Make AI Work for You: How to Build Your Own ChatGPT with No-Code

AI is evolving fast, and the days of needing deep coding expertise to build intelligent systems are fading. Thanks to Copilot extensibility, you can now create custom AI agents tailored to the task at hand - all with little to no coding required.

But what exactly are AI agents, and how can you build one? Let’s break it down.

What Are AI Agents?

Think of an AI agent as your own personal chatGPT, but customized for performing a particular task. Instead of just answering questions, these agents can be used to understand context, automate tasks, and interact with apps and data. Whether it’s replying to emails or generating reports, an AI agent can do the heavy lifting.

Microsoft Copilot allows you to extend its natural language capabilities by customizing AI agents for specific tasks, making AI work the way you need it to.

General Architecture of AI Agents extended from Copilot

The general anatomy of an AI Agent built by extending copilot looks something like the image attached below. Lets break it down

1. Foundation models - The underlying LLM model you want your AI Agent to use for decison making and understanding prompts.
2. Knowledge - Data supplied for the agent to learn
3. Skills - Abilities and responsibilities of the agent - eg: a customer support bot is different from a retail product inventory bot
4. Autonomy - This helps the AI think ahead. It can plan complex tasks, learn from past interactions, and escalate issues when it doesn’t know what to do.
5. User Experience – This is the part, user interacts with! Whether it's a chatbot, voice assistant, or an embedded AI in an app, this is how you communicate with the agent.
6. Orchestrator – Think of this as the brain's decision-maker. It takes your request, figures out what needs to be done, and coordinates between different parts of the system.

Types of Agents

You can build agents by using either a declarative or custom engine approach. This gives you, as a developer, the flexibility to choose how you want to build your agents, while providing the same experience to users.

1. Declarative agents are agents that user can Plug and Play. They can only supply the knowledge and skills part of the architecture. The underlying model used , UI and the orchestrator are all abstracted and handled behind the scenes. These agents are best to use for automation of small and easy tasks with less customization. Mostly used when low code approach is preferred. Example: A declarative agent that auto-summarizes emails and updates meeting notes in Microsoft Teams.
2. Custom engine agents are more adaptive and flexible allowing the developer to change all aspects of the architecture. The UI can be a custom web app with streamlit while the model can be Cohere or Mistral model based on the need. This is mostly pro code alllowing the developer to have more control over the agent. Example: A custom AI sales assistant that pulls data from CRM, personalizes emails, and books meetings automatically.

How can you get started?

1. Microsoft learn offers multiple learning paths and hands on labs to get users accustomed to building agents - login with an existing account and follow the instructions on the lab sandbox.
2. You will need an account for copilot studio - login and be mindul of the usage limits to ensure you dont run out

Some references

Copilot labs - https://developer.microsoft.com/en-us/microsoft-365/copilot

Learning paths - https://learn.microsoft.com/en-us/training/browse/?terms=copilot%20studio&products=ms-copilot

Docs - https://learn.microsoft.com/en-us/microsoft-365-copilot/extensibility/

#ADCBlogs #AzureDeveloperCommunity Azure Developer Community