Building low-code AI applications

Who this is for and what you will get

This guide is for people who want to turn a spreadsheet-style process into an app, an automated flow or an AI agent without writing much code, such as teachers, finance staff and operations teams. It also helps developers see where AI fits into low-code tools. By the end you will know what each AI-related part of Power Platform does, and you will be able to use Copilot to produce a first draft of an app, a flow or an agent.^[1]

What low-code means and what Power Platform includes

Low-code development platforms let people build apps and solutions with little or no code. They usually provide a visual environment in which you drag and drop components to assemble an app.^[1]

Microsoft's Power Platform has five key products: Power Apps, Power Automate, Power BI, Power Pages and Copilot Studio.^[1]

What Copilot and AI Builder each do

Copilot is an AI assistant. You describe your requirements in a series of conversational steps, using natural language, and it helps you build a Power Platform solution. Copilot is available in all five Power Platform products.^[1]

AI Builder is a low-code AI capability in Power Platform. It uses AI models to automate processes and predict outcomes. It can connect to data in Dataverse as well as cloud sources such as SharePoint, OneDrive and Azure, and it is available in Power Apps and Power Automate.^[1]

Build an app with Copilot in Power Apps

On the Power Apps home screen, describe the app in the text area. The example prompt is “I want to build an app to track and manage student assignments”. Click Send to pass the prompt to Copilot.^[1] Copilot generates a responsive canvas app from your description and suggests a Dataverse table with the fields you need and some sample data.^[1]

The steps are:

1. Describe the app on the Power Apps home screen and send the prompt.^[1]
2. Review the table and fields that Copilot suggests, and refine them through conversation.^[1]
3. Add fields or screens with a prompt. Examples include “I want to add a column to store student email” and “I want to add a screen to send emails to students”.^[1]
4. When the table is ready, click Create app. Save the app with Save, then use Share to invite others by entering their email addresses.^[1]

Create a table in Dataverse with Copilot

Dataverse is the underlying data platform for Power Platform. It is a fully managed service that stores data securely in the Microsoft Cloud, provisioned inside your Power Platform environment. It includes data governance features such as data classification, data lineage and fine-grained access control.^[1]

Tables work like multiple worksheets in one Excel workbook. Role-based security lets you control who can access the data in a table and how they can access it.^[1]

To build a table, select Tables in the left navigation bar, click Describe the new Table, describe the table in the text area, for example “I want to create a table to store invoice information”, and click Send. After you review the suggestion and refine it through conversation, click Create.^[1]

AI Builder prebuilt and custom models

AI Builder offers two types of AI model. Prebuilt AI Models are trained by Microsoft and ready to use in Power Platform. They include Key Phrase Extraction, Language Detection, Sentiment Analysis, Business Card Reader, Text Recognition, Object Detection, Document processing and Invoice Processing.^[1] Because they are trained already, you do not have to gather data and train and publish your own model before adding intelligence to apps and flows.^[1]

Custom AI Models let you bring your own model into AI Builder and train it with your own data. Once it is in AI Builder, it works like other AI Builder custom models in Power Apps and Power Automate. There are limitations for bringing your own model, which are documented separately.^[1]

Example: an invoice flow in Power Automate

The example scenario is a finance team that wants to pull invoice details out of emails and store them in a Dataverse table, with a status on each record.

1. Open the Power Automate home screen and describe the flow, for example “Process an invoice when it arrives in my mailbox”. Send the prompt.^[1]
2. Copilot suggests the actions. Click Next to go through them, set up the connections the flow needs, and click Create flow.^[1]
3. Update the trigger. Set Folder to Inbox, open Show advanced options, and set Only with Attachments to Yes so that the flow runs only for emails with an attachment.^[1]
4. Remove the actions the example does not use: HTML to text, Compose, Compose 2, Compose 3, Compose 4 and Condition.^[1]
5. Click Add an action, search for Dataverse, and select Add a new row.^[1]
6. In the Extract Information from invoices action, set Invoice File to the email's Attachment Content.^[1]
7. Select the table you created. Use dynamic content from earlier steps to fill ID, Amount, Date and Name. Set Status to Pending, and set Supplier Email to the From value from the trigger.^[1]
8. Save the flow, then send an email with an invoice to the trigger folder to test it.^[1]

Create text with GPT

The Create Text with GPT model in AI Builder generates text from a prompt. It is powered by the Microsoft Azure OpenAI Service.^[1] It can draft emails and product descriptions, and it can support chatbots and customer service apps that help agents respond to customer inquiries.^[1]

Copilot Studio: building AI agents

Copilot Studio, formerly Power Virtual Agents, is the low-code member of Power Platform for building AI agents. These agents can answer questions, take actions and automate tasks. You describe what the agent should do in natural language, and Copilot Studio helps scaffold its instructions, knowledge and actions.^[1]

Its main capabilities include:

• Generative answers: agents can draw on knowledge sources such as public websites, SharePoint, OneDrive, Dataverse, uploaded files, or enterprise data reached through connectors.^[1]
• Generative orchestration: instead of rigid trigger phrases, the agent uses AI to understand a request and decide which knowledge, topics and actions to combine.^[1]
• Actions and connectors: agents can use more than 1,500 prebuilt Power Platform connectors, Power Automate flows, custom REST APIs, prompts, or Model Context Protocol (MCP) servers.^[1]
• Autonomous agents: agents can be triggered by events such as a new email, a new Dataverse record or an uploaded file, and then work in the background.^[1]
• Multi-agent orchestration: agents can call other agents, including agents published to Microsoft 365 Copilot and agents built in Microsoft Foundry.^[1]
• Model choice: besides the built-in models, you can bring models from the Microsoft Foundry model catalog.^[1]
• Publishing: agents can be published to Microsoft Teams, Microsoft 365 Copilot, a website or a custom app, with security, authentication and analytics managed through the Power Platform admin experience.^[1]

Limits and common questions

Output from a low-code generator is a starting point, not a finished product. Review the apps, tables and flows that Copilot generates and change them to fit your needs.^[1]

Bringing your own model comes with limitations, so check the official limitations before relying on one.^[1] The invoice example depends on the Invoice Processing model in AI Builder. The fields the flow writes have to match the columns of the table you created, as the field mapping in the example shows; error cases are not covered here.^[1]

Next steps

To keep practicing, try the two homework tasks: use Copilot in Power Automate to build a flow that emails students automatically when they submit an assignment, and a flow that emails a supplier when an invoice status changes.^[1]

The Power Platform Copilot Prompt Library on GitHub contains prompts for building apps and flows with Copilot, and it is a good place to see how to phrase requests.^[1] For more on Copilot Studio, read the official Microsoft Copilot Studio documentation.^[1]

Adapted from lesson 10 of Microsoft's open-source course *Generative AI for Beginners*^[1], released under the MIT License.