Blog
Guide
The mess of OTel semantic conventions and why tracing CLI tools is still painful
An honest look at the state of OpenTelemetry semantic conventions for LLM applications, the specific challenges of tracing CLI-based AI tools, and practical workarounds that actually work in production.
Hendrix Liu · 3 days ago
Guide
Claude Code vs Codex: What traces reveal about how they actually work
A trace-level comparison of Claude Code and OpenAI Codex CLI: how they plan, call tools, recover from errors, and spend tokens, and what that means for choosing between them.
Frank Chen · 4 days ago
Guide
Your LLM Judge Has Opinions. They're Not About Quality.
A deep dive into the systematic biases in LLM-as-a-judge evaluation, including self-preference, verbosity, authority, agreeableness, and position bias, and how to design experiments that produce conclusions you can actually trust.
Frank Chen · 5 days ago
Guide
The 25 ways your LLM app is failing (and why you can't see most of them)
A catalog of 25 LLM failure modes in production: loud failures you can catch, silent failures that hide in plain sight, and slow failures that only reveal themselves in hindsight.
Frank Chen · 6 days ago