April 6, 2026
When vibes write the code
The Cult of Vibe Coding Is Insane
Leaked “English code” ignites roast: genius shortcut or lazy vibes
TLDR: Claude’s leaked “English code” sparked a brawl over hands‑off “vibe coding”: critics say it breeds messy, duplicate work, while defenders call it normal fast‑ship chaos and just another abstraction. The debate matters because it’s about who’s really building modern software: humans guiding, or bots coding under human direction.
A leaked chunk of Claude’s “English-written” internals set the internet on fire, and the comments are the main show. The author blasts “vibe coding” — the idea of building software by chatting with AI and never peeking under the hood — calling it a culty spin on dogfooding. Cue the popcorn: some readers are howling that this hands‑off approach led to messy, duplicated stuff; others say relax, this is just how fast‑moving products look before cleanup.
The strongest take? “AI isn’t the product — people’s feedback is.” One commenter argues the real magic is humans correcting the bot until it stops saying dumb things. On the other side, defenders say AI is just another layer of abstraction, like when coders moved from assembly to functions; every generation complains the next is “cheating.” Meanwhile, a crowd notes the author isn’t a rando — commenters point out he’s the BitTorrent guy — adding authority to the roast.
Humor flew too: people joked that Claude is at “AI Level 7: human writes the plan, robot does the homework,” and dubbed it “manifesting features.” The split vibe? Team Roast says “open the hood and fix it.” Team Ship-It says “millions use it; messy is normal, we iterate.” Drama level: premium.
Key Points
- •The article cites a leak of Claude’s source code and claims public criticism focused on poor code quality.
- •It attributes problems to excessive dogfooding and a practice termed ‘vibe coding,’ where developers avoid inspecting internal code.
- •The author argues pure vibe coding is unrealistic because AI needs human-provided structure such as plan files, skills, and rules.
- •A proposed remediation process includes auditing for items duplicated as both agents and tools, defining guidelines, and consolidating duplicates with AI assistance.
- •The article states AI can rapidly reduce technical debt when guided by prior ‘Ask mode’ dialogue to clarify edge cases and expectations.