Langchain is a publication focusing on programming languages, language design, and compiler development. Readers can explore articles covering topics such as language features, syntax design, and compiler optimization techniques. Additionally, they can learn about programming language theory, language implementation challenges, and practical applications of language design principles.

LangChain

LangChain introduces AgentMiddleware, a system for customizing agent harnesses by injecting logic at specific points in the agent loop. Middleware exposes hooks like before_agent, before_model, wrap_model_call, wrap_tool_call, after_model, and after_agent, enabling use cases such as PII redaction, dynamic tool selection, context summarization, model retries, and shell tool lifecycle management. LangChain ships prebuilt middleware for common patterns and allows custom subclassing. Deep Agents is presented as a case study built entirely on create_agent with an opinionated middleware stack. The post argues middleware is the right abstraction for decoupling business logic from core agent code and enabling reuse across teams.

How Middleware Lets You Customize Your Agent Harness

Why you would want to customize your agent harnesses