Recent News in Analytics and AI: August 2025 Edition

This month's summary of recent news in analytics and AI dives into the availability of org apps in Power BI Pro, the preview of Semantic Model Refresh Templates, Meta AI's release of DINOv3, details of NASA's Jet Propulsion Laboratory's use of DINOv2, the launch of OpenAI's GPT-5 across Microsoft 365 Copilot and Copilot Studio, and more.

Read on and get up to speed.

Power BI

• Org apps are now available in Power BI Pro workspaces. If you've built Power BI apps in a Pro workspace and you want to try out creating multiple org apps for reporting in your company, you can! This means you now get the capability to customise your report distributions for each group you're creating them for. Learn more.

• Microsoft Fabric Copilot's measures description writing feature is now generally available in both Power BI Desktop and web modeling. This means Copilot can write a description for your report creators at the click of a button. So when you're creating or making changes to your model, you can communicate what's new, enabling your report creators to use the measures in the right way. Learn more.

• As of this month (September 2025), the standalone Copilot experience for Power BI is set to be enabled by default. So if you've already got Copilot turned on, the chat with your data chat-based AI experience will be turned on as standard. But if you're still preparing your data for AI and you're not quite ready to go, you can also opt out. Learn more.

• Semantic Model Refresh Templates are now in preview. This function gives you the ability to execute semantic model refreshes using Fabric Data pipelines. This will be useful to you if you want to visualise and sequence refresh activities for common and advanced scenarios. The templates are now accessible in your refresh dropdown menu inside the Microsoft Fabric workspace. Learn more.

Microsoft Fabric

• There is a new level of support available for workspace identity in Fabric. This means that you no longer need to manage credentials in scenarios that support it. Instead, across Data Pipelines, Copy Jobs, Semantic Models, and Dataflows Gen2, you'll now have centralised, secure access control without the overhead of managing service principals and secrets. Learn more.

• Copy Job Activity is now in preview. With this functionality, you can use the Copy job item directly in your Microsoft Fabric Data Factory pipelines. This enables you to manage everything from your data movement to notifications in one place within the simple no-code experience. The preview also features a handy monitoring tool that'll give you real-time insight into your copy job status. Learn more.

• There's been a major update for User Data Functions in Microsoft Fabric. After receiving heavy feedback for testing and development during the preview, the Functions explorer now gives you two models to explore. With a new Develop mode and Run/View only mode, you now have dedicated spaces for users with different permissions to work on their tasks. Learn more.

• OneLake's capacity utilisation model has been simplified. The consumption rate has been reduced for OneLake transactions via proxy so that it now matches the transactions via redirect rate. This gives you the ability to consume data via either proxy or redirect, without worrying about the rate (or having to choose between the two). This soothes what was previously a major pain point for those consuming OneLake data from platforms and tools outside of Fabric and demonstrates Microsoft's commitment to OneLake's openness and simplicity. Learn more.

Microsoft 365 Copilot and Copilot Studio

• In August, OpenAI's GPT-5 became available in Microsoft 365 Copilot. Microsoft announced its commitment to ensuring OpenAI's latest models are available in Microsoft 365 Copilot, and GPT-5 became available on its release day. Copilot uses GPT-5 to take your prompt and craft the best response with speed, accuracy, and sufficient level of depth. Learn more.

• GPT-5 was also released in Microsoft Copilot Studio last month. With this introduction, you can also choose your agent's model for generative answers for greater flexibility and accuracy. This means your agents can now use your prompts as a guide for choosing the right model (out of the high-throughput model or deeper reasoning mode) for each response. Learn more.

• The new Microsoft Fabric data agents integration in Copilot Studio means your data can be incorporated into conversations seamlessly. Fabric data agents connect directly to your OneLake data, which means they can understand schemes, interpret business context, and more. This removes the need for you to switch between analytics and chat and enables your AI assistants to produce better insights. Learn more.

• A Principal PM Manager at Microsoft has written an insightful article on simplifying agent creation, adoption, and governance. This field guide breaks down the best way to adopt Copilot and agents in five simple steps. If you want to make the process easier, give this one a read. Learn more.

Azure Analytics and AI

• OpenAI's open‑source model gpt‑oss has now been released on both Azure AI Foundry and Windows AI Foundry. Microsoft has a vision for its AI platform–to build a full-stack AI app and agent factory that spans cloud to edge. Azure AI Foundry is there for building and fine-tuning, while Foundry local enables the use of open-source models. Now, you have the ability to run and deploy OpenAI's gpt‑oss models in your own way. Learn more.

• Microsoft has launched a technical guide called “Accelerating Generative AI Innovation with Cloud Migration”. Here, they're outlining super useful insights into how IT and digital transformation leaders can utilise Azure to improve generative AI outcomes. Take a look at the use cases they're sharing to see the benefits. Learn more.

• Agent Learning from Human Feedback (ALHF) is a new machine learning paradigm that powers Databricks Agent Bricks. In their latest case study, which dives specifically into Agent Bricks Knowledge Assistant (KA), discover how responses are continuously improved thanks to expert feedback. If you want your agents to adapt faster and more intuitively, this case study will be a valuable read. Learn more.

• If you frequently experience errors performing repetitive SQL tasks like cleaning data, SQL Stored Procedures in Databricks will change that. Rather than constantly copying and pasting code, you get to store the logic once. You can then run it whenever you need it. Storing the procedure under Unity Catalog makes the process much easier, more consistent, and efficient. Given that notebooks are the primary container for code in Databricks, this appears to be a move to sooth migrations away from SQL-centric data warehouse solutions in other platforms. Learn more.

Open-Source Analytics and AI

• Under the Apache 2.0 license, OpenAI has now released two open-weight language models. Positioned as high-performance at a low cost, gpt-oss-120b and gpt-oss-20b were built to outperform similar models, and are trained using OpenAI's most advanced in-house models, including o3. If you want to give them a whirl within agentic workflows, read on to find out how. Learn more.

• DuckDB has published a super handy post to demonstrate how to perform some of the essential machine learning data preprocessing tasks you might want to work on. If you want to expand your capabilities with categorical encoding, feature scaling, and more to simplify your workflows and make your data prep more efficient, dig into the post. Learn more.

• Meta AI has introduced DINOv3. DINOv3 is a computer vision model trained with self-supervised learning that has the ability to create superior high-resolution visual features. It removes the need to rely on human-generated metadata like captions for training, freeing up both time and resources when you want to scale. There are also some sample notebooks for you to try. Learn more.

• If you're a fan of SciFi or planetary exploration, Meta has a great article you need to read. NASA's Jet Propulsion Laboratory has been sending robotic explorers to Mars for decades, but now they're in need of more, smaller alternatives to further research. This is where DINOv2, released by Meta Fundamental AI Research (FAIR) in 2023, comes in. Read the piece to find out the details. Learn more.

Stay in the Know

Get notified when we post something new by following us on X and LinkedIn.