Breaking the Spell of Vibe Coding

2026-05-15 · source · source ↗ · raw capture · wiki page

Author: Rachel Thomas Published: January 28, 2026 Publication: fast.ai

Summary

Rachel Thomas critiques "vibe coding" — producing large volumes of AI-generated code without human reading — drawing parallels to gambling addiction and Csikszentmihalyi's concept of "dark flow." She works at an AI company (Solve) and uses AI daily but argues the practice mirrors the addictive mechanics of slot machines: unclear feedback, "loss disguised as win" (LDW) dopamine hits, and a false sense of control over outcomes.

Key Points

Agent Psychosis

References Armin Ronacher's "agent psychosis" post describing being "hooked on Claude," building tools he "did not end up using much."

Dark Flow / Junk Flow

Csikszentmihalyi described "junk flow" as addiction to superficial, non-growth-producing experiences that are "attractive and seductive." Thomas argues this aligns with many people's vibe coding experiences.

Loss Disguised as a Win (LDW)

Multi-line slot machines reward partial "credits" — betting 20¢ and getting 15¢ back is a 5¢ loss, but celebratory sounds trigger dopamine. Research shows these induce flow-like states similar to actual wins.

Three Flow Violations in Vibe Coding

  1. Unclear performance feedback with misleading "losses disguised as wins"
  2. Murky match between challenge and skill level
  3. False sense of control over outcomes

METR Productivity Study

"When developers used AI tools, they estimated that they were working 20% faster, yet in reality they worked 19% slower" — nearly a 40% gap between perceived and actual time.

Failed Tech Predictions Catalog

Automated Coding vs. Software Engineering

"We have automated coding, but not software engineering." AI produces syntactically correct code but not useful abstractions, meaningful modularization, or improved codebase organization.

Closing Warning

Jeremy Howard: People who go all in on AI agents risk "guaranteeing their obsolescence" by outsourcing thinking and stopping upskilling.

References