Azure Functions vs Microsoft Foundry Routines: Which One Should You Use for AI Agents?

-

If you're scheduling AI agents in Azure, you've probably defaulted to Azure Functions. But Microsoft Foundry just introduced something that might change your approach — Foundry Routines.

I recently built the same AI agent health check using both Azure Functions and Foundry Routines to see how they compare. The results were eye-opening.

The Use Case

I created an AI agent called HealthCheckAgent in Microsoft Foundry with a web search tool. 

The goal: ping the agent every 10 minutes, verify it responds, check if it used the web search tool, and log a status summary.

Simple enough. But the setup complexity was very different depending on which approach I used.

Azure Functions: Powerful but Heavy

To get a Python timer-triggered Azure Function running, I needed a Function App (Flex Consumption plan since Python requires Linux), a Storage Account, Managed Identity with Entra ID authentication, IAM role assignments on the Foundry resource, three environment variables, and a deployment pipeline.

That's six infrastructure steps and roughly 70 lines of Python — before the function even runs once.

The upside? Full control. Custom health check logic with latency tracking, tool usage verification, and structured console output with clear healthy/unhealthy indicators. Plus App Insights integration for production monitoring.

Foundry Routines: Simple but Limited

The same scheduling task with Foundry Routines took 0 (about 15, if you want to do using code) lines of Python. No storage account. No Function App. No deployment pipeline. No environment variables.

The downside? No custom logging, no console output, limited run history in the portal, and no way to add custom health check logic like tool usage verification. Routines are currently in preview with timer and schedule triggers only.

Which Should You Choose?

Use Azure Functions when you need custom monitoring, non-time-based triggers (HTTP, blob, queue), response processing, or production-grade observability.

Use Foundry Routines when you need zero-infrastructure scheduled agent runs, you're prototyping, or the agent's built-in response is sufficient.

For most teams, you'll end up using both — Routines for simple scheduling, Functions for custom orchestration.

Learn More

I've covered this topic in detail with full code walkthroughs and live demos:

Watch the full YouTube tutorial here:


Read the in-depth Medium article HERE

If you're building AI agents on Azure, understanding both tools will save you hours of setup time and help you pick the right approach for each scenario.

Comments

Post a Comment

Popular posts from this blog

Create Chatbot Based Using GPT-Index/LlamaIndex | OpenAI | Python

-
Image
In this article, I’ll show you how can you create a basic chat bot which utilizes the data provided by you. Here we will be using GPT-Index/LlamaIndex, OpenAI and Pytho n. Let’s get started by installing the required Python module. Install modules/packages We need to install, two packages named  llama-index and langchain and this can be done using below lines: pip install llama-index pip install langchain Importing packages Next, we need to import those packages so that we can use them: from llama_index import SimpleDirectoryReader , GPTListIndex , GPTVectorStoreIndex , LLMPredictor , PromptHelper , ServiceContext , StorageContext ,load_index_from_storage from langchain import OpenAI import sys import os Please note that, here, we don’t need an GPU because we are not doing anything local. All we are doing is using OpenAI server. Grab OpenAI Key To grab the OpenAI key, you need to go to https://openai.com/, login and then grab the keys using highlighted way: Once you got the...
Read more

Generate The Streamed Response Using Semantic Kernel

-
Image
Semantic Kernel , a powerful tool for integrating large language models into your applications, now supports streaming responses. In this blog post, we’ll explore how to leverage this feature to obtain streamed results from LLMs like AzureOpenAI and OpenAI. Why Streamed Responses Matter When working with language models, especially in conversational scenarios, streaming responses offer several advantages: Real-time Interaction: Streaming allows you to receive partial responses as they become available, enabling more interactive and dynamic conversations. Reduced Latency: Instead of waiting for the entire response, you can start processing and displaying content incrementally, reducing overall latency. Efficient Resource Usage: Streaming conserves memory and resources by handling data in smaller chunks. How to Use Streaming Responses I've published a complete video on how to generate this using Python and that can be found on my YouTube channel named Shweta Lodha . Here, I'm jus...
Read more

Azure Model Router: The Smart AI Traffic Controller

-
Image
Imagine you're at a busy airport, and planes from different airlines are landing and taking off. To keep everything running smoothly, air traffic controllers decide which runway each plane should use. Now, think of AI models as those planes—each one has different strengths, speeds, and capabilities. The Azure Model Router  acts like an air traffic controller for AI models, ensuring that every request gets handled by the best model available. What is Azure Model Router? Azure Model Router is a smart AI system that automatically selects the best AI model to respond to a request. Instead of developers manually choosing which AI model to use, the Model Router does it for them, optimizing for speed, cost, and accuracy. It’s part of Azure AI Foundry , a platform that helps businesses and developers deploy AI models efficiently. Why Do We Need It? AI models come in different types—some are great at answering questions, others are better at reasoning, and some are super-fast but less detai...
Read more