Home / Articles / Practical notes: Microsoft Agent Framework 1.0 vs LangGraph vs CrewAI: A

This article is published in English.

langgraph agents llm architecture typescript python

Practical notes: Microsoft Agent Framework 1.0 vs LangGraph vs CrewAI: A

Operable walkthrough of Practical notes: Microsoft Agent Framework 1.0 vs LangGraph vs CrewAI: A: contracts, checks, and drop-in code slots for teams shipping this pattern.

2397 words

Use this as an operator-facing rebuild of the ideas in “Microsoft Agent Framework 1.0 vs LangGraph vs CrewAI: A Production-Ready Comparison (2026)”: clear stages, ordered code slots, and recovery notes that survive a handoff. The Overview stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

TL;DR

For the TL DR stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

Why This Comparison Matters Right Now

For the Why This Comparison Matters stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

What Changed Since 2025

For the What Changed Since 2025 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the What Changed Since 2025 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

How Each AI Agent Framework Works: The Core Philosophy

When working through the How Each AI Agent stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

# LangGraph: explicit graph control
from langgraph.graph import StateGraph
from langgraph.checkpoint.sqlite import SqliteSaver

graph = StateGraph(State)
graph.add_node("classify", classify_doc)
graph.add_node("extract", extract_data)
graph.add_conditional_edges("classify", route_by_type)
app = graph.compile(checkpointer=SqliteSaver("agent.db"))

# CrewAI: role-based simplicity
from crewai import Agent, Crew, Process

researcher = Agent(role="Researcher", goal="Find accurate data",
                   llm="gpt-4o")
writer = Agent(role="Writer", goal="Draft clear reports",
               llm="gpt-4o")
crew = Crew(agents=[researcher, writer], tasks=[...],
            process=Process.sequential)

# MS Agent Framework: simple agent in Python
import asyncio
from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient

async def main():
    client = AIProjectClient(
        endpoint="https://your-project.services.ai.azure.com/api",
        credential=DefaultAzureCredential()
    )
    agent = client.agents.create_agent(
        model="gpt-4.1",
        instructions="You are a concise data analyst."
    )
asyncio.run(main())

Head-to-Head: What Actually Matters in Production

When working through the Head-to-Head What Actually Matters stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

State Management & Persistence

When working through the State Management Persistence stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the State Management Persistence stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Memory

The Memory stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Debugging & Observability

The Debugging Observability stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Protocol Support (MCP & A2A)

The Protocol Support MCP A2A stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Expose tools with narrow schemas and explicit side-effect labels. Hosts need to know which calls mutate state before they auto-approve. The Protocol Support MCP A2A stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Two Paths for Python on Azure

For the Two Paths for Python stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Separate client construction from the message loop so providers can be swapped without rewriting the conversation state machine.

Path 1: MS Agent Framework (Python)

For the Path 1 MS Agent stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Separate client construction from the message loop so providers can be swapped without rewriting the conversation state machine.

# MAF Python: working agent in a few lines
import asyncio
from agent_framework import Agent
from agent_framework.openai import OpenAIChatClient

async def main():
    agent = Agent(
        client=OpenAIChatClient(),
        instructions="You are a helpful data analyst.",
        tools=[get_weather, get_menu_specials]
    )
    response = await agent.run("What's the weather in Amsterdam?")
asyncio.run(main())

Path 2: LangGraph + `langchain-azure-ai`

For the Path 2 LangGraph langchain-azure-ai stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the Path 2 LangGraph langchain-azure-ai stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

# LangGraph + Azure Foundry via the bridge
from langchain_azure_ai.chat_models import AzureAIOpenAIApiChatModel
from azure.identity import DefaultAzureCredential

model = AzureAIOpenAIApiChatModel(
    project_endpoint="https://myproject.services.ai.azure.com/api",
    credential=DefaultAzureCredential(),
    model="gpt-5.2"
)

How to Choose Between Them

When working through the How to Choose Between stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

The Cost of AI Agent Deployment

When working through the The Cost of AI stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

the Decision Flowchart

When working through the the Decision Flowchart stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the the Decision Flowchart stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

The Pattern you See Across Clients

The The Pattern you See stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Beyond the Framework: What Actually Determines Success

The Beyond the Framework What stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Operational checklist

When working through the Operational checklist stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest.

Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion.

Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Pin dependency versions and record the image digest that ran the demo. Reproducibility beats tribal knowledge.

Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.