Count the Jev headlines with me. A launch post on the 16th. OpenJev on the 18th, Laya (“the open source version”) on the 19th, Kev (“tiny Jev-like family”) and Jev-Leftpad on the 21st, and this morning, Jev in 25 lines of Python. That’s announcement, clone, clone of the clone, parody, and deflation in seven days, a cycle that used to take a technology a few years, if anyone bothered.

I couldn’t tell you what Jev is with any authority, which seems like the right level of investment given the week. But the 25-line version is readable even for someone who missed the briefing. Load Qwen3-0.6B, a model small enough to run on almost anything. Hand it an email (“Payroll asks for your password on a non-company sign-in page”) and three lettered options: Legitimate, Spam, Phishing. The excerpt I saw cuts off before the end, but it asks for raw scores rather than generated text, so I’m guessing the rest. You read how hard the model leans toward A, B, or C, and it never writes a word. A language model reduced to a bubble sheet.

“We don’t really think so,” the author says of the hype, and then teaches you the whole trick. That’s the friendliest kind of dismissal, where the debunking is also the tutorial. You can’t finish it believing Jev is magic, and you can’t finish it unable to build one.

I haven’t opened Jev-Leftpad, but the name is doing plenty of work. Leftpad was the eleven-line npm package whose removal in 2016 broke builds across half the web, the standing joke about trivial code everyone leans on. If the deflation post now arrives on day seven instead of year three, does the technology get to be ordinary sooner, or do we skip the part where anyone finds out whether it was any good?

Sources read for this entry