Usutaku shows several Jev experiments, from classifying generated customer emails to selecting browser actions. The Japanese-language video combines his own demo pages and extensions with examples by other developers.
Sort emails and page elements
- At 4:52, he generates customer-service emails and routes them into inquiry, complaint, thanks and other categories.
- He reports processing 100 generated emails in 27 seconds in one run. The video does not establish labeled accuracy, and the result was not independently reproduced.
- At 7:22, his browser extension removes ads from a coffee article he built to contain them. One constructed page cannot establish performance across the web.
Use decisions to drive a browser
- At 9:57, he demonstrates a personal-information form, an apartment inquiry and a simulated shopping flow on pages he made for testing.
- The apparent checkout is on a mock storefront, not a verified purchase from Amazon. The video does not test real-world permissions, payment safety or reliability.
- At 13:26, a separate speech-recognition system transcribes commands; a Jev-based voice browser then operates a company website. He says the voice tool was published by someone else.
- At 15:45, he shows another developer’s Flappy Bird comparison. That clip is a third-party demonstration, not his own controlled benchmark.
「実際のサイトでやるとそのサイトにね、迷惑がかかる」 — Usutaku explains why he uses his own test pages: running the demos on real sites could impose a burden on them. (10:12; translated)
Caveat: The presenter’s speed and price comparisons are not an independently reproduced benchmark. Typed choices and apparent completion do not establish that a decision was correct or safe.