LangChain and LangGraph Agent Frameworks Reach v1.0 Milestones (blog.langchain.com)

LangChain and LangGraph have reached v1.0, marking a stability milestone and a clearer split of responsibilities: LangChain 1.0 is the high-level, provider-agnostic fast path for building agents (now centered on a new create_agent abstraction and a trimmed API surface), while LangGraph 1.0 is the lower-level, durable runtime for production-grade, long-running and highly controllable workflows. LangChain tightens the core agent loop (model → tool calls → final answer), folds structured output generation into that loop to cut latency and cost, and introduces standardized content_blocks so outputs (reasoning traces, citations, tool calls) stay consistent across providers. Key new capabilities include middleware hooks (built-ins for human-in-the-loop approval, summarization for context management, and PII redaction), and a response_format/ToolStrategy pattern for typed structured outputs. LangGraph 1.0 provides durable state, built-in persistence, observability, and first-class human-in-the-loop patterns so interrupted workflows resume cleanly and multi-day or sensitive processes can be implemented without bespoke DB glue. LangChain agents run on LangGraph, letting teams start fast and drop to graph-level control when needed; the releases preserve backward compatibility (with legacy APIs moved to langchain-classic), deprecate langgraph.prebuilt/create_react_agent, require Python 3.10+, and promise no breaking changes until 2.0. With wide adoption in production and consolidated docs, these 1.0 releases signal maturation of agent tooling—faster shipping for common patterns and stronger primitives for production reliability and compliance.