The idea
Vlad Terin presents Jev Browser, an adapter in which a coding agent plans a task, Jev selects actions and browser tools execute them. The demo starts from a spoken instruction.
Inside the decision loop
Transcribed request + browser state
Select the next available action
Execute through browser tools
What to take from it
This combines three responsibilities: interpreting the overall request, choosing a local action and operating the interface. A clear boundary makes each piece easier to replace or evaluate.
The source describes a real-time recording, but the entire experience also depends on speech transcription, browser tooling and the planning model. Jev alone is not the complete system.
Follow the original work
This case is an editorial interpretation of a public community demonstration. It has not been reproduced or benchmarked by this publication.
Original post by @VladTerin