diff --git a/articles/TOC.yml b/articles/TOC.yml index 2643bb529..1c80a3012 100644 --- a/articles/TOC.yml +++ b/articles/TOC.yml @@ -37,7 +37,13 @@ - name: Export and import a non-solution flow href: export-import-flow-non-solution.md - name: Drafts and versioning for cloud flows - href: drafts-versioning.md + href: drafts-versioning.md + - name: Generative actions (preview) + items: + - name: Overview + href: generative-actions-overview.md + - name: Create generative actions + href: create-generative-actions.md - name: Automated flows items: - name: Create a cloud flow from a template @@ -1264,6 +1270,8 @@ href: faqs-copilot-in-pad.md - name: FAQ for Repair with Copilot at runtime in Power Automate for desktop href: faqs-repair-copilot.md + - name: FAQ for generative actions in cloud flows + href: faqs-generative-actions.md - name: FAQ for generating code with natural language href: faqs-copilot-nl-to-code.md - name: FAQ for suggested actions in Power Automate for desktop diff --git a/articles/create-generative-actions.md b/articles/create-generative-actions.md new file mode 100644 index 000000000..b57d5223c --- /dev/null +++ b/articles/create-generative-actions.md @@ -0,0 +1,142 @@ +--- +title: Create generative actions in cloud flows (preview) +description: Learn how to create generative actions in cloud flows. +ms.date: 12/20/2024 +ms.topic: conceptual +author: HeatherOrt +contributors: + - HeatherOrt + - v-aangie +ms.author: heortaol +ms.reviewer: angieandrews +ms.collection: + - bap-ai-copilot +--- + +# Create generative actions in cloud flows (preview) + +[!INCLUDE[cc-preview-features-definition](includes/cc-preview-features-top-note.md)] + +Generative actions are a new type of action that is authored, tested, and executed through an AI runtime. Generative actions can simplify and accelerate the automation process by reducing the need for specifying the action sets and sequencing to execute. Generative actions can also adapt to changing scenarios and data sources by leveraging the AI runtime's intelligence and reasoning capabilities. Generative actions can handle complex and dynamic tasks that involve multiple steps, conditions, loops, branches, and integrations. + +[!INCLUDE[cc_preview_features_definition](includes/cc-preview-features-definition.md)] + +## Create a generative action + +To create a generative action, you start in cloud flows designer. + +1. Sign in to [Power Automate](https://make.powerautomate.com). +1. On the navigation menu to the left, create a cloud flow using one of the following options: + 1. Select **Create** > one of the following options under **Start from blank**: + + - Automated cloud flow + - Instant cloud flow + - Scheduled cloud flow + - Describe it to design it + - Desktop flow + + :::image type="content" source="media/create-generative-actions/start-from-blank.png" alt-text="Screenshot of the create page in Power Automate"::: + + Or + + 1. Select **My flows** > **New flow** > one of the following menu options under **Build your own from blank**: + + - Automated cloud flow + - Instant cloud flow + - Scheduled cloud flow + - Describe it to design it + - Desktop flow + +1. Give your flow a name, select a trigger, or follow the directions to build a flow. +1. Go to the designer by selecting **Create** or **Create flow**, depending on the flow type you selected. +1. In the designer, select the plus sign (**+**) to create an action, and then select **Add generative action (preview)**. +1. In the **Parameters** tab, select **+ New generative action**. + + :::image type="content" source="media/create-generative-actions/new-generative-action.png" alt-text="Screenshot of the option to create a new generative action."::: + +1. To describe the automation you want to create, enter the intent of your automation in as many details as you like. Then, get an AI suggested action plan by selecting **Generate**. + + :::image type="content" source="media/create-generative-actions/generate-action.png" alt-text="Screenshot of a generative action description and the 'Generate' button."::: + +### AI generated suggestions + +AI generates suggested input and output types to use in the flow. AI also generates suggested connectors and actions that it can use to achieve the automation goal. + +Inputs are texts that the generative action uses to execute the plan. This text can be passed dynamically from previous cloud flow actions. Inputs are limited to 2,500 characters. + +Outputs are text that the generative action creates that can be used in succeeding actions in the flow. + +> [!TIP] +> To remove additional detail or HTML and reduce input character count, review and use the **Compose** action on dynamic input. + +:::image type="content" source="media/create-generative-actions/inputs-outputs.png" alt-text="Screenshot of a computer showing the AI-generated inputs and outputs"::: + +### Finalize your generative action + +Finalize your generative action and preview it before execution. This allows you to make sure that the generative action is working as expected, and make necessary changes before adding it to your cloud flow. + +1. Review, accept, or reject these suggestions. + + - If you don't accept suggestions, they're deleted when you preview or add your generative action to the flow. + - You can add custom inputs, outputs, connectors, and actions if they aren't suggested. + - The generative action doesn't execute on actions that aren't first added to the plan. + +1. To preview your generative action, select the **Preview** tab. +1. Provide sample inputs for your generative action, such as an order ID or a customer name. + + The generative action shows you its thought process as it analyzes the inputs you provided against the intent of the automation. The generative action runs and pauses before taking any action while it's in preview. The logic, the data behind each action, and the expected outputs display. + +1. Modify the inputs, outputs, or actions if needed, and run the preview again. Once you're satisfied with the preview, exit the preview mode. +1. When you're satisfied with your generative action, select **Add** to add it to your cloud flow. + +## Add references to a generative action + +You can add up to three Microsoft OneDrive documents as references. The generative action can use them as part of its execution. + +The Microsoft Word document must be labeled as non-confidential and reside in the owner's OneDrive. Each Word document must be fewer than 10 MB. + +## Monitor a generative action + +You can monitor the run history and the performance of your generative action from the **My flows** page in Power Automate. You can also edit or customize your generative action at any time in the cloud flow designer. + +To monitor and edit a generative action, follow these steps: + +1. From the **My flows** page, select the **Generative actions** tab. + + You can display the list of your generative actions, their status, and their last run time. + +1. To monitor the run history of a generative action, select the generative action name > **Run history**. + + The details of each run, such as the start time, end time, duration, status, inputs, and outputs, display. The reasoning behind each action that the generative action called, and the data that was used also display. + +1. To edit or customize a generative action, select the generative action name > **Edit**. + + You can modify the instruction set or intent of the generative action, or the inputs, outputs, or actions that the generative action uses. + +1. Save your changes and preview your generative action again. + +## Known limitations + +The following limitations apply to generative actions in cloud flows: + +- Only text based inputs are supported. +- Cancelling the cloud flow doesn't cancel the running generative action. It must be canceled separately. +- Currently limited to the following connectors: + - SharePoint + - Office 365 Outlook + - OneDrive for work or school + - Planner + - Microsoft Teams + - Notifications + - Office 365 Users + - Approvals + - Excel online for Business + + > [!NOTE] + > if you wish to include more connectors in your generative actions use case, contact your representative. + +- Data loss prevention (DLP) policies for cloud flows are supported exclusively at the generative action level. Differentiating connectors as Business or Non-Business between cloud flow connections and generative action connections isn't currently supported. + +## Related information + +[FAQ for generative actions in cloud flows](faqs-generative-actions.md) \ No newline at end of file diff --git a/articles/faqs-generative-actions.md b/articles/faqs-generative-actions.md new file mode 100644 index 000000000..21cdd8627 --- /dev/null +++ b/articles/faqs-generative-actions.md @@ -0,0 +1,63 @@ +--- +title: FAQ for generative actions in cloud flows +description: This FAQ provides information about the AI technology used in Power Automate generative actions, along with key considerations and details about how AI is used, how it was tested and evaluated, and any specific limitations. +ms.date: 12/18/2024 +ms.custom: + - responsible-ai-faqs +ms.topic: article +author: HeatherOrt +ms.author: heortaol +ms.reviewer: angieandrews +--- + +# FAQ for generative actions in cloud flows + +These frequently asked questions (FAQ) describe the AI impact of the generative actions feature in Power Automate. + +## What are generative actions? + +Generative actions are a new type of action that's authored, tested, and executed through an AI runtime. You specify only the intent of the automation and the AI chooses the right set of actions in the right order based on your input, context, and intent. + +[!INCLUDE[cc_preview_features_definition](includes/cc-preview-features-definition.md)] + +## What are capabilities of generative actions? + +Generative actions simplify automation by reducing the need for detailed specifications, adapt to changing scenarios using AI intelligence, and handle complex tasks involving multiple steps and integrations. + +- Generative actions can create a set plan for flow execution through natural language. Generative actions can simplify and accelerate the automation process by reducing the need for specifying the conditions and action sets to execute. It surfaces recommended inputs, outputs, and actions to complete the described task. The user is then asked to validate these recommendations before running the generative actions and is allowed to make any changes necessary. + +- Generative actions can adapt to changing scenarios and data sources by leveraging the AI runtime's intelligence and reasoning capabilities. Generative action can orchestrate actions within the flow based on reasoning over inputs and knowledge documents. It then executes the plan by filling in the required inputs for the flow. + +- Generative actions can handle complex and dynamic tasks that involve multiple steps, conditions, loops, branches, and integrations. Once it is complete it shares the actions it executed and the reasoning behind parameter filling. + +## What is the intended use of generative actions? + +You can use generative action to intelligently orchestrate, and parameter fill actions within your flow. The generative action can act based on the defined inputs. + +## How was the generative actions feature evaluated? What metrics are used to measure performance? + +The generative actions feature is evaluated for end-to-end quality at each step of the process. Quality is measured in terms of how well the system creates and executes a plan that successfully addresses the user prompt. We evaluate quality over various user queries, prompts, and actions. We also evaluate how well the system does at ignoring malicious content from users and authors, and how well the system avoids producing harmful content. + +## What are the limitations of generative actions? How can users minimize the impact of generative action limitations when using the system? + +For best results, make sure your generative action prompt includes high-quality descriptions. The generative action can only act on the inputs and reference document given, be sure that the content is complete for the best performance. The generative action can also only orchestrate the actions that were defined and authenticated. + +## What operational factors and settings allow for effective and responsible use of generative actions? + +The generative actions feature is currently *English only*. Once you create a generative action, you can test the system to see how well it performs using test. You can run the test and manually trigger actions to understand how your generative action will perform. You can always edit or modify the prompt to change the way the generative action performs. + +## What kinds of issues might arise when using generative actions? + +Actions might not always work as intended. Errors might occur when preparing the input for the action or when generating a response based on the action's output. Your generative action might also call the wrong action during the orchestration of the given actions. To mitigate the risk of such errors, make sure you have high quality, relevant, and unambiguous descriptions configured for the actions. Also limit what actions the generative action might be able to use in the execution of the plan. + +## What protections does Power Automate have in place for responsible AI? + +There are many mitigation features in place to protect your flows. You can configure your generative action with a set of references and input values defined within your flow. Generative actions never take an action that isn't part of their plan. Generative actions respect organizations' Data Loss Prevention (DLP) policy. Any disallowed action for the flow is blocked in the generative action. + +In addition, we have classifiers that look at input to the system to detect harmful content and jailbreak attacks. According to our tests, these classifiers have a high success rate at blocking harmful content and jailbreak attacks, while also having high success at not blocking content that isn't harmful or a jailbreak attack. However, classifiers can't be perfect so there are risks of a generative action producing harmful content or responding to a jailbreak attack. These risks include cross-domain prompt injection attacks, where instructions could be added to the output of an action or a knowledge source that the generative action then tries to follow. + +## Related information + +[Create generative actions in cloud flows (preview)](create-generative-actions.md) + +[!INCLUDE[footer-include](./includes/footer-banner.md)] \ No newline at end of file diff --git a/articles/generative-actions-overview.md b/articles/generative-actions-overview.md new file mode 100644 index 000000000..6a1e6b1db --- /dev/null +++ b/articles/generative-actions-overview.md @@ -0,0 +1,47 @@ +--- +title: Overview of generative actions (preview) +description: Get an overview of generative actions, including benefits, how to author, monitor, and edit. +ms.date: 12/20/2024 +ms.topic: conceptual +author: HeatherOrt +contributors: + - HeatherOrt + - v-aangie +ms.author: heortaol +ms.reviewer: angieandrews +ms.collection: + - bap-ai-copilot +--- + +# Overview of generative actions (preview) + +[!INCLUDE[cc-preview-features-definition](includes/cc-preview-features-top-note.md)] + +Generative actions are a new type of action that is authored, tested, and executed through an AI runtime. You specify only the *intent* of the action and the AI chooses the right set of actions in the right order based on your input, context, and intent. + +[!INCLUDE[cc_preview_features_definition](includes/cc-preview-features-definition.md)] + +## Benefits of generative actions + +Generative actions simplify automation by reducing the need for detailed specifications, adapt to changing scenarios using AI intelligence, and handle complex tasks involving multiple steps and integrations. + +- Generative actions can simplify and accelerate the automation process by reducing the need for specifying the conditions and action sets to execute. +- Generative actions can adapt to changing scenarios and data sources by leveraging the AI runtime's intelligence and reasoning capabilities. +- Generative actions can handle complex and dynamic tasks that involve multiple steps, conditions, loops, branches, and integrations. + +## Author generative actions + +You can author generative actions in the cloud flows designer in Power Automate. When you give the system an intent, the AI generates suggested inputs, outputs, and actions to use in the action. You can choose to accept the AI suggestions, or reject them and add your own inputs and outputs. + +In addition to inputs and outputs, the system also provides connectors and actions that it can use to achieve its intended automation goal. You can choose to accept, reject, or add your own. + + Once you create the action, you can preview how it behaves by providing sample inputs and verifying what the action does. In the preview phase, the action pauses before taking any action. You can run this multiple times. + +## Monitor a generative action + +You can monitor your generative actions runs from the cloud flows run history. From the **My Flows** page, you can select any generative action under the **Generative actions** tab to get to its manage page. This page is where you can monitor the run history for the related generative action. To examine the process behind why a generative action called a specific action, you can dive into any run history instance. + +## Edit or customize a generative action + +To modify the instruction set or intent of a generative action at any time, select **Edit** in the flow design page. + diff --git a/articles/media/create-generative-actions/generate-action.png b/articles/media/create-generative-actions/generate-action.png new file mode 100644 index 000000000..08277ace5 Binary files /dev/null and b/articles/media/create-generative-actions/generate-action.png differ diff --git a/articles/media/create-generative-actions/inputs-outputs.png b/articles/media/create-generative-actions/inputs-outputs.png new file mode 100644 index 000000000..5fd8286e7 Binary files /dev/null and b/articles/media/create-generative-actions/inputs-outputs.png differ diff --git a/articles/media/create-generative-actions/new-generative-action.png b/articles/media/create-generative-actions/new-generative-action.png new file mode 100644 index 000000000..ae3f2d985 Binary files /dev/null and b/articles/media/create-generative-actions/new-generative-action.png differ diff --git a/articles/media/create-generative-actions/start-from-blank.png b/articles/media/create-generative-actions/start-from-blank.png new file mode 100644 index 000000000..3e89ec013 Binary files /dev/null and b/articles/media/create-generative-actions/start-from-blank.png differ diff --git a/articles/responsible-ai-overview.md b/articles/responsible-ai-overview.md index c5dc66826..6c9d0b73c 100644 --- a/articles/responsible-ai-overview.md +++ b/articles/responsible-ai-overview.md @@ -21,6 +21,8 @@ ms.collection: bap-ai-copilot - [FAQ for Copilot in cloud flows](faqs-copilot.md) - [Create, update, and fix your expressions with Copilot expression assistant (preview)](flows-designer.md#create-update-and-fix-your-expressions-with-copilot-expression-assistant-preview) - [FAQ for Copilot expression assistant](faqs-copilot-expression-assistant.md) +- [Create generative actions in cloud flows (preview)](create-generative-actions.md) +- [FAQ for generative actions in cloud flows](faqs-generative-actions.md) - [Copilot in Process Mining ingestion (preview)](process-mining-copilot-in-ingestion.md) - [Copilot in Process Mining process analytics (preview)](process-mining-copilot-in-process-analytics.md) - [FAQ for Copilot in Power Automate Process Mining](faqs-copilot-in-process-mining.md)