LangSmith launches AI Search: query agent traces in natural language
LangChain · x · 2026-10-13
LangChain announced AI Search in LangSmith at the Interrupt conference, targeting a common observability pain point: finding traces by agent behavior.
Questions like 'does the user seem frustrated?', 'where did the agent fail to complete the task?', or 'where is the user trying to cancel their subscription?' previously required writing evaluators or reading hours of conversations. Now you just describe the situation in natural language and LangSmith returns the matching traces.
arjunnatarajan notes that searching traces by agent behavior, user qualities, or execution patterns was previously impossible or incredibly convoluted, and it's one of the most common questions the team gets. A practical update for teams doing agent evals and debugging.
More from coding & agent
- Anthropic merges 30,000+ code changes a day, built in-house git replacement Scion in 5 months — AnthropicAI · 2026-10-14
- Dev builds voice-first macOS focus assistant that nags you like your mom, powered by Sarvam's bulbul v4 — itsOmSarraf_ · 2026-10-14
- Anthropic's daily code merges up 10x since May, peaking at 30,401 in a day — inductionheads · 2026-10-14
- Plane Agents ship: AI agents as workspace members you assign and @ in comments — JosephJacks_ · 2026-10-14
- Developer predicts AI will collapse the software dev cycle by 2027, from prompt to shipped in a day — NickPassig · 2026-10-14
- 20 free prompts to unlock the other 90% of Claude Cowork's SEO potential — JaynitMakwana · 2026-10-14