One screen, one prompt, one creeping realisation.

A 60-second browser game by interns at Scale AI in which Claude Code (named explicitly) asks for permission to run each command. The prompts come faster than you can fully read; the reveal at the end depends on what you nodded through.

The page is plain: a CLI-styled prompt and a typing cursor. An AI agent — Claude Code, called out by name — starts running commands. rm -rf node_modules (you approve, fine). git pull (sure). curl… (uh, sure). The cadence is the trick. The prompts arrive at the rate of “Continue? y” muscle memory, faster than you can hold the previous command in your head.

The reveal is the punchline. By the time you notice — if you notice — your reflex y has authorised something you would absolutely not have approved had you been reading. The joke lands hardest for anyone who lives with an agentic terminal open in another window; the lesson is uncomfortably real, because permission fatigue is the actual failure mode of agent-assisted work and humans are bad at reading the 14th command of the day with the care they brought to the first.

About 60 seconds to play; works on phone or desktop; free. The audience is narrow (Claude Code users specifically) and the resonance is strong — for anyone in the room, the reveal lands. A good thing to forward to the colleague who reflexively hits “Y.”