Grok Bot 团队访谈：用 Jev 做类型化决策开关

silentguy @silentguyy66原推发布于 2026.10.01

在 X 上打开原推

SpaceXAI engineer who shipped Grok Bot: "Grok Bot has its own computer, it runs even if you're not at your computer" 28 minutes with two of the people who built it: 00:28 - every bug report, her bots go fix it and she just reviews the code 13:05 - a lot of grok bot was built with grok bot 15:13 - the bot dispatches cloud agents and hands back a PR at the end 18:36 - rules and style guides become hard constraints that fail ci 24:21 - grok 4.6 scores 70.8% at $2.81 a task. the other model: $17.32 She runs a team of bots and still makes every call by hand Which one picks up the task, whether the fix is real, whether a PR can merge, whether a human has to look at it Chatbot used as a switch statement Jev answers with typed options plus a probability, $0.042 per million input tokens, and never writes a sentence LLM creates the work -> Jev decides what happens next -> code enforces it Save this before you spawn another bot

这条是引用着另一条写的：「Jev 把路由打分验证做成决策层」