How LangChain shortens debug loops on agent workflows with Sazabi.

Read the story
Go back
#Agents#Open source

LangChain

By Sazabi Team

LangChain is the framework for building applications with LLMs through composability. The same composability that makes agents powerful also makes them hard to observe — a workflow is a graph that rewrites itself at runtime.

Industry

Open-source AI tooling

Founded

San Francisco, CA

Switched

April 2026

The challenge

When an agent chain produces a wrong answer, the root cause is rarely the final step. It's a tool that returned malformed JSON four hops back, or a retry that silently swallowed an error and kept going. Traditional dashboards show you the shape of the run, not the reason it went sideways.

LangChain's team was spending its sharpest hours reconstructing agent runs by hand, copying spans into a doc to follow the thread of a single bad execution.

The solution

By streaming run logs into Sazabi, the team turned "reconstruct the run" into "ask about the run." The follow-up questions — did that tool ever return valid output? which retry masked the failure? — get answered inline.

Sazabi gives us answers our dashboards never could. Debugging an agent stopped being archaeology and started being a conversation.

Harrison Chase
Harrison ChaseCo-founder & CEO of LangChain

Results

  • faster root-cause analysis on failed agent runs
  • 60% fewer "cannot reproduce" tickets closed without a fix
  • Debugging sessions that took an afternoon now resolved before lunch

Read next

Anthropic

Anthropic x Sazabi

How Anthropic's product teams debug agentic systems with Sazabi.

Read the full story
XMTP Labs

XMTP Labs x Sazabi

How XMTP Labs cut bug detection from days to minutes with Sazabi, catching failures before they reached users across its agent-powered consumer messenger.

Read the full story