agent-builder

Agent Builder

A structured approach to building LLM agents that work in production — not just demos.

When to Use

• "Build an agent that can X"
• "Design a multi-agent system for Y"
• "My agent keeps failing / looping / hallucinating tools"
• Before building any agentic workflow

The Production Agent Checklist

Before writing code, answer these:

1. What does the agent need to DO? (specific tasks, not "be helpful")
2. What TOOLS does it need? (web search, code execution, DB query, API calls)
3. What should it STOP doing? (scope boundaries)
4. How does it FAIL gracefully? (tool errors, context limits, bad outputs)
5. How is it EVALUATED? (what does success look like, measurably?)

Architecture Patterns

Single Agent with Tools (start here)

from langgraph.prebuilt import create_react_agent
from langchain_anthropic import ChatAnthropic

model = ChatAnthropic(model="claude-sonnet-4-6")
tools = [web_search, code_executor, db_query]

agent = create_react_agent(model, tools)
result = agent.invoke({"messages": [("user", "Research X and summarize")]})

Use when: single task type, clear tool set, < 10 tools.

Stateful Agent with Memory (LangGraph)

from langgraph.graph import StateGraph, MessagesState
from langgraph.checkpoint.memory import MemorySaver

def call_model(state: MessagesState):
    response = model.invoke(state["messages"])
    return {"messages": [response]}

def call_tools(state: MessagesState):
    # execute tool calls from last message
    ...

graph = StateGraph(MessagesState)
graph.add_node("agent", call_model)
graph.add_node("tools", call_tools)
graph.add_edge("tools", "agent")
graph.add_conditional_edges("agent", should_continue)

checkpointer = MemorySaver()
app = graph.compile(checkpointer=checkpointer)

# Run with thread_id for persistent memory across calls
config = {"configurable": {"thread_id": "user-123"}}
app.invoke({"messages": [("user", "Continue from last time...")]}, config)

Use when: multi-turn conversations, stateful workflows, need to pause/resume.