Three to four months ago, Microsoft introduced Copilot Studio, replacing Power Virtual Agents. In fact, since the arrival of Copilot, a major shift in Microsoft’s strategy has occurred. Beyond just a name change, it was predictable that more AI-driven features would be integrated into the Power Platform.
In this article, I will focus on the plugin action feature.
New AI-Based Features
For those of us who were already using Power Virtual Agents, this solution was well-integrated with the Power Platform and stood out as one of the best applications for creating topic-based chatbots. For example, you could configure a scenario where, if a user requested a "report," the system would retrieve data from Dataverse via Power Automate.
With Copilot Studio, Microsoft is taking this integration to the next level.
The "Knowledge" feature is a major asset. It allows the copilot to leverage various information sources, such as:
- Uploaded files
- SharePoint
- Public websites
- Dataverse
Additionally, you can enable the option to "Allow AI to use its own general knowledge," giving the copilot the ability to respond autonomously by relying on the global knowledge of the language model (LLM). Learn more.
What is a Copilot Action ?
Beyond knowledge sources, Copilot Studio offers the Actions feature. But what exactly is an action?
An action is an operation, or a set of operations, that you define to enable the copilot (the assistant) to act in response to a user request. Actions extend the copilot’s capabilities by integrating custom interactions with your data, applications, or processes.
Basically, you can consider an action like a connection point between the copilot and an underlying logic or functionality, such as a Power Automate flow, an API command, a custom action in dataverse or a data manipulation.
When the user makes a request in the copilot interface, the copilot analyzes the intent and, if necessary, executes a defined action to respond efficiently.
An action includes a name, a display name, and a description. These fields are crucial because they guide the copilot in using generative APIs to align with the user's intent during interactions. Additionally, actions can accept input and output parameters.
The copilot autonomously handles requesting these parameters by generating relevant questions for the user. This makes actions significantly more powerful than traditional "topics," as the copilot dynamically engages with the user to collect all required information. Once all the expected parameters are provided, the action is executed independently, streamlining processes and enhancing user interactions.
Conclusion
In conclusion, actions in Copilot Studio represent a major step forward in enriching interactions between users and systems. They offer remarkable flexibility by allowing the copilot to autonomously collect the necessary parameters and execute tasks without additional intervention, while significantly reducing the number of interactions required in "topics".
For my part, I am eager to further explore the possibilities offered by this feature and to discover how it will evolve in the future. It already brings real added value by optimizing exchanges and making processes more fluid and efficient. Actions are a real promise of innovation in the automation of interactions.
Français
English