The Great Adaptive Card Showdown: Copilot Actions vs. Topics!

Welcome to another exciting episode of “Raife Bytes,” where we dive into the tech trenches and emerge with nuggets of wisdom, this time Copilot & Adaptive Cards!

Today, we’re pitting two titans against each other in the ultimate showdown, well not quite as much of a show down as Anthony Joshua vs Daniel Dubois but in the tech world it’s pretty yuge,

I’m talking Adaptive Cards in Copilot Actions vs. Topics! of course…

Oh yes, It’s like watching a boxing match between two great behemoths for the IBF heavyweight title, well kinda. The main premise of me wanting to use an adaptive card was to find a nice way to present a list of clinical “observations” from a healthcare data services instance in Azure. Ok, random I hear you say, there was a reason for this! You can read about the why in my healthtech article series.

I must admit I didn’t plan to write a blog on this topic however as part of the configuration for my upcoming session at #scottishsummit2024 I stumbled across this nuance and thought it worth sharing, which after all is caring!

So humour me as all the best things in life are slightly niche, so let’s dive into the world of Adaptive cards and Copilot Extensibility!

So what is a Copilot “action” I hear you say? This is a very good question probably a good place to start given all the various Copilot nomenclature, of which even I don’t fully understand.

An Action is essentially capability in Copilot that is wrapped with additional information to describe its purpose, allowing the Copilot orchestrator to use its clever generative AI to generate contextual questions.

These questions are required to fill the inputs needed to perform the action. Therefore, you don’t need to manually author question nodes to gather all inputs needed, such as the inputs required to run of example a Power Automate workflow. Inputs are handled during your action runtime, these are then AI filled using “slot filling”.

Once triggered an action will then generate a contextual response to a user’s query, using the results of the connected workflow. Alternatively, you can explicitly author a response for the action.. Here’s the clever bit; you can use adaptive cards to display a response, the adaptive cards can be populated with the inputs & output variables. The AI magic captures inputs effortlessly, making the user experience smoother than a freshly ironed pilots shirt.

However, this comes with a catch as the output customisation is as limited as a Joshua’s diet in the run up tonight’s title match. For example parsing a Power Automate response into a table, just like I needed for my observations… Forget about it we are a string only world here!

So in summary the AI capabilities of Copilot actions enhance the user experience by providing intelligent suggestions and auto-completions. However, once the data is captured, and a response is generated from a connector the ability to manipulate and present it in a customised manner is restricted. This limitation can be a significant drawback when dealing with complex data structures or when specific formatting is required, other than a simple string response.

So now we have the detail, it’s fight night, the AI-powered Actions based wizard in the red corner and in the blue corner, Topics, the meticulous record handler. Who will win? It depends on what you need. If you want quick, AI-driven interactions, Copilot Actions is your champ. But if you need detailed, complex data handling, Topics takes the crown.

Let’s break it down:

• Ease of Use: Copilot Actions wins with its AI-driven input capture and user-friendly interface.
• Customization: Topics takes the lead with its ability to handle complex data and provide detailed responses.
• Integration: Copilot Actions struggles with integrating complex data sources, while Topics excels in this area.
• Flexibility: Topics offers more flexibility in terms of data manipulation and presentation.

So now you know creating adaptive cards in Copilots is an art and a science. Here are some best practices to ensure your cards are as effective as they are beautiful:

• Be Clear: Use concise and straightforward language. Make sure users understand what is required of them. Clear instructions reduce the likelihood of errors and improve the overall user experience.
• Responsive Design: Ensure your cards look good on all devices. Test them on different screen sizes. A responsive design ensures that your adaptive cards are accessible and functional across various platforms.
• Consistent Layout: Maintain a consistent design language across your cards. This helps users know what to expect. Consistency in design improves usability and creates a cohesive user experience.
• Accessibility: Make sure your cards are accessible to all users, including those with disabilities. Use appropriate contrast and provide text alternatives for images. Accessibility is crucial for ensuring that all users can interact with your adaptive cards effectively.
• Avoid Overloading: Don’t cram too much information into one card. If necessary, break it down into multiple cards. Overloading a card with information can overwhelm users and reduce the effectiveness of the card.