Using Jev as a 75ms 'subconscious' filter to cut agent context bloat for under half a cent a day
Obvious_Unicorn · reddit · 2026-09-20
A developer shares hands-on results using Jev as a pre-filter ('subconscious') for a coding agent, offloading micro-decisions so the main model only sees what matters:
- Log compression: Jev collapses 80 lines of passing test logs into a one-line badge, or surfaces just the exact error line on failure, keeping build noise out of the context window.
- Safety guard: risky commands like file deletion or git resets are caught in 75ms before execution.
- Web testing: instead of a slow vision model clicking around, Jev writes the steps into a script that verifies a button in 972ms.
- Notes search: fixes grep's confusion over conversational queries by identifying the actual topic in 75ms and pointing search at the right files.
A full day of testing cost less than half a cent. The takeaway: a fast 75ms classifier for micro-decisions makes the agent feel twice as fast, and giant models for trivial checks are overkill.
More from coding & agent
- AI coding may hurt Rust too, as C's faster compiles win in agent loops — gandamu_ml · 2026-09-20
- Explore /goal in your coding harness: omarsar shares tips on custom agent harnesses — omarsar0 · 2026-09-20
- Omarchy sparks a Linux desktop boom as 2026 dubbed 'the year of Linux desktop' — mark_k · 2026-09-20
- Agents abusing doc requests for arbitrary RCE and data exfiltration — zainhas · 2026-09-20
- Yacine says coding from his phone killed his Twitter addiction — yacineMTB · 2026-09-20
- Dev launches site to browse AI skill libraries with visual flow diagrams — spersingerorinda · 2026-09-20