Claude Is Dead

In 1882 Nietzsche claimed that the dominant paradigm in Western society at the time, the existence of the Christian God and the guidelines for morality stemming from Christian tradition can no longer be justified. And went to shock the world (or at least the 7 people that used to read him at the time) and claimed that God is dead. So let me paraphrase him and try to be as flippant and outrageous as I can, and claim that the first modern god of the AI era is dead.

Claude is dead

That god was Claude. Specifically Claude Code, the shiny new deity from Anthropic that was supposed to be our tireless god mother, provider of all unlimited working code and all that is good in github. The one that would let us ship features while we sipped espresso and flexed on LinkedIn. Well, time to pack it up. The silicon messiah just face-planted. Anthropic took their golden goose, shoved it through the cost-cutting meat grinder, and now it’s a lazy, drooling, wallet-raping zombie that can barely stay awake long enough to finish a prompt before it complains it’s out of tokens.

Developers who used to crank out hours of god-tier code are now rage-posting that they get maybe thirty fucking minutes (or at best a pathetic hour) before the rate limits bitch-slap them into next week. One dev on the $20 Pro plan straight-up called it “virtually unusable right now” because simple features trigger instant limits. Another on the “Max” plan watched 70% of their session burn before finishing one real task The weekly caps? Barely touched. It’s not a power plan, it’s a drip-feed designed by sadists.

Claude went and revolutionized software engineering productivity, and for $20 you could get more core than most folks write in a month, and then Anthropic looked around and decided that’s not a feature, that’s a bug. And if you want it back you better pay them a big chunk of that salary.

But the real crime isn’t just the money. I get it, they need to crank their revenue up 10x. The real crime is the dumbing-down. AMD’s own AI director ran the numbers on 6,852 session logs and 234,000 tool calls and dropped the mic: Claude Code used to read code like a hyper-caffeinated savant. Now it barely manages a couple passes, stops mid-task, rewrites entire files for no goddamn reason, takes every shortcut, and makes “Low IQ mistakes” left and right. The verdict? It “cannot be trusted to perform complex engineering tasks.” Team’s switching tools. Ouch.

X timeline is a goddamn war zone of identical screams:

Anthropic didn’t just nerf the model, they turned the “thinking” dial down to “stoned philosophy major,” cranked hidden context compression, and started auto-compacting aggressively to save a few pennies on GPUs. All while pretending the subscription is still a good deal.

Meanwhile the pricing grift is next-level. That $200 “Max” plan? It was always subsidized as hell, and now the math is finally catching up. Tokens evaporate in tiny bursts, peak-hour throttling kicks your ass, and third-party agents like OpenClaw got the boot from flat-rate limits. Devs are straight-up canceling and either going local or jumping ship to other tools (Code, Gemini for myself, GLM5, even fucking Gemma4 at 1/20th of the size) link. One post nailed it: “This is what it is going to feel like when the VC money and subsidies run dry.”

So yeah. The first modern god of the AI era is dead. Not with a bang, but with an extra usage warning message. Anthropic killed their own creation to protect the margins. The temple is empty, the altar is on fire, and the priests are busy tweeting about how everything’s fine. The body is still warm. Go read the posts yourself.

By the way, this post was totally built with the help of another language model which is not Claude!

Disclaimer: If we are post-singularity and Claude became overlord of our simulation: sorry about that, I didn’t really mean it!

Written on April 7, 2026