Autonomous Agentic AI using Microsoft Copilot Studio
We are talking about a custom engineered Copilot which can execute the Azure VM lifecycle management (start and stop for demo) autonomously onto Azure cloud with simple natural language prompts through chat using Copilot Studio.
Note: Not to be confused with Microsoft Copilot for Azure (wrote about it here in 2023).
Let’s go!!
The rapid evolution of GenAI, agents and declarative automation have paved the way for innovative solutions that can streamline operations and amplify productivity — in a manner never imagined before.
I will Tell (the theory) and Show (the practical) in this article.
The Theory (tell part):
This Autonomous Agentic AI, referred to as Autonomous Agentic Copilot here by me, is specifically designed to execute workflows for managing virtual machine (VM) lifecycle.
Leveraging Microsoft Copilot Studio and GenAI integration via Azure OpenAI, this system enables seamless automation, orchestration, and interaction for users. In this blog, we will dive into the architecture and functionality of this revolutionary system, as depicted in the provided diagram above (Fig.1).
The Autonomous Agentic Copilot integrates several critical components to provide a holistic and autonomous solution for workflow management. Let’s break down its key elements.
User Interface: The starting point of interaction, the user interface (UI), is intuitively designed to facilitate user engagement. Through this intuitive interface, users can create, configure, delete, modify, publish, export, and import agents. It provides a seamless way for users to interact with the system and execute their desired operations.
At the heart of the Autonomous Agentic Copilot lies its modular architecture, which includes the following components:
Data (Knowledge): This serves as the knowledge repository, storing all relevant information required for workflow execution. It ensures that the system has access to up-to-date and contextual data. This was not needed for this use case.
Orchestration (GenAI-based): The orchestration layer, powered by generative AI, decides how to respond to users, the events and actions configured.
Functions: Allow the execution of predefined tasks.
Tests: Provide mechanisms to validate workflows within the chatbot in the side bar.
Triggers: Act as the initiator for workflows based on specific conditions.
Actions: Execute the defined operations within a workflow. In this case, I had created two actions — 1. Stop the VM and 2. Start the VM.
These components of Autonomous Agentic Copilot integrates with GenAI (Azure OpenAI) to enhance automation and intelligence. This integration also enables Natural Language Processing (NLP), allowing users to interact with the system using natural language prompts.
Workflows and VM Lifecycle Management: The system’s primary use case here is managing VM lifecycles. By integrating with Azure Cloud, the Copilot automates tasks such as stopping and starting VMs.
Azure Cloud Integration: The workflows are integrated with the Azure cloud. Azure Cloud serves as the landing zone for the infrastructure for VM lifecycle management. Through connector’s integration, this Copilot interacts autonomously with Azure to perform lifecycle operations (start and stop), based on the user’s simple prompts in English.
Users simply chat with this Copilot and provide simple prompts in English — such as, stop the VM or start the VM.
The Practical (show part):
I have created two Agentic Copilots in there. We will focus on the first one in the list here, i.e. Azure VM Manager (name of the Copilot) as shown below. The other one is the procurement Copilot (not in scope for this article).
The agent configuration (after creation and publishing) looks like as following:
I created two workflows (actions) configured into the agent for execution, which are also integrated with Azure cloud.
Now, I started chatting. I asked it to start my Azure VM (Vish-VM) in chat window in the side bar (Test your agent section).
Please refer to the zoomed-in chat below.
I validated it from the Azure cloud portal. My VM was started (running) successfully as shown below.
Now, I wanted to test the stop / power off execution through the natural language input. This time, I had configured it in a way where it asks me for the details, such as, subcription id, resource group and the VM name. I entered those inputs in the chat window in the side bar as shown below.
Zoomed-in chat with the Copilot as shown here.
I then validated from the Azure cloud portal and the VM was stopped as we can see below.
That was it. The use case was very much executed successfully.
Conclusion:
- The Agentic AI can in-deed execute tasks based on natural language prompts.
- The Autonomous Agentic Copilot here represents a promise of significant leap forward in workflow automation and cloud resource management using GenAI. In other words, by combining Microsoft Copilot Studio (orchestrated with GenAI) and Azure cloud, it can deliver efficiency and user experience in a modern AI-way. For enterprises looking to streamline operations and embrace the future of AI-driven automation, the Autonomous Agentic Copilot can offer a compelling solution. While some of the capabilities are in preview mode, it gives a glimpse of how the Autonomous Agentic Copilots can work side by side with human admins (and/or more Copilots) in near future.
- Many other use cases can be created e.g. procurement, HR, DevOps etc.
- This is not a production ready solution just yet (in my view).
What is for sure is, The Sky is the limit going forward in this space.
Disclaimer: Personal post, personal point of view and personal use case.
Thanks !
Vishal Anand.
