New improvements coming to the AI Skill
Additional Authors: Amir Jafari, Shreyas Radhakrishna, Nellie Gustafsson
We’re excited to unveil significant enhancements to the AI Skill in Fabric! These updates make it easier than ever to build a powerful generative AI data agent, helping you to unlock deeper insights and streamline decision-making.
With improved flexibility, intelligence, and user-friendliness, the updated AI Skill is now more capable than ever. Let’s explore what’s new and how these upgrades can transform the way you work with data in Fabric.
What is an AI Skill?
AI Skills empower users to create their own conversational agent within Fabric using generative AI. AI Creators can easily tailor the AI Skill by providing instructions and examples, allowing the AI to understand their organization’s unique data context. The AI Skill allows users to provide instructions and examples, helping the AI understand their organization’s unique data context. This empowers AI Creators to build a data agent that acts as an expert on organizational data, which can be shared broadly to help consumers quickly gain insights.
What’s new?
The AI Skill now supports multiple data sources, integrates KQL Databases and Semantic Models, enhances reasoning capabilities for multi-step queries, introduces a fully conversational experience, and features a redesigned chat canvas for improved usability.
Conversational support
The AI Skill now supports fully conversational interactions, making the experience more seamless and intuitive. Conversations persist across sessions, allowing users to pick up where they left off. The AI Skill also remembers previous questions and answers, making follow-ups easier without requiring users to repeat themselves. Improved context retention enables fluid and iterative data exploration, making engagement with the AI Skill feel more like a natural conversation.
Add multiple data sources
The AI Skill now supports multiple data sources, giving users greater flexibility in exploring and analyzing their data.
Users can:
- Add up to five different data sources.
- Select relevant tables within those sources.
- Ask natural language questions across all connected datasets.
The AI Skill intelligently determines the best source for answering each question, ensuring more accurate and relevant responses. Additionally, AI Creators can provide instructions to fine-tune results and guide the AI on which sources to prioritize. This enhancement enables richer insights from diverse datasets, all within Fabric.
KQL database integration
Users can now link their KQL Databases as a data source within the AI Skill, bringing real-time streaming and analysis capabilities to natural language querying.
AI Creators can configure their AI Skill with a KQL Event house data source by:
- Providing AI Instructions to contextualize the AI Skill.
- Supplying example KQL queries to improve the AI generated queries.
- Selecting relevant tables for querying.
Once set up, AI creators and consumers can submit natural language questions, and the AI Skill will generate a curated response with an interpretation of the query and the Kusto query used to generate the answer. This integration combines the real-time power of Kusto with the customizability of the AI Skill, making complex data insights accessible without requiring deep technical expertise.
Semantic Model integration
The AI Skill now supports Semantic Models, enabling AI Creators to integrate structured business data for natural language querying. Semantic Models define business logic, relationships, and aggregations, creating a curated data layer for more intuitive data exploration. AI Creators can add Semantic Models and determine which tables are available for querying.
Once configured, both AI Creators and Consumers can ask natural language questions, which the AI Skill translates into DAX (Data Analysis Expressions) queries using LLMs. Users can track the entire query process within the chat canvas, viewing input interpretations, transformations, and the exact DAX queries executed. This feature makes structured business data more accessible and actionable for all users.
Enhanced reasoning capabilities
The AI Skill now breaks down complex questions into multiple steps to find the most accurate answers. Instead of responding in a single step, it can recognize when multiple sources are needed, pull relevant information from different datasets, and take the right steps to generate a complete response.
For example, if a question requires checking both Lakehouse data and a KQL database, the AI Skill will automatically query each source. Additionally, users can see every step of the reasoning process, including each executed query and how the final answer was derived, increasing transparency and reliability.
New & improved chat canvas
The updated chat canvas provides AI Creators with full transparency into the AI Skill’s reasoning process. Users can now track each step taken by the AI, from reformulating the initial question to generating queries and retrieving results. Every step includes a restated version of the question, the executed query, and a sample of the returned data, ensuring clarity and insight into how responses are generated.
Try it out now
The latest AI Skill capabilities are now rolling out and are expected to be fully available by the end of the week. These features will be accessible to all customers using Fabric F64 or larger capacities – learn more.
Please note that your tenant admin must enable this preview experience and the Fabric AI cross-geo setting before you can try it out.
We encourage you to try out this new public preview generative AI experience and let us know what you think. Submit your feedback here.
We’ll be demoing these capabilities and more at the upcoming FabCon event. Register now to secure your spot. We can’t wait to see how this experience empowers users in your organization!
