Jev Isn't for Coding: It's a General-Purpose Semantic Judgment Function for Agent Loops

d3n2el · reddit · 2026-10-12

Observing the Jev ecosystem, the author argues that almost all Jev projects are coding projects — not because the pattern is about coding, but because developers build coding loops first.

His core thesis:

Jev's real role isn't generation; it answers small, bounded questions inside a loop: which action, which file, keep or drop, which model to route to, safe or not
Big models still own long-form generation and deep reasoning; Jev is the judgment node called repeatedly in between
Swap the tool step for docs, spreadsheets, CRM records, or email and the structure is identical — and those scenarios vastly outnumber coding
He demonstrates with WorkBuddy, his non-coding harness: Jev decides whether a document routes to the expensive model, whether a spreadsheet row needs human confirmation, which queue a message joins

Conclusion: Jev isn't a chatbot competitor but a general-purpose semantic judgment function — state plus bounded question in, probability/score/options out, and your software decides what happens next. What determines output quality isn't the model but the standards you embed in the loop. Open question: which steps stay outside the loop? His only answer so far: acceptance criteria.

Original post →

Jev Isn't for Coding: It's a General-Purpose Semantic Judgment Function for Agent Loops

More from coding & agent