Today, we’re introducing closed-loop learning for agents connected to the Power Apps MCP server, starting with the data entry tool. Every correction a user makes through the Agent feed persists as structured memory. On future runs, the agent retrieves that memory and applies it. Over time, those corrections consolidate into organization-wide patterns the agent applies across tasks. The feedback loop runs automatically in production. Nothing to configure, no data pipelines to build.

How closed-loop learning works

Say a finance team has an agent to process vendor invoices. The agent uses the data entry tool in the Power Apps MCP server to extract supplier names, addresses, and totals from PDFs. Most fields come through correctly, and the agent correctly fills in “UK” from the document, but the organization requires it to be normalized to “United Kingdom.” A user corrects it in the Agent feed. After a handful of corrections like this, the pattern becomes the agent’s memory. The next invoice with “UK” gets “United Kingdom” automatically. Over time, the system learns the pattern: abbreviations like “USA” or “DE” get expanded too. The agent also learns by example: it retrieves similar memories to the current request and applies the same processing steps used previously.   

Think of it like a driving instructor in the passenger seat. A student merges without checking the blind spot; the instructor corrects them, and the student remembers it for next time, because they wrote it down. That’s the first layer: memory-based optimization, the mechanism that captures individual corrections and applies them on future runs.

Corrections get you through the driving test. Patterns make you a confident driver. That’s the second layer: Genetic-Pareto optimization, an evolutionary prompt optimization technique. Memory-based optimization generalizes from the corrections it remembers. Genetic-Pareto goes further, distilling those corrections into rules compiled into the agent’s instructions, so the principle becomes the agent’s default behavior rather than a pattern it must retrieve each time. Both layers are live today for the data entry tool.

Both mechanisms run with the rigor of a production machine learning pipeline. Unlike the memory features built into most AI assistants, which personalize the experience for an individual user, closed-loop learning improves task accuracy for everyone using your organization’s agent. The learning stays scoped to your business process and governed by your tenant.

Inside the research

The approach

Today’s techniques for adapting Large Language Models don’t close the feedback loop between deployment and improvement. Retrieval-augmented generation (RAG) surfaces documents at inference time but doesn’t learn from outcomes. Fine-tuning adapts the model but requires a training and deployment cycle for every update. Open-loop prompt optimization improves a fixed prompt offline but doesn’t incorporate live feedback.

Closed-loop learning closes the gap through two complementary mechanisms. Memory-based optimization captures user corrections as structured memories retrieved at inference time, giving the agent immediate recall. Genetic-Pareto optimization periodically distills those memories into generalized rules using evolutionary prompt optimization, so the agent applies what it’s learned to cases it hasn’t seen. The agent acts, a user corrects, the system learns, the next action improves.

Closed-loop learning builds on two open-source projects. Memory-based optimization takes inspiration from Memento, though our implementation has evolved significantly to fit the Power Apps MCP server’s architecture. Genetic-Pareto optimization is implemented via GEPA, integrated through DSPy, Stanford’s framework for programmatic LLM optimization.

The evaluation

The benchmark had to reflect enterprise data entry in practice. One of our early customers, the UK Electoral Commission, processes thousands of invoices annually from suppliers based in the UK, US, Ireland, and other countries. Each invoice demands structured extraction of supplier name, address, country, and total expenditure, with conventions that vary by supplier country and source document. The corrections the agent must learn are the organization-specific conventions that turn a correct extraction into a usable one.

We evaluated four configurations on a dataset of 100 invoices across 10 independent runs, reporting the average score across runs with 95% confidence intervals. The primary metric is F1 score, balancing precision (are the predictions correct?) with recall (is the model predicting all the expected fields?). Our quality bar is strict: did the user save exactly what the agent predicted, with zero edits? By that measure, a prediction of “UK” when the organization’s records use “United Kingdom” is a quality gap, because the user still had to correct it.

The results

These results are from pre-rollout offline simulation on UK Electoral Commission dataset. The data entry tool extracts content from invoices reliably. The gap is between that raw extraction and the format and conventions the organization expects. Closed-loop learning bridges that gap, calibrating the agent to the business. Across 4277 field instances, closed-loop learning decreased the share of fields users had to edit from 64% to 48%: 1045 fewer fields requiring manual correction.

In 10 independent runs, Genetic-Pareto improved F1 score over the baseline from 66.4% to 74.6%, an 8.2 percentage point improvement.

The optimized prompt outperformed the baseline in all 10 runs, with non-overlapping confidence intervals and a positive confidence interval for the difference.

How Genetic-Pareto calibrated the agent

The F1 number tells part of the story. The baseline typically extracts the right information from the invoice; the gap is in how the business needs that information presented. A supplier name shown as a brand tag rather than the legal entity. An address combined into one line rather than splitting across the fields the organization expects. A total reflecting the ex-VAT amount rather than the gross invoice total.

In one sample run, Genetic-Pareto addressed 76 of 583 gaps in the baseline (a 13% reduction):

These are taste and preference gaps. Genetic-Pareto closes them by calibrating the agent to how the organization structures its data. The gains were most pronounced when specific corrections mapped to a generalized principle. Country accuracy alone jumped from 11% to 78% after Genetic-Pareto learned to expand abbreviated country names. A broader evaluation of additional customers is in progress to validate and expand these findings.

Closing the loop

This evaluation and rollout loop a data science team typically runs happens automatically inside your tenant, through the Power Apps MCP server. The system generates a candidate prompt, runs a shadow experiment (each request uses the current prompt for the user-facing result while the same input is scored in parallel on the candidate), and uses statistical validation (hypothesis testing and power analysis) to decide whether the candidate is measurably better. When it clears the threshold, the candidate automatically becomes the new baseline, and every subsequent request runs on the improved version.

What comes next

Closed-loop learning will extend across more agentic workflows on the Power Apps MCP server over the coming weeks.

The gap between passing the driving test and feeling confident behind the wheel is experience. Closed-loop learning gives agents that experience.

Get started: Add the data entry tool to your agent on the Power Apps MCP server and give your agent a memory.

The post Power Apps MCP server introduces closed-loop learning for enterprise agents appeared first on Microsoft Power Platform Blog.

Save and reuse charts with confidence

Whether you’re tracking inventory trends, monitoring customer orders, or reviewing sales performance, saving visualizations means your analysis doesn’t disappear when you move on. With just a click, you can preserve the charts Copilot generates so you and your team can revisit the same insights, apply them across records, and stay aligned around the same data story.

Legacy charts get usability enhancements

This update also brings a fresh look to all charts not just those created by AI. Whether created in the chart designer manually or with Copilot, your visualizations now share the same clean, modern styling that enhances readability and consistency across your apps. By default, this experience is enabled for everyone, but admins can turn it off in environments where custom chart styling is in use.

This update is rolling out on-by-default for all customers that have Visualize with Copilot enabled, so you can start saving and styling visualizations right away—no setup required. It’s an enhancement with big impact: turning answers into persistent, shareable insights that keep your team aligned on the same data.

Visit our documentation to learn more.

The post Visualize with Copilot now allows you to save your chart appeared first on Microsoft Power Platform Blog.

Introducing Visualize with Copilot

Data exploration is a task that can involve manual steps like exporting data to a spreadsheet, creating reports, and fiddling with filters. But, with Power Apps’ data exploration agent, you can now visualize your view as a chart to spot trends, patterns and relations your data with just one click. Every visualization comes with an AI-generated title and clear reasoning, giving you full context of the thought process.

Effortlessly refine AI-generated chart to fit your needs

The data exploration agent works for you – that’s why we’ve made it easier than ever to refine and personalize the AI-generated visualization with just a few clicks. You can change the chart type or adjust columns or modify aggregation to tailor the chart AI-generated according to your preferences.

Interact with visuals that stay in sync with your data

Data exploration is most powerful when the visualization is interactive, supporting drill downs. With the AI-generated visualization dynamically linked to the grid, you can interact with your data without any friction. Select a chart segment to filter the grid, refine your focus, and analyze data in real time. Together with natural language filtering, just type what you need, and watch as the data exploration agent filters your grid and updates your chart turning raw data into meaningful insights.

Use Visualize with Copilot in your model-driven apps by updating the following two feature settings in Power Platform Admin Center.

1. Under Natural language grid and view search, set Enable this feature to All users immediately.
2. Set Allow AI to generate charts to visualize the data in a view to On.

Get started with Visualize

Visit our documentation to learn more and get started today. This feature, deploying with build 9.2.25013, is currently rolling out and is expected in all regions worldwide by the end of this month. We’d love to hear your feedback on this experience, please let us know in our community forum post.

The post Introducing Visualize with Copilot (preview) appeared first on Microsoft Power Platform Blog.

Power BI quick reports in Power Apps represent a seamless integration of Power Apps, Power BI and Dataverse into a single experience to enable every business user turn data into insights inside business apps. Of the data visualization tools available in Power Apps, Power BI quick reports are the most powerful, as they enable everyone to create visually appealing, meaningful, interactive reports based on a view with just one click.

Generate insights in apps without losing context of your work

When exploring data on a table’s home page grid in a model-driven app, select Visualize this view command.

This generates a Power BI report based on the underlying state of the grid.  Any filter or column modifications are added to the report to preserve the context of your data exploration process.

Columns in the view are available in the Your data pane, with the same localized display names as in the underlying grid column header. There are several options available to personalize the visualizations in ways that are meaningful to you. Here’s the post on Power BI blog that details the ways you can interact with the visual.

Save your report to the Power BI service

You can now save the report into the Power BI service easily by giving it a name and choosing a workspace. This persists the report and the related dataset in Power BI, which you can access just like any other Power BI report, and even embed it as a dashboard or on a form.

Getting started

You can enable this capability for a model-driven app through the app’s settings area, available in the modern app designer. Enabling this capability will display Visualize this view command on the grid pages of the tables in the model-driven app.

Review documentation to learn more about this capability and the options available to interact with the Power BI visual.

Let us know your feedback in comments or on our Power Apps community forum post.

The post Announcing general availability of visualizing views in Power Apps with Power BI appeared first on Microsoft Power Platform Blog.

Explore data through advanced filters on any table in the app

Using the global search bar in the header, you can now choose any table in the app to perform a structured search. Choosing a table navigates you to the familiar grid page, with filter options immediately available to work with.

View selector enhancements to help you create and find views easily

After you have edited the filters and included the right columns in the view, you can find the options to create a personal view within the view selector, along with other options to manage the view list.

As you create more views, you can browse them easily in the view selector, with personal views annotated with an icon. You can also search for a view by name so you can be more productive in navigating between different pivots of data.

Personalization options to see the most relevant views

If you have created a lot of personal views over time, finding the right view quickly can be difficult. With two new powerful personalization options available with modern advanced find, you can make the view list your own, so the most relevant views are easily accessible.

Hiding a view in the view list

You can hide one or more views in the view list to ensure only the most relevant views show up in the view selector. Views are hidden only for you across all devices for that table in the environment.

Changing the order of views

You can make the list of views appear in one of three preset orders. If you have tables in which you prefer to use system views over personal views, you can change the order of views in the view selector across all devices for that table in the environment.

View management options to collaborate with views

You can also collaborate with your team members by sharing personal views just like you would share any other record.

Administrators can enable modern advanced find through a feature setting in Power Platform Admin Center.

Review the documentation to learn more about all the enhanced view management capabilities that are part of this feature setting. Please share your feedback on our Power Apps community forum post.

The post Modern advanced find with enhanced view management in model-driven apps appeared first on Microsoft Power Platform Blog.

Power BI quick reports add to the flexibility of data visualization options already available in Power Apps. While charts and custom components allow you to create specific, custom visualizations in Power Apps, Power BI quick reports generate a report automatically based on the page context, combined with an understanding of underlying data types, with no prior set up required.

Generate insights in Power Apps without losing context of your work

To get started, simply navigate to a table’s home page grid in a model-driven app and select Visualize this view command at the top of the page.

This button generates a Power BI report based on the underlying view. Any additional filters or column modifications on top of the view are also automatically added to the report to transfer that context from data to visualization.

There are several options available to personalize this visualization to help you visualize the data in ways that are meaningful to you. Here’s the post on Power BI blog that details the ways you can interact with the visual.

You can close the dialog to go back to the view page to continue your work in the model-driven app from where you left.

Enable Power BI quick report visualization in a model-driven app

You can enable this capability for a model-driven app through the app’s settings area, available in the modern app designer in preview. Enabling this capability will display Visualize this view command on the grid pages of the tables in the model-driven app.

Review documentation to learn more about this capability and the options available to interact with the Power BI visual.

Let us know your feedback in comments or on our Power Apps community forum post.

The post Visualize data in model-driven apps with Power BI (preview) appeared first on Microsoft Power Platform Blog.

The new Dataverse search name is visible in the Power Platform Admin Center and the Power Apps maker portal within the updated solution explorer experience, which is in public preview.

Dataverse search enabled by default

Dataverse search is enabled by default for all production environments. Dataverse search remains an opt-in feature for non-production environments and environments using their own encryption key. You can change the setting at any time in Power Platform Admin Center.

21 tables are in the default scope of Dataverse search and reflect the most used tables across Dataverse users. You can customize the search experience based on the needs specific to your business easily and at any time within the updated solution explorer experience.

Enhanced Maker experience for configuring Dataverse search

Power Apps Makers can configure the global search experience in model-driven apps by selecting the tables and fields to be searchable through Dataverse search – all in the context of a solution, within the updated solution explorer. Managing the Dataverse search index inside a solution enables you to export/import solutions that have model-driven apps optimized for search.

View snapshot of Dataverse search on the solution overview page

Get a quick overview of the number of fields indexed for Dataverse search in the current solution.

Configure Dataverse search with updated solution explorer

Select the tables to be indexed for search and use the table’s quick find view to select the searchable fields.

End users can explore data with search in all model-driven apps

With Dataverse search enabled, model-driven apps have a new search experience that enhances productivity by making information discovery fast and easy. These search-centered usability enhancements help you explore data in a familiar and recognizable way, making it an essential part of how you navigate through model-driven Power Apps.

Search bar in the header

Search moves to a prominent place in header. You can consistently see it on any page in any model-driven Power App.

Suggested results, as you type

View search results inline as you type, minimizing keystrokes and simplifying navigation.

Quick actions with search

Some of the commonly used tables are configured to show a set of most used commands to help you complete your task without losing context

All the above end user capabilities are enabled by default with 2021 release wave 2.

Learn more about how to enable and configure Dataverse search and ensure you have the right tables enabled for search.

Review end user documentation to discover all the ways Dataverse search can be used in model-driven apps.

Please share your feedback on our Power Apps community forum post.

The post Goodbye Relevance search, hello Dataverse search! appeared first on Microsoft Power Platform Blog.

The key to looking up a record quickly is context. This context is expressed by Power Apps Makers configuring a lookup field with filters to refine the data and views to display necessary columns. With advanced lookup, you can apply all of the available filters easily to search for, explore and select records to populate a lookup field.

Let’s take a closer look at the advanced lookup usability enhancements.

View more columns when looking up a record

Advanced lookup offers an immersive lookup experience, so you can view all columns of a view to identify the right record to select. The columns can also be sorted on, to view records in the order you prefer.

See results quickly with typeahead search

You can view results as you type with advanced lookup. Switch to another view or a completely different table and view results without having to retype your query.

Explore data from within advanced lookup without losing context

Browse a record’s details or add a new record from within advanced lookup without having to navigate away with unsaved changes.

Apply filters on the data easily

Advanced lookup supports filters made available by the Power Apps maker, along with Client APIs that are supported by the lookup control.

Looking up multiple records takes significantly fewer clicks

Special lookups like party lists support looking up multiple records across multiple tables. With advanced lookup, you can add select more records or unselect records and populate it with fewer clicks and cleaner layout.

Turn on advanced lookup experience at Power Platform Admin Center

Administrators can enable the advanced lookup experience through a feature setting in Power Platform Admin Center. Advanced lookup will be enabled by default for all environments with 2021 release wave 2.

Review the documentation for more details. Please share your feedback on Power Apps community forum.

The post Advanced lookup experience in model-driven Power Apps appeared first on Microsoft Power Platform Blog.

Today we are excited to share the next evolution and an important facet of quick actions: the ability to customize and extend them in ways that is appropriate for your processes.

What are quick actions?

Quick actions are commands that help you complete your task without losing context of your work.

Over the last few months we released a number of capabilities around search, starting with a search bar in the front and center of model-driven Power Apps. This search bar is powered by Dataverse relevance search, that shows results as you type and has robust intelligence built in.

We believe that presenting contextual quick actions with search will be the basis of a “search-first” means of discovering and acting on information in model-driven Power Apps, enhancing your productivity and easing navigation.

Understanding quick actions under the hood

Quick actions are contextual and relevant so that you can act on your information as you look for them with search. And hence, quick actions are a subset of a table’s homepage grid’s item-level commands.

For example, when you select an account in its homepage grid, Account table’s quick actions are derived from the set of commands at the top of the page. This is important to understand the customization options available to configure quick actions.

Using rules to hide/show quick actions

You can use the ribbon’s Enable Rule to hide or show quick actions for a table. The following three new rules give you the flexibility to optimize quick actions:

• ShowOnQuickAction
  Use this rule to make a command appear only as a quick action.
• ShowOnGridAndQuickAction
  Use this rule to make a command appear on the homepage grid as well as a quick action.
• ShowOnGrid
  Use this rule to make a command appear on the home page grid only.  You can also use this enable rule to hide an existing quick action.

  

Things to remember

For usability reasons, quick actions are capped at FIVE commands per table. Quick actions show up only in the context of search – alongside suggestions and in results page on the primary column.

These capabilities will be available with build 9.2.21034.00126, starting 5 Apr 2021. We’d like to hear your feedback.

Please share your feedback on our Power Apps community forum post.

Review more details on how to enable relevance search and how to configure quick actions and learn more about benefits of the new search to end users.

The post Search quick actions are now customizable in model-driven Power Apps appeared first on Microsoft Power Platform Blog.

New experience rolled into relevance search

With 2021 release wave 1, the new experience is rolled into relevance search so the enhanced search helps you look for your information in a familiar and consistent way in model-driven Power Apps. A quick recap of the usability improvements:

Search bar in the header

Search is moved to a prominent place in header. It can be easily discovered from any page in all model-driven Power Apps on the web consistently.

Suggested results, as you type

View search results inline as you type, minimizing keystrokes and simplifying navigation.

Results page that is easy to understand

Search results are ranked and presented grouped by entity type, with more fields displayed for distinguishing records.

Quick actions with search

Introducing quick actions, commands that help you complete your task without losing context of your work.

Some of the commonly used tables are configured to show a set of most used commands to help you complete your task without losing context. Assign an account or share a contact with your team with the most minimal keystrokes and just one click.

You can also see quick actions in results page on the primary column, so you can act on your information without navigating.

With relevance search and 2021 release wave 1 preview enabled, you can try out all of these capabilities.

Learn more about how to enable relevance search and how to enable 2021 release wave 1 preview.

Also review end user documentation for more details.

Please share your feedback on our Power Apps community forum post.

The post New search experience and quick actions in Power Apps appeared first on Microsoft Power Platform Blog.