多助理实测填写韩国入境表,仅 Instinct 一次成功
作者在飞机起飞前让多个助理填韩国入境表:Dot、Hermes、Openclaw 因敏感操作拒填,Muse 找不到按钮,Codex 反复确认,只有 Instinct 一次完成。



tried one shotting Korean immigration (arrival) forms with several agents as my flight left the tarmac only one agent came through Dot failed Hermes’ & Openclaw failed Muse horribly failed Dot, Hermes’ and openclaw came close but refused to do anything sensitive like submitting forms. This guardrail only started in the gpt6 era. Muse kept saying it couldn’t find certain buttons to click codex w/ 5.6 came really close. It kept asking if I was sure if I wanted to submit everytime it got to a sensitive step. It would submit but only after a “yes do it” so which product one shotted it for my family and I? Instinct. i wasn’t sure what it submitted is even correct but I went through Korean immigration without a hitch. @noahrshinn cooked on this one