zeeg: UIs are dead, use traces — recommends vitest-evals over UI-based evals
zeeg · x · 2026-09-12
Reacting to LukeParker's OpenEval project, zeeg argues "UIs are dead, you need traces" and suggests just using vitest-evals to view full session traces and tool calls — making it easy to spot agent cheating and bad judge decisions.
The exchange touches on two philosophies of agent evaluation: a polished local web UI vs. baking evals into a test framework with trace-first inspection.
Related event: Debating agent evals: traces over UI(2 posts)→
More from coding & agent
- ADSP Ep. 303: Mark Saroufim on Open vs Closed Models, AI GPU Kernels and Autoresearch — blelbach · 2026-09-12
- Google releases free 2-hour course on agent graph & loop engineering — goyalshaliniuk · 2026-09-12
- Gemini CLI nightly patches indirect prompt injection and hardens sandbox filesystem — gemini-cli-robot · 2026-09-12
- Repurpose an old low-VRAM GPU just for mmproj in llama.cpp — an order of magnitude faster — inthesearchof · 2026-09-12
- Ollama vs llama.cpp: same request counts 85k vs 164k tokens in OpenCode — RadianceTower · 2026-09-12
- Open-source Pentest Copilot runs autonomous pentests on a Kali box, 1.3k GitHub stars — tom_doerr · 2026-09-12