felt like watching the industry slowly lose its mind.
Everyone suddenly loves AI
Five years ago most people outside tech barely cared about machine learning. It was a niche thing. Researchers, data people, some weird startup founders, a few engineers arguing on forums at two in the morning. Now every company on earth acts like it discovered fire. Doesn't matter if they build accounting software, search engines, note-taking apps, or smart fridges. Somehow AI ended up in everything.
The big companies are easy to understand. They need stories. They need growth. They need something shiny for investors who get bored every six months. "We're integrating AI across the platform" sounds much better than "revenue is flat". Smaller companies don't even try to hide it. They're scared. If they don't slap AI on the landing page, somebody else will, and suddenly they look old.
So now your text editor wants to brainstorm with you. Your email client wants to rewrite your personality. Your search engine doesn't search anymore - it "understands". Customer support became a chatbot that confidently invents policies that don't exist. Nobody asked for any of this. But somehow we're all supposed to clap.
And the funny part? Most of it is bullshit.
Take away the branding, the demo videos, the TED-talk voices, and what's left is not intelligence. It's token prediction. That's it. A giant probability machine guessing what word should come next, then the next one, then the next one, wrapped in enough polish that people forget what's actually happening underneath.
People talk about context like these systems remember things. They don't. They juggle text inside a window, drop pieces when it gets full, compress things when they run out of room, and fake continuity just well enough that most users don't notice. Sometimes they contradict themselves ten messages later. Sometimes they forget things five lines later. Sometimes they explain something brilliantly, then completely invent the next fact with absolute confidence.
And hallucinations? I love how people talk about them like bugs. Hallucination is not a bug. It's what happens when you build a machine whose only job is to generate something statistically plausible. If it sounds right, it might be right. Or not. The machine doesn't care.
Developers are being dragged into it
This part annoys me the most, because this one hits close to home. Developers aren't being asked if they want these tools. They're being told. Copilot in the IDE. AI reviews. AI documentation. AI sprint planning. AI architecture suggestions. AI summaries of meetings nobody wanted to attend in the first place.
And sure, sometimes it spits out something useful. Sometimes it also spits out race conditions, broken edge cases, fake APIs, security holes, and code that looks clean until production traffic touches it. That's the dangerous part. Bad code used to look bad. Now bad code often looks polished.
I know people who quietly disable these tools and never mention it. I know people who use them just enough so management sees the icon in screenshots. I know senior engineers who spend half their day cleaning up "productivity gains" generated by systems that have never maintained software for five years. Nobody says much. Everyone knows why.
The money still doesn't add up
This whole thing is supposed to be the future, but for something so revolutionary, it just burns cash. Training costs are insane. Running inference at scale is insane. GPUs, cooling, datacenters, networking, power - none of this is cheap, and none of it magically becomes cheap because the UI has rounded corners and a gradient background.
Subscriptions help, sure. But every time someone generates a wall of text, somebody somewhere pays for that compute. And from what I can tell, a lot of these companies are still living on investor patience, not actual healthy business.
Maybe the numbers work someday. Maybe they don't. Right now it feels less like a mature industry and more like everybody agreed not to ask difficult questions while the music is still playing.
And somehow nobody wants to talk about the damage
People are losing jobs. Sometimes because AI replaced them. Sometimes because AI gave management an excuse they already wanted.
At the same time, the quality of everything feels worse. More bugs. More fake content. More autogenerated tutorials written by systems that don't understand what they're explaining. More products shipping features nobody tested properly because "the model looked confident".
And maybe that's the strangest part. Everyone sees it. Engineers see it in logs. Designers see it in broken flows. Customers see it in support chats that make no sense. And honestly... some nights it really feels like the world has gone completely insane. I come back to my Emacs with newly polished fingertips.