The human-in-the-loop feature covers four aspects:

• Providing feedback: Users and makers provide feedback on AI prompt responses when the responses aren't accurate.
• Getting notifications for reviews: Makers get notified when new feedback is available.
• Listing reviews: Makers review the feedback sent for a given AI prompt.
• Few-shot learning: Makers use the feedback they receive or create samples to drive AI prompt responses and increase accuracy.

When you use prompts in Copilot agents, you set specific logic to extract and augment information from documents. With this feature, you can:

• Save hours for each Copilot user who doesn't need to manually extract, augment, and format specific data.
• Perform the exact same document processing logic for each Copilot user in one or two chat interactions.
• Reuse the prompt across multiple agents, apps, and automated flows.

To use this feature, create a prompt in Power Apps or Power Automate of Copilot Studio. Then, add an input type, such as image or document. Supported types are PNG, JPG, JPEG, and PDF. Now you're ready to include the prompt in a Copilot agent using topics (actions aren't supported yet). Finally, test the agent and publish it.

This feature is enabled by default without any admin configuration.

In prompt builder, the model selection capability gives you a secure connection to a custom model deployed in Azure AI Foundry. With this integration, you can seamlessly add your own fine-tuned and optimized models tailored to specific business needs directly into your Power Automate flows or Copilot agents.

When you use models trained on domain-specific data, such as legal, financial, or customer service information, you can create highly specialized AI actions that go beyond generic language capabilities. This approach opens new opportunities for smarter, context-aware automation and conversational experiences that reflect your organization's unique knowledge and tone.

In prompt builder, you can choose your preferred large language model (LLM) for each prompt. In this release, you can run your prompts with GPT-4 or GPT-3.5. You can also set the model temperature for the LLM to customize how creative the output response is.

GPT-4 is more capable and has better multilingual skills. It has a larger context window, so it can handle the equivalent of more than 300 pages of text in a single prompt. GPT-4 also performs better than earlier models on tasks that require careful following of instructions, like generating specific formats such as JSON formatting.

The validation station is an interactive feature in document processing that you use to review, verify, and correct data extracted from documents before final submission or integration. After AI-driven extraction captures data from invoices, receipts, contracts, or other documents, the validation station shows this data in a user-friendly interface. It highlights fields with low confidence scores or potential errors for manual confirmation. You can quickly inspect extracted fields, make corrections as needed, and approve data to ensure accuracy and compliance. This feature enhances quality control by allowing human intervention in critical points, reducing errors in automated document workflows, and increasing overall trust in the document processing system. With streamlined navigation and an intuitive layout, the validation station offers an efficient way for you to confirm AI results. This feature improves the reliability and precision of document processing outcomes.

You reach the validation station through an approval email or its own entry point. Use the validation station to monitor and validate what the system extracts and processes. The package includes this feature, so you don't need to enable it manually.

The document processing agent integrated into Copilot Studio enhances business workflows by automating the collection, extraction, categorization, and validation of data from diverse document types. By leveraging AI capabilities, it seamlessly processes invoices, contracts, and receipts. This agent transforms traditionally manual tasks into efficient, automated actions. The document processing agent includes preconfigured flows and a validation station. With these features, you can easily set up document processing workflows that work without extra configuration.

Azure Document Intelligence 4.0 brings advanced features like improved Optical Character Recognition (OCR), natural language understanding, and layout detection. With these features, AI Builder can process a wider variety of document types, from structured forms to complex unstructured content. With this integration, AI Builder users can use Azure's latest advancements in machine learning models directly within their workflows. This integration also supports returning table, row, and cell confidence scores.

Automating document-related tasks—such as summarizing reports, classifying documents, or extracting content from invoices—saves time and significantly reduces the manual work involved in managing large volumes of data. When you streamline these document and image-based workflows, your organization boosts productivity and efficiency. With automation, you can spend more time on higher-level decision-making and strategic initiatives instead of routine administrative tasks.

With advancements in AI-driven tools, you can define how to process documents or images in an end-to-end workflow by using simple natural language commands. By using the AI Builder prompt in Power Automate or Power Apps, you can easily set parameters for automatic document or image processing. Examples of parameters include key fields to extract, document categories, content summaries, and specific elements to detect within a document or image. With these capabilities, you can automate complex workflows without advanced technical skills.

This innovation eliminates the traditionally complex and time-consuming steps of labeling data and training AI models. You don't need specialized skills in machine learning or data science to automate document processing. Instead, you can seamlessly integrate these capabilities into your existing workflows by using natural language instructions. This approach makes the setup process faster, more intuitive, and accessible to a wider range of users. By democratizing AI-powered automation, you can quickly deploy intelligent document and image processing solutions that drive operational efficiency and enhance the accuracy and speed of routine tasks.

This capability helps your organization check how well generative AI prompt actions work across agents, apps, and workflows. When you create test datasets, set evaluation criteria, and run batch tests on prompt actions, you can check results and make data-driven decisions about the maturity or development of these prompts. Prompt actions act as intelligent operations within workflows, so they keep outcomes consistent and reliable. Finding issues early reduces risks, minimizes disruptions, and improves reliability. This capability also speeds up deployment times and simplifies workflows. With this capability, you get smarter AI-powered solutions that boost user satisfaction, improve operational efficiency, and support your organization's goals.

This feature makes prompts actionable with external sources by using logs for validations and better exception handling. This feature also helps to overcome limitations of a native large language model (LLM). The native prompt builder experience in Microsoft Copilot Studio is built on Fluent UI 9 (non-iFramed experience). It removes the wizard for prompt actions by integrating code interpreter elements to work with the new prompt builder experience.

Here are some tasks you can perform with this feature: You can turn on the toggle to generate the code from Adv. Setting. You can turn off the toggle to stop generating the code. You can select the checkbox to write data to Dataverse tables in Adv. Setting. You can open the Preview code to autogenerate Python code and run it. You can edit the Python code and re-run the code. You can copy and paste the code in a code interpreter you find on the internet. You can undo code changes by using the Undo button. When you turn on the setting to write data to Dataverse tables, you can use the Run button to write actions to Dataverse tables.

Prompts are the instruction language for agents. An agent uses prompts when it runs directly from a prompt or as a flow running behind the scenes. These instructions play a critical role in how the agent performs its various steps. Use prompts to perform data generation tasks in Power Automate cloud flows or Power Apps applications.

This feature enables prompts to use connector data. Previously, prompts could use only data in Dataverse. This enhancement makes your agents, cloud flows, and applications more functional and flexible.

Select + Add or type / to insert table fields from existing connections. Or, use Power Fx formulas in the prompt builder to query data.

To create and use prompts in Power Automate, select AI Hub > Build your own prompt in the navigation menu on the left, or select Create text with GPT action when you author a cloud flow.

You can access the Power Fx designer from prompt builder. Define an expression and name it for inline insertion in the prompt. The prompt dynamically includes information from the expression's execution, enabling more specific responses.

To filter your knowledge, use Power Fx expressions along with the current attribute and value selection. With this approach, you get more control over the data used as the prompt's context.

The prompt accuracy scoring feature in AI Builder’s prompt builder helps you build a test suite and check your prompt performance across different versions of prompt development. With these detailed assessments, you can make informed decisions about using prompts in agents, apps, and flows, moving capabilities to production, and improving prompts. This feedback on the effectiveness of your AI prompts helps you optimize clarity and precision.

As you create or refine prompts, the feature analyzes the prompt structure, language, and relevance to the task. It assigns a confidence score to each test case prediction that shows the expected performance of the prompt. The score comes from factors such as specificity, complexity, alignment, and custom assertions. With this score, you get actionable insights to improve prompt phrasing or reduce ambiguity.

By giving you a clear, quantifiable measure of prompt quality, the accuracy scoring feature simplifies the prompt engineering process. It enhances model outcomes and reduces iteration time. With this feature, you get more efficient and reliable AI interactions across use cases.

With prompts, you can easily create sophisticated and custom AI capabilities for business agents, applications, and workflows using natural language instructions. The Azure OpenAI reasoning model (starting with the o1 model) is available in prompt builder, so you can use deep reasoning for advanced AI prompts. With this model, you can analyze large datasets, generate accurate forecasts, extract meaningful insights, and perform detailed analyses.

Draft with Copilot is a feature that helps you write or rewrite your prompts using Copilot. The prompt builder makes it easy to start Copilot so it can help you create your prompt. Copilot turns your intent into an effective and sophisticated prompt that follows the best practices for prompt engineering. Along with writing suggestions, Copilot gives you optimization tips, a completeness score, and sample test data to help you run validations faster.

With prompt builder, you can create and test specific output formats directly in the AI Builder prompt builder experience. By using prompt builder, you can address multiple scenarios that go beyond text generation. These scenarios include automation context or app behavior extensions where reliable data formatting is crucial. During public preview, JSON formatting is the first supported output format. AI Builder prompts that generate JSON formatting are available from Power Automate, Power Apps including PCF controls, and custom copilots.

With this capability, you can add Dataverse data records as an input source to your AI Builder GPT prompts. When you add Dataverse data to GPT prompts, you customize the knowledge from GPT with enterprise data stored in Dataverse.

In prompt builder, select the option to add knowledge to GPT prompts. For example, select specific Dataverse records to include in the prompt. Add specific instructions to filter the data according to the target scenario. This approach amplifies GPT with data knowledge. Use prompts to perform key scenarios like data summarization and classification. You can trigger these prompts from copilots, Power Automate, and Power Apps.