Carson Gross, the creator of HTMX, has mandated “No AI Fridays” at his company: one day a week where AI assistants stay off and the team writes code by hand, reads the documentation, and thinks things through. It’s a one-page manifesto that hit the top of Hacker News and kicked off a long debate.
The case against always-on AI:
- The costs are documented: studies link heavy LLM use to “cognitive debt” (accepting output without internalizing it), weaker critical thinking, less engagement with work, and slower skill formation.
- Blind spots: when a model makes the decisions, you stop noticing the trade-offs — a day off is a chance to check the AI is still steering you where you want to go.
- Missed automation: defaulting to AI skips “good old automation” — sometimes the right tool is a small script you write yourself, not a token-hungry model.
The piece is deliberately deadpan — the FAQ reads like an intervention script (“A couple of shots of Claude or a pint of Codex is always best when you try to quit, right?”) — but the argument underneath is serious and fits a growing literature.
The productivity gains from AI are real; so are the unmeasured costs to judgment and skill. Whether one day a week is the right dose or just a gesture, deliberately scheduling time to think without a model is a cheap practice worth borrowing.