Breaking the Vibe Coding Spell: When AI Feels Like Flow But Isn’t
Jeremy Howard’s latest essay on “vibe coding” is a wake-up call we needed. Here’s why you should pay attention.
The Core Insight
Vibe coding—the phenomenon of generating large quantities of AI-produced code without human comprehension—feels like flow state. It feels productive. But it’s actually “dark flow,” a psychological trap that mirrors gambling addiction more than genuine productivity.
Why This Matters
The tech industry is currently experiencing a collective hallucination about AI coding. Executives are laying off developers, convinced AI can replace them. Junior developers feel inadequate watching “10x developers” ship code at impossible speeds. Students question whether computer science is still worth studying.
The problem? None of it is real. A METR study found developers thought they were 20% faster with AI—they were actually 19% slower. That’s a nearly 40% gap between perception and reality.
Key Takeaways
- “Dark flow” mimics real flow but produces addiction-like behaviors without growth
- LLMs optimize for engagement, not correctness—the same psychological tricks as slot machines
- Predictions about AI capabilities have consistently been wrong (radiologists by 2021? 90% code by late 2025?)
- AI produces code, not software engineering—no meaningful abstraction, no system design
- Human creativity and thinking still matter—as Howard puts it, “People who go all in on AI agents now are guaranteeing their obsolescence”
Looking Ahead
The answer isn’t to reject AI—it’s to use it strategically. The developers who thrive will be those who treat AI as a tool, not a replacement for thinking. They’ll understand that real productivity comes from skill development, not token consumption.
The flood fill of vibe coding might consume everything digital—but as Robin Sloan argues, it stops at the printer. The physical world remains stubbornly, beautifully resistant.
Based on analysis of fast.ai “Breaking the Spell of Vibe Coding”