AI agents in Power Apps: what the agent does on its own and what a person approves
Microsoft has built AI directly into Power Apps business apps. Copilot answers questions about the app's data, pre-built skills fill in forms and write summaries, and agents take actions on the user's behalf. To keep agents in check there's Agent Feed: the agent handles minor actions on its own and sends important ones to a person for approval.
What Microsoft announced
The update has four parts, each with its own availability status.
- Microsoft 365 Copilot in apps. Users ask questions right inside the app and get answers grounded not only in the records on screen but in broader work data through Microsoft Work IQ. Generally available in model-driven apps, in public preview for canvas apps.
- App skills. Built-in AI capabilities for data entry, exploration, visualization, and summarization. Generally available.
- Agent Feed. A feed where users see what the agent is doing and approve the important actions. Generally available in model-driven apps since May 4, 2026.
- App-based forms in Copilot Chat. Power Apps forms available directly in Copilot Chat. Generally available since July 2026.
Four skills that work out of the box
Data entry
AI turns an email or a document into filled-in form fields. Instead of copying details by hand, the user reviews a ready record.
Exploration
Plain-language queries instead of filters: "show me open high-priority tickets this week."
Visualization
Reshapes the data view to match the user's request, without an admin setting up new views.
Summarization
Condenses a long activity history on a record: what happened, who did what, and what's still open.
These skills can be exposed to agents through the Power Apps MCP server (in preview). Microsoft's example is a recruiting app: an agent built on it accesses the same records, enforces the same rules, and operates under the same controls as a person using the app. In other words, you don't have to rewrite business logic separately for AI.
Agent Feed: how a person stays in control
The first question executives ask about AI agents is "what if it does the wrong thing?" Agent Feed answers that at the product level.
The app maker sets the threshold
They decide which actions the agent takes on its own and which need approval.
Low-risk actions run in the background
Things like filling in fields or updating a status, the agent completes quietly.
High-impact actions go for approval
Actions such as sending an email to a customer appear in Agent Feed as an explicit approval request.
Decisions are made in context
The user sees side-by-side comparisons, deep links to records, and agent performance signals right in the app, not in a separate monitoring tool.
Microsoft's example
At an insurer, an agent extracts data from emails and pre-fills the claim record. The claims adjuster reviews the proposed changes in Agent Feed and approves them before the data goes into the system. The agent takes the routine work; the person keeps the responsibility.
Security and access rights
- Copilot in the app respects the same permissions, roles, and business logic the app already enforces. Users can't see records through Copilot that they can't access in the app.
- Admins enable Copilot at the tenant level, and makers configure it in a few clicks.
- Approval thresholds for agents are set by the app maker, not by the agent.
Setting up roles and approval thresholds is the part Progresia handles during a pilot: together with the process owners, we decide which actions the agent takes on its own and which need approval.
For a broader look at how Copilot handles company data, see Copilot security and confidential data. For rules on using AI across the company, see our article on AI governance.
Where it pays off most
Built-in AI works best where lots of similar records come in from emails and documents, but a person still makes the call.
- Requests and cases. The agent processes incoming emails, creates records, and drafts a reply; a manager approves.
- Contracts and documents. Details, amounts, and dates from a document land in the form automatically. More on document management alongside CRM in our ECM article.
- Recruiting and HR. Parsing CVs, updating candidate statuses, summarizing interviews.
- Projects and tasks. Project summaries, finding overdue tasks with a plain-language query, updating statuses from emails.
Progresia builds Power Platform apps for exactly these scenarios, including our own project management solution Progresia.TM, and can add AI skills to the apps you already run.
Where to start
You can go through these steps with Progresia, from choosing the app to a report on the pilot's results.
Pick one app with routine data entry
The best candidate is a model-driven app where staff copy data from emails or documents by hand.
Review roles and access rights
Copilot inherits the app's permissions. If everyone can see everything, AI will only make that more visible.
Decide what the agent can do on its own
List the actions and, for each one, decide: run in the background or require approval in Agent Feed.
Run a pilot and measure the impact
Within 3–4 weeks you'll see how much time is saved on data entry and how many agent suggestions are accepted without edits.
If you need a standalone AI agent for customers or employees outside the app, look at Copilot Studio. To automate document processing in Power Automate flows, there's AI Builder.
FAQ
Is Copilot available in all Power Apps?
Microsoft 365 Copilot is generally available in model-driven apps and in public preview for canvas apps. Admins enable it at the tenant level.
What is Agent Feed?
It's a feed in model-driven apps where users see what an AI agent is doing and approve its important actions. The app maker decides which actions the agent takes on its own and which require approval. It's been generally available since May 4, 2026.
Can an agent see data the user doesn't have access to?
No. Copilot and agents in the app operate within the same permissions, roles, and business logic already configured in Power Apps.
How is this different from Copilot Studio?
Copilot Studio is a tool for building standalone agents and bots. AI in Power Apps is built into an existing business app and uses its data, forms, and rules. App skills can be exposed to agents, including ones built in Copilot Studio, through the Power Apps MCP server.
Add an AI agent to your Power Apps app
Progresia will help you choose a process for a pilot, set up Copilot and app skills, define approval rules in Agent Feed, and measure the impact on your real data.