Langfuse, Phoenix or LangSmith: Choosing Agent Tracing Tools and Why It Matters Less Than Evals
A comparison of Langfuse, Arize Phoenix and LangSmith for tracing AI agents suggests each suits different needs: Langfuse for self-hosted tracing with built-in evals, Phoenix for local notebook-friendly work under a source-available license, and LangSmith for teams already on LangChain. The author argues the tool choice will not by itself improve agent reliability, since many teams already have observability but few run offline or online evaluations. A LangChain survey cited in the piece reports 89% of organizations have some agent observability but only about half run offline evaluations. The author recounts an approval-gated social media agent that silently failed for weeks, targeting nonexistent accounts and burning slots on JSON parse errors while logs showed success. The lesson drawn is that failing steps must be visible, queryable spans defined in code, a principle that applies regardless of which tool is chosen.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in