AI Agent Frameworks Compared 2026: LangGraph, AutoGen, CrewAI, OpenAI Agents SDK

# Core agent loop — works with any LLM, no framework needed
def agent_loop(task: str, tools: list, max_steps: int = 20) -> str:
    messages = [{"role": "user", "content": task}]

    for step in range(max_steps):
        # 1. LLM decides what to do
        response = llm.call(messages=messages, tools=tools)

        # 2. If done, return
        if response.stop_reason == "end_turn":
            return response.content

        # 3. Execute tool calls
        tool_results = []
        for tool_call in response.tool_calls:
            result = execute_tool(tool_call.name, tool_call.input)
            tool_results.append(result)

        # 4. Add to history, loop again
        messages.append({"role": "assistant", "content": response.content})
        messages.append({"role": "tool", "content": tool_results})

    return "Max steps reached"

from langgraph.graph import StateGraph, END
from typing import TypedDict, Literal

class AgentState(TypedDict):
    messages: list
    tool_results: list
    iterations: int

def call_model(state: AgentState) -> AgentState:
    response = llm.invoke(state["messages"])
    return {"messages": state["messages"] + [response]}

def should_continue(state: AgentState) -> Literal["tools", "end"]:
    last_message = state["messages"][-1]
    if last_message.tool_calls:
        return "tools"
    return "end"

# Build graph
workflow = StateGraph(AgentState)
workflow.add_node("agent", call_model)
workflow.add_node("tools", tool_executor)
workflow.add_conditional_edges("agent", should_continue, {
    "tools": "tools",
    "end": END,
})
workflow.add_edge("tools", "agent")
workflow.set_entry_point("agent")

app = workflow.compile(checkpointer=memory)  # Persistent state!

from openai_agents import Agent, handoff, guardrail

# Define specialized agents
triage_agent = Agent(
    name="Triage",
    model="gpt-5.4",
    instructions="Route queries to the right specialist.",
    handoffs=[
        handoff(billing_agent, description="For billing questions"),
        handoff(tech_agent, description="For technical issues"),
    ]
)

billing_agent = Agent(
    name="Billing",
    model="gpt-5.4-mini",  # Cheaper for simpler tasks
    instructions="You are a billing specialist...",
    tools=[lookup_invoice, process_refund]
)

# Run
result = await triage_agent.run("I was charged twice last month")

from crewai import Agent, Task, Crew

researcher = Agent(
    role="Senior Research Analyst",
    goal="Find comprehensive information on {topic}",
    backstory="Expert at synthesizing complex information from multiple sources",
    tools=[search_tool, web_scraper],
    llm="claude-sonnet-4-6"
)

writer = Agent(
    role="Content Writer",
    goal="Write a compelling article based on research",
    backstory="Skilled at turning research into engaging long-form content",
    llm="claude-sonnet-4-6"
)

# Tasks
research_task = Task(
    description="Research the latest developments in {topic}",
    agent=researcher,
    expected_output="Bullet-point summary of key findings"
)

write_task = Task(
    description="Write a 1500-word article based on the research",
    agent=writer,
    context=[research_task],  # Uses researcher's output
    expected_output="Full article in markdown"
)

crew = Crew(agents=[researcher, writer], tasks=[research_task, write_task])
result = crew.kickoff(inputs={"topic": "agentic AI in 2026"})