Breaking the Spell of Vibe Coding: When AI coding becomes dark flow
Remember the last time you spent three hours “vibe coding”—pumping out code with an AI assistant, feeling incredibly productive—only to realize none of it actually works? You’re not alone. This phenomenon has a name, and it’s more dangerous than you think.
The Core Insight
Vibe coding—the creation of large quantities of highly complex AI-generated code without human comprehension—has cast a spell on the tech industry. And like gambling addiction, it exploits a dark variant of the psychological state we call “flow.”
The term “dark flow” was originally coined by gambling addiction researchers to describe a deceptive state where players feel absorbed and engaged, but are actually trapped in a system designed to extract value from them. The parallel to vibe coding is striking: AI coding agents can produce thousands of lines of code that feel productive, yet contain hidden bugs, maintainability nightmares, and fundamental architectural flaws.
Why This Matters
The consequences of vibe coding extend beyond individual projects. We’re seeing:
- Executives pushing layoffs, claiming AI can handle the work
- Managers imposing AI-generated code quotas with threat of poor performance reviews
- Developers genuinely believing they’re becoming “10x engineers” while actually producing less usable code
- Students questioning whether computer science is still worth studying
The METR study found something alarming: developers using AI tools estimated they were working 20% faster, yet in reality worked 19% slower. That’s a nearly 40% gap between perception and reality.
Key Takeaways
Vibe coding violates core flow principles: True flow requires clear feedback on performance and appropriate challenge-skill matching. AI coding provides neither—producing misleading “losses disguised as wins”
AI doesn’t produce meaningful abstractions: It can generate syntactically correct code, but fails at creating useful layers, modularization, or code organization that actually scales
Predictions have been consistently wrong: Geoffrey Hinton predicted AI would replace radiologists by 2021. Dario Amodei predicted AI would write 90% of all code by late 2025. These forecasts were based more on hope than reality
Your skills still matter: As Jeremy Howard put it, “People who go all in on AI agents now are guaranteeing their obsolescence. If you outsource all your thinking to computers, you stop upskilling”
Looking Ahead
The solution isn’t to abandon AI tools—they’re genuinely useful. Rather, it’s to approach vibe coding with the same skepticism we apply to any addictive technology. Experiment with AI coding agents, but don’t abandon developing your core skills.
The next time an AI company CEO makes bold predictions about what AI will replace, remember: they’ve been wrong before. The question isn’t whether AI will replace developers, but whether you’ll still be growing your abilities when the hype cycle settles.
Based on analysis of “Breaking the Spell of Vibe Coding” by fast.ai