March 7, 2026
Codependence, but make it Claude
Addicted to Claude Code–Help
Is it genius or slot-machine brain? Most say play on, a few yell “intervention”
TLDR: People say they’re “addicted” to Claude Code, the AI coding helper, and the community splits: some call it harmless fun and skill‑building, others warn it’s gambling‑brain vibes. Fans gush over charts and demos, while skeptics note it’s error‑prone for serious work and hard to explain to non‑tech folks.
Confession time: folks are getting hooked on Claude Code, the AI that writes and runs code, and the comments came in hot. The original post says it’s probably fine—for now—because everyone’s building a mental “how to use it” model, like learning to Google. But then the mood whiplash hits: one reply warns this pro‑AI crowd is like asking a slot‑machine buddy if you have a gambling problem—spoiler, you won’t get a real intervention. Others wave it off with a shrug: “There are worse things to be addicted to.”
There’s real anxiety too. One veteran dev admits it feels like a new toy, but worries what happens if it disappears. Another compares Claude Code to a supercharged coding tool, urging people to chase long‑term joy, not dopamine sprints. Skeptics throw shade at AI “agents” (AI helpers that can do steps for you and even run code): they’re power‑user tech that can go wrong, and you still need to read the actual papers for serious work. Also: the AI can sound brilliant and be confidently wrong—like mixing up niche physics categories—so don’t trust it for high‑stakes stuff.
Meanwhile, meme city: chart nerds demanded receipts—“Link the charts!”—and one commenter derailed into Sankey obsession, daring someone to beat the classic steam engine diagram. The vibe? Part playground, part rehab, all drama.
Key Points
- •The author reports a surge in intensive use of Claude Code after a noted performance improvement in November 2025, alongside Codex.
- •Early heavy use is framed as helpful for building a practical mental model of capabilities and limits, with concern warranted only if it persists for months.
- •A concrete limitation is highlighted: the model produced a subtle but incorrect claim about Rydberg and neutral atoms in quantum computing.
- •Advice for non-technical audiences focuses on demonstrations and a safe subset of tasks, such as executing code on uploaded files via ChatGPT or Claude.
- •The piece notes rapid evolution in best practices as models advance, and the significant capability gap between frontier and free models.