Copilot AI updates across the Power Platform | Automate, apps, pages & virtual agents

Posted by

This post has been republished via RSS; it originally appeared at: New blog articles in Microsoft Community Hub.

Use AI Large Language Models with Microsoft’s Power Platform to create automated workflows, apps, web pages and bots — without knowing how to write code. AI and Copilot help build fully functional experiences. Generate workflows using only natural language prompts in Power Automate, create apps in seconds in Power Apps, build professional websites with Power Pages, and use the new Boost Conversations capability with GPT to create FAQ bots with Power Virtual Agents. 



Stephen Siciliano, Vice President of Microsoft Power Automate, joins Jeremy Chapman to tour the latest Power Platform updates.


Build workflows using only natural language prompts. 

1 workflow.png

Ask Copilot to automatically create approval flows with conditions and emails. Check out new Copilot capabilities with Power Automate.


Copilot moves data from Excel files into Dataverse.

2 Excel.png

Build an app to manage and track invoices. Copilot creates column headings & data types automatically. See it here.


Add a bot to your site — instantly. 

3 bots.png

Use the new Boost Conversations capability along with GPT to quickly create FAQ bots with Power Virtual Agents. Click to watch.


Watch our video here.


00:00 — Introduction 

01:03 — New AI and Copilot experiences 

02:42 — Use Copilot to build and automate workflows 

06:07 — Copilot and AI Builder for document analysis 

07:19 — Use Copilot to move data from Excel while building apps 

08:55 — Customize an existing app 

09:53 — Use Copilot with Power Pages 

11:06 — Add GPT powered bots to your site 

13:37 — Wrap up


Link References: 

Get started at 

Watch our 3-part series on how to add Copilot and GPT to your apps 

To build an app using our workbook, start in the Power Apps Studio at


Unfamiliar with Microsoft Mechanics? 

As Microsoft’s official video series for IT, you can watch and share valuable content and demos of current and upcoming tech from the people who build it at Microsoft. 


Keep getting this insider knowledge, join us on social: 

Video Transcript:

- Can AI help build your automation workflows, your apps, webpages and bots without you knowing how to write a line of code? Well, let’s find out. Today, we’re looking at how the fusion of AI large language models with Microsoft’s Power Platform make things easier than ever, from AI-assisted workflow generation in Power Automate, how you can use Copilot with Power Apps to build apps from data in seconds, where this also applies to generate your websites with Power Pages, and of course, how it translates to chat experiences with Power Virtual Agents, and joining us again to tour the latest Power Platform updates is Stephen Siciliano, welcome back. Yeah, thanks for having me here again.


- Thank you, so every time we have you and Charles Lamanna on the show, just when we think we’ve reached the pinnacle of low code app and logic development, things get even better. So this time last year, we were generating an app from a sketch on a piece of paper and now we’ve shown recently, we’re able to go from just an idea to a finished app complete with backend logic in the time it takes us to describe what we want using Copilot, which is leveraging large language models like GPT. In fact, most of the Power Platform components have their own Copilot experiences now, or coming soon, so how does this change things?


- In many ways, it’s a great way to explore all the capabilities the Power Platform has to offer, while saving you a lot of time and effort. The AI and Copilot experiences across the Power Platform can increasingly assist you with whatever you may be trying to do without you needing to be a code or process expert. Now if you have an idea, it can come to life in the best way possible. Not only will the AI build out an experience based on its knowledge and/or any data you present it with, you can take advantage of suggestions to customize your app, or you can ask for what you need in terms of logic and workflow and have it generated for you. This can work whether you want to automate a repetitive process or build an app or website experience.


- And something to point out here is what you can end up with using Copilot and AI isn’t just a basic experience, these are fully-functional experiences.


- Yeah, we’re giving you the power of AI to reach the experiences and outcomes faster and this can help enrich what you’re trying to do. In fact, let me give you a taste of how you can transform a common scenario like invoice processing and go from process automation to integrated and interactive internal and external experiences. Most of us can relate to data entry and approval workflows. Here, you’re looking at an Excel spreadsheet that a finance team might use to process invoices, you know, the ones that arrive as an email attachment or even in snail mail. Somebody manually parses this information, sends it to someone else for approval, enters the information into an Excel spreadsheet, then the payment process can finally happen. Of course, this is something that Power Automate has been able to do for awhile now, but you needed a fair amount of knowledge to make it work. Now I can use Copilot to build out my basic workflow using a simple natural language prompt. I’ll type, “When an email with an invoice file arrives, start and wait for an approval.” And it generates this nice preview flow in just a few seconds to manage intake and approval. The nice thing about this is that if I want to iterate on the flow a bit more, I can ask Copilot to generate another variant if I want it to. For example, I’m going to add more details after when an email arrives. I’ll add “then for each file” to my prompt, and you’ll see it adds the “for each step.” Once I feel good about the flow, I can click next and that will ask me to sign in to connect the required services to run my flow. And what you’re seeing here is the new, modern canvas with zooming controls so I can easily navigate everything on the screen and focus in on what I want. In fact, I’ll dig into this trigger card to see what’s happening and you’ll see that as an email with an attachment arrives, it will trigger the flow, but from the red message, it looks like it might be missing from the parameters. We’ll get to that in a second. But first, I want to make sure my flow is only scoped to invoice attachments. Right now it’ll try to process everything so I can ask Copilot to only trigger on invoice-related emails. I’ll paste in, “trigger the flow only if the incoming email title contains invoice” as a new prompt. Keep your eye on the email parameters on the left. Copilot automatically added that parameter filter to the trigger and now the flow is more accurate. I didn’t need to know how to add that filter. I’ll give Copilot a like for that. But let’s keep going and fix up that approval step we saw before. I’ll get that step back into view. There it is. I’ll specify approve/reject, first to respond and the red text is gone, but I need to add a few more parameters. I can do that with Copilot and use variables too so I’ll paste in another prompt. Update the title of approval to be “New invoice needs approval, assign the approval to my manager.” Now, let’s check how it did. Title looks good. Now let’s check if it got the assigned to parameter right. When I hover over mail, you’ll see it’s pulling my manager info automatically. Remember in our manual process, after the invoice was approved, we updated our Excel file? Let’s automate that step too. I’ll paste in another prompt. If approved, write the invoice data to Excel Online. Now let’s see what it did. It added that as an action, but I want to specify it’s only for approved invoices so I just need ask Copilot to move it into the right condition for true. Now let’s take a look and you’ll see it’s there. I just need to tell it which Excel file I want, so to save a little time, I’ve already defined the file attributes we need. Now I can select a few more parameters for our column and it’s done. From here, I just need to save my flow. And one more thing Copilot is really good at is describing what a flow does. All you need to do ask. So, I’ll ask it, “What does this flow do?” And Copilot gives me a nice summary that anyone can understand.


- This is going to save a lot people a lot of time in terms of building the automation and the workflows, but what about the invoice content itself? Can we extract the data from that?


- Yeah, so we’ll be able to use AI Builder with its forms processing to get all of that information right into our workflow. This is something that we’ve had for a few years now, so I won’t walk through the entire process. By the way, everything I’ve shown now is something you can do today. Now, I want to give you a first look at what’s coming with Copilot and AI Builder. First, you’ll see this step to extract the data from incoming invoices. And this is the cool part. Here’s where we’re using the large language model to extract approval justification and automatically generate it for you in an email. In fact, let’s run this so that you can see what it does. This is the approver’s inbox. An approval request just arrived and when I open it, you’ll see that the LLM in this case authored a comprehensive email with justification context, and of course, the AI Builder extracted all those invoice details and I can approve it right here from inside Outlook. I’m going to switch desktops quickly because this all happens pretty fast. You’ll recall that if approved, our flow needs enter the new row into Excel. There it is just in time.


- This is pretty impressive stuff. I know the team has also been working on using GPT to make the app-building experience a lot faster. In fact, we just published a three-part series with Emma Cooper on this topic specifically that you can check out at


- Yeah, there’s a lot of exciting new stuff on the app side too so we can use that same Excel workbook and walk through that scenario. Let’s say our finance team wants to build an app to help manage and track the invoicing process instead of just using an Excel spreadsheet. Now it’s super easy and this is something you can do today. I’m going to import this Excel file, the same one we saw before, directly into a Power App. Notice this spreadsheet doesn’t have a header row and that’s okay, the AI will take care of that. To build an app using our workbook, I just need to start in the Power Apps Studio at Now, I’ll click start with data and upload our Excel file by dragging and dropping it onto the canvas. Copilot processes that data for a moment and moves the data from an individual Excel spreadsheet into a Dataverse environment. This way it’s also a more scalable central place to manage invoice data where everyone can work simultaneously on the same underlying shared dataset. And here’s why I mentioned there was no header row. The AI is able to process all of the data in the spreadsheet and determine the column headings and data types automatically. You can open these up and edit them if it doesn’t get everything right. I’ll open the status to check this one, but these look perfect to me. I can also check the table properties and these also look good, it even added a description. From there, I can create the app. That brings me to the Power Apps Studio authoring canvas and the app looks pretty good already. You’ll also see that our Dataverse table is already linked in the tree view. This is a working app. I can hit play and browse though the app in full screen, selecting any of these line items on the left and seeing their corresponding invoice data on the right.


- I really like that it also added the ability to search and create a new record, so where else can Copilot help as we build out our app?


- Yeah, let me show you how Copilot can help beyond these initial creation steps. So we’re going to customize this existing app. From the edit view, I’ll choose the Copilot option from the top bar, and from there, I can make changes to my app using prompts. I’ll paste in add a submit invoice button to the selected container and Copilot places the button on the app canvas. It even moved my edit and delete icons automatically to make space for the button. From here, I could edit this control to upload an invoice document and run a flow similar to what we just built.


- All right, so in just a couple of minutes, you’ve also managed to automated a manual workflow, you created a new app experience with a more robust Dataverse backend, and combined your app with a robotic process automation using generative AI and Copilot in a bunch of places that’s going to save a lot of time.


- Yeah, and these were just a couple of ways that Copilot will be able to help out all of your existing processes. And we can use Copilot and AI with your public-facing sites and experiences too. Using Copilot with Power Pages, you can get started building professional websites even faster than before. Let me show you how. I’m in the Power Pages design studio and I’m editing an existing site. You can see I already have a few pages built out. Just like we saw with Power Apps, you can open up Copilot with this control on the top right, which opens up the right pane where I can add prompts. In my case, I want to add a page and I’ll type “Create a page to allow our partners and suppliers to get in touch with us.” And with that simple prompt, Copilot generates HTML and CSS for each section of the page, as well as contextually relevant images. And this is looking pretty good, but I want to expand the subtext and describe how we work with our partners and suppliers. So I’ll select it and bring up the component-level Copilot experience. Now I’ll paste in my prompt and you’ll see it generates a nice paragraph to start with. Of course, I can edit this further if I want to. I’m going to stop there, but you can even use Copilot to generate forms too like a contact us form, but we’ll save that for a different show.


- I really like here how this uses the large language model is actually able to both write the code as well as text-based content for the page. Switching gears though, one thing has been apparent the last couple of months with tools like ChatGPT, it’s that the chat and conversation experience can be more intelligent than ever, so how would you add something like that to your website or Power Page?


- Yeah, that’s very easy to do with Power Virtual Agents. Anyone who visits your website could have additional questions where they might not be able to find the right information if it’s buried deep in your website. Let me show you how we solve for that. I can use the new integration between Power Pages and Power Virtual Agents to add a bot to my site instantly. With one click here in set up, a bot is created for me and added to my site. Now I can open Power Virtual Agents, make a few changes and test it out before publishing. The bot is already configured to provide generative answers to user questions powered by OpenAI and GPT, just by pointing to our website. You have the option to add more sites here, including internal sites, such as SharePoint, and everything respects the site and content permissions you have in place. Coming soon you’ll be able to upload documents and choose the tone for dynamically generated responses and there are additional controls for content moderation. As soon as this bot is created, it’s instantly useful and is able to have rich, multi-turn conversations. Let’s open the test canvas and give it a try. If I ask, “How can I book travel?”, you can see the bot has searched across our website and then summarized a response for me from a knowledge-based article. Each answer also includes one or more attribution links, ensuring site visitors can always find more information and easily review the source of the answer. Then, if I ask a follow up question, the context of the previous conversation is taken into account to make sure I get the answer I need. I could now use Copilot in Power Virtual Agents to create topics to handle key use cases for my bot just by describing the topic I want to build, using natural language, the entire topic is instantly generated for me. I can then use Copilot alongside the powerful authoring canvas to iterate further on my topic. Copilot in Power Virtual Agents and the new authoring canvas are both out of preview now too, so let’s add this to our website. Back in Power Pages, I can publish the bot and make it available to our customers in one click. Now if I open our website as a user, I can open the bot and test to ensure everything is working. This is another example of how we’re leveraging AI to reduce what might have been a multi-month effort to develop a bot, down to just a few minutes. And to be clear, I’m really just scratching the surface with what AI can do for you using the Power Platform.


- There’s really so much to try out and I just can’t wait to get started.


- Yeah, there’s a lot there today and a lot more on the way. It’ll save you a ton of time. And please, give us your feedback. Just let us know what you’d like to see. Go to to get started.


- Thanks so much for joining us today, Stephen. All of the end-to-end updates on AI in the Power Platform. Of course, keep checking back to Microsoft Mechanics for all the latest updates. Subscribe to our channel if you haven’t already, and as always, thank you for watching.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.