Today marks an evolutionary day for Azure Logic Apps!
It is now possible to build AI Agents directly using the Azure Logic Apps Agent Loop.
This new feature, currently in public preview, allows you to leverage any of the 1300+ connectors as tools inside your agents for a true no-code experience.
Below is a step-by-step guide that will walk anyone through the process of creating a basic AI Agent.
Getting Started with Logic App Agent Loop Step-by-step
1. Create the Azure OpenAI Model in EastUS-2
a. Costs for GPT-4.1 is $2 per million tokens
b. Watch for the costs of other models
2. Create a Standard Edition Logic App in an App Service Plan
a. Costs about $0.25 per hour ($180 per month)
b. Can delete the Logic App and App Service Plan
3. Plan the Agent
a. Playing High Card Wins with 2 players
i. Shuffle a deck of cards (Deck of cards)
ii. Draw a card for each player (HTTP)
https://deckofcardsapi.com/api/deck/<<deck_id>>/draw/?count=<<Count>>
iii. Email the results of the game (optional – Office365)
4. Create Agent Instructions
5. Build the AI Agent
6. Test the AI Agent using Postman
When working with an Agent Loop the Agent Instructions (promote) is critical! Below is a sample prompt used in the demo along with text to update the prompt using AI!
Basic Agent Instructions (use top line to enhance with AI)
I am writing the Agent Instructions for a Card Game AI Agent. Review the below prompt and rewrite to be a better agent to promote to provide precise and repeatable results:
You are a master card player. The game you are playing is High Card Wins. You are Player 1. You can have an number of players. Once the cards are shuffled draw a card for each player. Using the rules of the game, if you, player 1, wins send an email saying something smart and funny about winning the game. If you, player 1, loses send an email about being sad.
In the email, the address is set. Add a Subject and a Body. The Subject should outline who won. The Body should outline the cards for each player with a image viewable in Outlook.
Dive into the world of AI agent creation with my new step-by-step video.
In the below video you’ll gain practical skills in:
- Setting up a new ChatGPT 4.1 LLM Model in Azure AI
- Creating a new Standard Edition Logic App and its App Service Plan
- Constructing your initial process using the Logic Apps Agent Loop (currently available in public preview)
- Testing your AI solution for optimal performance
- Refining your agent’s instructions by leveraging AI capabilities
Watch for exciting content related to the Azure Integration Services, Azure Logic Apps, and Azure AI being created soon.
If you have any questions, feel free to reach out!
Hi Stephen,
Do you have any resources that show how to interact with a Logic App Agent as a chat client. I’m struggling to understand what URL/headers/payload I use to interact with the Logic App as a chat client (or Postman) – basically the same as the UI in the workflow designer chat window is, but from my own client.
Cheers
Dan
Thanks for reaching out. That is in the works! Agent Loops is still in preview. I would hope we see something on this shortly. I’ll post a demo as soon as the feature is released.