LangChain AI Updates: July 23, 2026

1. LangChain Launches an Eval Engineering Skill to Automate Test Creation

LangChain. LangChain introduced the Eval Engineering Skill, a capability that helps coding agents build evaluations by analyzing repository code and production execution traces. The skill runs an interactive interview to propose testable agent abilities and then generates executable evaluations in Harbor format. It is framed around a continuous loop of mining traces, identifying failures, building evals, and improving the agent. Source

2. LangChain Reflects on Three Years of Graph Engineering with LangGraph

LangChain. Sydney Runkle and Harrison Chase published a retrospective on three years of building LangGraph, LangChain’s framework for representing agents as graphs. The post argues that graph representations strike a balance between deterministic paths and agentic flexibility, letting builders encode domain knowledge into workflows. It contends that the approach has become more powerful now that individual agents are reliable enough to embed as nodes within larger graphs. Source

3. Apollo Rebuilds Its GTM Assistant on Deep Agents

LangChain. LangChain published a case study on how Apollo rebuilt its AI Assistant using Deep Agents to unify a fragmented go-to-market platform into a single chat interface. Apollo replaced a supervisor-based system with a skill-driven architecture and now exposes the assistant as a headless API and MCP server, with more than 40,000 teams adopting it. The company used LangSmith for evaluation across six layers and reports development cycles reduced by roughly 80 to 85 percent. Source