How to not use AI. Or staying sane in a frenzied world
summary
Argues that low-cost experimentation from transformer models reduces the cost of being wrong but increases decision fatigue and burnout, citing a 600‑participant hackathon where tool use jumps from ~5% using Cursor in 2025 to near‑ubiquitous agents in 2026. Describes a migration initially 40% over client budget where a two‑hour AI draft cuts effort by about one third and unlocks the deal. Recommends guardrails: OKF docs for AI context, pre‑commit checks, browser verification, and one attention stream at a time.
AI-written summary · about
