Jev Projects Roundup Thread

The author says the coolest projects he could find on X that day, with no inclusion criteria and no verification. Every item is a link to somebody else's post, so the quality bar is whatever those posters claimed, and none of the numbers are checked in the thread itself.

No, they are self-reported and scoped differently: one project claims 129 rows judged in about a second, another 500 emails in seconds for 3.5 cents, another about $0.0002 per voice decision, another 7 seconds for a flight search. Different workloads, so the numbers do not aggregate.

Several, and the thread is best used as a link list for that purpose. The PostgreSQL natural-language search extension, the Claude Code compaction plugin, and the browser-use agent are all open source in the linked posts, while the voice and keystroke entries are described as experiments with no repository linked.

The latter. It is a directory of demos from the first days of the launch, and the author says he was still working out how the technology fits his own business. The rebuild of Tesla full self driving in under an hour is the clearest case of enthusiasm rather than result.

Yes. Item five, browser controlled with voice, is his own, described with probabilities returning in about 300 milliseconds and a cost of $0.0002 per decision. The post closes by pointing to his newsletter and saying he will test Jev further, so the list mixes one of his projects with other people's.

Three days. It was posted on September 18, 2026, after Jev launched on September 15, and the framing is explicitly that these are the coolest projects found on X that day. That recency is the thread's strength and its limit: it captures the first wave before anything had time to mature.

He says he will test it extensively that week and work out how the technology fits his own business, promising to share learnings in his newsletter. So the thread is partly a curated roundup and partly positioning from a creator who is about to publish his own findings.