# Kush pulled Jev out of Puffle's harness

DevFeed: [Kush pulled Jev out of Puffle's harness](<https://devfeed.tech/articles/kush-pulled-jev-out-of-puffle-s-harness-100486.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-26-w7>)

Author: Jev Weekly

Published: 2026-10-26T14:17:35.931057Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [human review](<https://devfeed.tech/topics/human-review.md>), [benchmark overfitting machine learning](<https://devfeed.tech/topics/benchmark-overfitting-machine-learning.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [qwen](<https://devfeed.tech/topics/qwen.md>)

Tags: [about](<https://devfeed.tech/tags/about.md>), [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [all](<https://devfeed.tech/tags/all.md>), [also](<https://devfeed.tech/tags/also.md>), [also-on-the-timeline](<https://devfeed.tech/tags/also-on-the-timeline.md>), [article](<https://devfeed.tech/tags/article.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [best](<https://devfeed.tech/tags/best.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [false-negative](<https://devfeed.tech/tags/false-negative.md>), [file](<https://devfeed.tech/tags/file.md>), [flash](<https://devfeed.tech/tags/flash.md>), [harness](<https://devfeed.tech/tags/harness.md>), [jev](<https://devfeed.tech/tags/jev.md>), [making](<https://devfeed.tech/tags/making.md>), [qwen](<https://devfeed.tech/tags/qwen.md>), [reading](<https://devfeed.tech/tags/reading.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [worth-reading](<https://devfeed.tech/tags/worth-reading.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

## AI overview

Kush removed Jev from the Puffle harness and returned to Qwen 3.7 flash after Jev repeatedly misjudged whether files were worth reading, leading the agent astray. Jev was intended to scout files and improve RAG precision, but Kush says false negatives--skipping relevant files--were especially costly and hurt their benchmarks. The excerpt also describes Lightpanda using Jev in judge() to classify page content, and Composio Firewall evaluating agent actions with Jev plus tuned SLMs; it does not explain how those uses performed.

## Source excerpt

Kush says Jev kept making wrong calls on whether a file was worth reading, so Puffle went back to Qwen 3.7 flash. Where a false negative is expensive, a cheap...