How Salesforce Standardizes Agent Evals with LangSmith @LangChain
How Salesforce Standardizes Agent Evals with LangSmith  @LangChain
Uploaded July 2026 | Updated November 2026, 1 day ago
Ben Sklar, Director of Product Management at Salesforce, explains how the team behind Agentforce Vibes, Salesforce's enterprise vibe coding product, moved from teams building their own one-off MCP tools to a shared evaluation standard with LangChain and LangSmith.

In this video, Ben walks through how LangSmith lets them:
✅ Run evals at scale before code reaches customers
✅ Standardize scoring across instruction following, coherence, factuality, and deployability
✅ Support the milestone of over 100 million lines of accepted code

Chapters:
0:00 The problem with every team building its own MCP tools
0:22 Making sure agent output does what you expect
0:35 Finding LangChain and LangSmith
0:53 Running evals at scale before code reaches customers
1:19 Setting one scoring standard across every team
1:40 100 million lines of accepted code

Get started with LangSmith: langchain.com/langsmith-platform
How Salesforce Standardizes Agent Evals with LangSmithTrace Every Codex Session in LangSmith in MinutesAgents in the Enterprise with MongoDB | Interrupt 26Evaluating Agents in Production: Traces, LLM-as-Judge, and Prompt Management at WonderThe One-Line Integration That Got Clays Agents Into LangSmithThe LangChain Team Answers the Most Searched Questions About AgentsShould Agents Say I Dont Know? | AirBnB at Interrupt 26Why the Harness Matters More Than the ModelHow Bridgewater Built an AI Analyst That Does Hours of Expert Research in MinutesThe 3 Things Clay Measures on Every AI AgentMCP Explained: The USB-C Cable for AI AgentsBuilding Clinical AI Agents with LangGraph: Abridges Eval Stack for High-Stakes Healthcare
LangChain |

How Salesforce Standardizes Agent Evals with LangSmith

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER