LLMs can already do everything. But we absolutely need Jev
Hugo Brua uses a personal support experience to explain why a chatbot needs a route to a person. He describes a Jev router that labels example messages for human or LLM handling instead of writing the reply itself.
Original by Hugo Brua | AI EngineerClassificationBeginner9 min 1 secPublished
Separate the choice of who should handle a request from the model that writes its answer.
Test escalation with concrete examples; Hugo's router sends a reused-card complaint and a fake-profile report toward a person.
Keep exact calculations in ordinary code rather than sending them to a language model.
Worth knowing
The personal account and routing outputs are the presenter's reports, not verified customer-support outcomes. LiteLLM's linked benchmark compares 80 authored cases repeated three times per classifier; its tier-match rate is not general support accuracy. Confidence needs testing on your own requests. No router implementation was linked or executed.