First Jev waitlist invitations go out
TypeSafe says it emailed a first batch of waitlisted users and will invite more people as access expands.
The post notes that with Jev Skills, Codex still writes code while Jev handles quick judgments like multiple choice, scoring, and allow/block, cutting latency from seconds to tens of milliseconds and illustrating a division of labor between expensive and fast models.
先別被「不用工作了」騙到。 Codex 裝 Jev Skills,不是 Agent 突然會做人。 是它終於不用每一步都請貴模型思考。 寫代碼的還是 Codex。 做單選、打分、放行/攔截的是 Jev。 一個 3 秒,一個幾十毫秒。 大多數人還在用最貴的模型判斷最便宜的問題。 這篇值得看的,不是標題,是這個分工。