* This blog post is a summary of this video.

Power Virtual Agent Levels Up with Copilot Studio AI Plugins for Smarter Bots

Table of Contents

Introduction to Copilot Studio and AI Plugins

Copilot Studio, formerly known as Power Virtual Agents, has evolved to include all the functionality from Microsoft's AI Builder product. This means classic bots built with Power Virtual Agents can now be supercharged with AI capabilities using plugins.

There are at least seven different types of plugins available in Copilot Studio. These reusable pieces of code act as extensions that give copilots added functionality through easy integration with external data sources and services.

Copilot Studio Overview

Copilot is a term used to describe several different types of virtual assistants created by Microsoft. There are Microsoft 365 copilots for productivity tasks inside Teams, 1st party copilots from Microsoft products like Dynamics 365, and custom copilots built for specific organizational tasks. Custom copilots can be embedded into Teams, Power Apps, websites, and more. They are similar to classic bots built previously but with far more advanced AI capabilities.

Types of AI Plugins

There are over 7 plugin types available. These include:

  • Conversational plugins - Built with low-code tools and used to extend Microsoft 365 copilots
  • AI plugins - Leverage GPT tech for content generation, custom automation, custom connectors, etc.
  • OpenAI plugins - Easy integration using OpenAI framework and manifests
  • Bot Framework plugins - Extend custom copilots built with the Azure Bot Framework
  • Action plugins - Call custom connectors from within custom copilots using natural language

Using AI Plugins with Custom Connectors

One way to extend copilot capabilities is by building custom connectors that integrate with external systems. These connectors expose operations that can be mapped to AI plugins.

For example, a custom connector could integrate with an e-commerce order database. The operations might include retrieving order details or updating order statuses.

By mapping these operations to copilot plugin actions, you enable the bot to understand requests like "Get the status of order #1001" or "Update order #1001 status to shipped". The copilot handles interpreting the natural language, extracting order IDs and parameters, calling the right connector operation, and formatting the response.

This helps overcome limitations with linear, static conversations in classic bots. The AI chatterbot handles dynamic topic switching and parameter passing automatically through a feature called "dynamic chaining".

Action Plugins for Custom Copilots

Action plugins specifically relate to custom copilots. They allow invoking custom connectors through natural conversation. You describe the connector, operations, parameters etc. in easy to understand language. Copilot's AI then dynamically determines when and how to call the operation based on the user's input. Responses can be standardized text or even adaptive cards that mimic human-like exchanges.

Dynamic Chaining for Natural Conversations

With dynamic chaining enabled, copilots leverage the GPT-3.5 language model to process conversations naturally without the rigid structures required previously. There's no need to manage topic switching or slot filling - the AI handles this automatically based on the action plugin configuration. This makes interactions far more human-like. Users can change subjects fluidly and the bot follows along, requesting clarification only when needed.

Creating Adaptive Cards with Plugin Responses

While plugin actions can display simple text responses, generating adaptive cards provides far more flexibility for rich, engaging exchanges.

Using Power Fx, developers can iterate over plugin response data to build adaptive card payloads dynamically. This works similarly to creating reactive UIs in canvas Power Apps.

With cards, copilots can display images, graphs, tables, clickable actions and more. It's even possible to invoke additional plugin operations through button clicks on these cards.

Formatting Output with Adaptive Cards

Adaptive cards make plugin action output more visually engaging and usable for end users. Developers control card design and can use Power Fx to access plugin response data. Formatting options include text formatting, images containers, fact sets, graphs and charts. Buttons with actions allow invoking other plugins. According to Microsoft, these cards may soon auto-generate dynamically based on example layouts given by the developer.

Multi-Language Support

An exciting capability showcased is how adaptive cards from plugins gracefully handle multi-language scenarios. Users can switch between languages when interacting with the copilot. The copilot automatically translates text on cards to match user preference. Additionally, plugin responses seem to support multiple languages without extra effort. This simplifies developing multi-lingual chatbot experiences significantly, while retaining full customization capabilities.

Key Takeaways and Next Steps

Copilot Studio's extensive AI plugin capabilities help elevate simple rule-based chatbots into intelligent and natural conversational experiences.

Leveraging custom connectors and adaptive card responses, developers can build highly engaging custom copilots that users will love interacting with.

As Microsoft continues enhancing their AI offerings, we can expect even smarter and human-like copilots in the future. But even the currently available functionality opens tremendous possibilities for improving customer and employee conversational interfaces.

Hopefully this overview gives you inspiration on the art of the possible today. Dive deeper into Copilot Studio, explore all the different plugin actions available, and unleash your creativity to build the next generation of intelligent chatbots!

FAQ

Q: What is Copilot Studio?
A: Copilot Studio is the new name for Power Virtual Agents in Microsoft's AI Builder tools. It provides a low-code interface to create AI-powered conversational bots called copilots.

Q: How do AI plugins work with copilots?
A: AI plugins allow you to extend your copilots with reusable pieces of AI functionality. There are plugins for natural language generation, custom automation, connectors, and more.

Q: Can I use my custom connectors?
A: Yes, you can create action plugins to integrate your custom connectors into your copilots. This allows copilots to call your connectors to access external data and systems.

Q: What is dynamic chaining?
A: Dynamic chaining is a new copilot feature that uses AI to have natural, context-aware conversations without predefined dialog flows. The copilot dynamically chains together topics.

Q: How do I customize responses?
A: You can format plugin responses using adaptive cards for richer visuals. Adaptive cards can be multi-lingual to support conversations in different languages.