LangChain Agents

Agents merge language models with tools to build systems capable of reasoning about tasks, deciding which tools to suit the situation, and iteratively working toward solutions. The create_agent function provides a production-ready implementation.

Tools enable agent actions. Unlike simple binding, agents facilitate multiple sequential calls, parallel execution, dynamic selection, retry logic, and state persistence. Static Tools: Defined at creation using the @tool decorator: `python from langchain.tools import tool @tool def search(query: str) -> str: """Search for information.""" return f"Results for: {query}" agent = create_agent(model, tools=[search]) ` Dynamic Tools: Modified at runtime through two approaches -- filtering pre-registered tools or registering tools discovered at runtime (from MCP servers or external registries).

Agents follow a loop pattern: input flows to the model, which decides on actions, calls appropriate tools, receives observations, and repeats until reaching a stop condition (final output or iteration limit).

Static Models: Set once at creation and remain constant throughout execution. `python from langchain.agents import create_agent agent = create_agent("openai:gpt-5.4", tools=tools) ` For granular control, instantiate model objects directly: `python from langchain_openai import ChatOpenAI model = ChatOpenAI(model="gpt-5.4", temperature=0.1, max_tokens=1000) agent = create_agent(model, tools=tools) ` Dynamic Models: Selected at runtime based on state and context using the @wrap_model_call decorator for sophisticated routing and cost optimization.