Episode 300 – Domain Tags, Data Agents Meet Copilot Studio & Cosmos DB Arrives

Celebrating episode 300 at the tail end of July (with Jason’s eldest starting senior year band camp and John fresh from his son’s wedding), the duo tackled the Microsoft Fabric July 2025 feature summary—a summer release focused on governance enhancements, AI agent integration, and the long-awaited Cosmos DB preview.

Domain Tags Enhance Governance & Findability

Fabric domains gained tagging capabilities, adding metadata layers within domain structures to improve filtering and discovery. Originally, domains provided only top-level segmentation; now tags enable more granular organization.

“The nice thing about domains is you can go to the domain page and find the things you’re looking for,” Jason explained. “Now you’re able to filter better and find things in a much easier way.”

John views domains as his long-desired workspace folders—with the key difference that workspaces can exist in multiple domains. Tags within domains help manage the “massive number of items that you can start to accumulate as you use fabric.”

Jason emphasized the governance-usability connection: “Governance that doesn’t lend itself to usability is a waste of time in my opinion. I think one has to precede the other.”

The feature helps admins manage and govern data per business context with delegation settings, though Jason still grumbles about “domain admins” terminology conflicting with the Microsoft identity term.

Data Agents Integrate with Copilot Studio

The month’s biggest advancement: Fabric data agents now integrate with Microsoft Copilot Studio, enabling users to query data from anywhere in the Microsoft 365 ecosystem rather than exclusively within Fabric.

Jason had been testing data agents since Build but hit frustration with their limited scope. The Copilot Studio integration changes everything: “Now you can go into copilot studio and actually build an agent, a copilot studio agent that on the back side talks to a fabric data agent.”

He demonstrated minimal setup—just connecting the Copilot Studio agent to a Fabric data agent with instructions already configured. “It works. It’s not perfect. It loses its grounding a little bit periodically here and there, but as a starting point, I will happily take it.”

The integration supports Microsoft’s “don’t make me move” philosophy for M365. Rather than forcing users into Fabric to ask data questions, they can interact through Teams or other M365 entry points.

Jason shared a client story about AI expectations versus reality: spending time “figuring out the right way to fail a whole bunch of different ways” before recommending proper Fabric data architecture. His conclusion aligned with advice from Mark Rackley: build data correctly first, then layer AI interactions appropriately rather than forcing AI to unstructure well-structured data.

Data Agent Improvements: Instructions & Reasoning

Beyond Copilot Studio integration, data agents gained refinement capabilities:

• Enhanced data source instructions let authors provide more specific guidance about handling particular question types
• Reasoning transparency (the “show your work” feature) exposes how agents arrive at conclusions—though Jason questions whether end users need this visibility versus just admins

John noted these capabilities tap into broader LLM-level features appearing across AI chatbots industry-wide.

Activator Gets Easier & More Flexible

Real-time intelligence updates focused entirely on Activator, addressing usability gaps that made rule creation daunting for casual users. Previously, users had to structure everything from scratch; now the system prompts for rule creation directly from streams and walks through the process.

New capabilities include:

• Teams integration for group chats and channels rather than just individual notifications
• Parameter passing to Fabric items (preview) enabling hardcoded or dynamic values from data sources
• Enhanced action types beyond alerting—triggering notebooks, pipelines, or flows based on conditions

“From a business process perspective, important to be able to deal with,” Jason noted about the Teams channel capability, which lets team members claim and handle alerts collaboratively.

John acknowledged implementation variations depending on entry point (Fabric interface, event streams, or Power BI visuals), with some workarounds needed for incomplete features like Teams subcategory selection.

Pipeline Auto-Refresh Controls

Data pipelines gained manual control over the status window auto-refresh that shows execution progress. Previously defaulting to 30-second updates, users can now adjust frequency or disable entirely.

John explained the usability problem: auto-refresh can “steal context” while users continue working in the pipeline during background execution. The controls prevent workflow interruption—a small but meaningful quality-of-life improvement.

Cosmos DB Finally Appears (In Preview)

The Build announcement finally materialized: Cosmos DB (NoSQL) entered preview as a native Fabric database, joining SQL Database to make the databases category plural.

“It’s a big announcement because now the databases category in Fabric is not alive, it’s not database,” John noted.

Jason sees strategic significance: “It brings more and more into light where Microsoft is going with fabric and that it is going to be the data platform.” Rather than building in Azure then pulling into Fabric, the goal becomes building directly in Fabric from the start.

Critical implementation detail: Cosmos DB in Fabric requires enabling a preview feature toggle in the admin center—it won’t appear automatically. John and Jason discovered this while preparing their TechCon Atlanta deck, finding the toggle had rolled out recently but wasn’t visible until activated.

Important distinction: searching “Cosmos” in new items now returns two options—mirrored Azure Cosmos (connecting existing databases) and native Fabric Cosmos (creating new databases). The same pattern exists for SQL (mirrored vs. Fabric SQL).

OneLake Catalog Default Categories

The OneLake data catalog updated default categories based on product switcher context (Power BI vs. Fabric):

• Power BI users default to “insights” category
• Fabric users default to “data” category

Jason’s reaction: “Nice enhancement, but well, [I wouldn’t have noticed].” John agreed it’s the kind of invisible improvement that simply makes sense without moving cheese.

Conference Circuit Continues

Recording at month’s end, the pair prepared for TechCon 365 Atlanta in two weeks, followed by Branson’s Collaboration Summit in September. Jason noted life returning to normal as kids head back to school—band camp starting for his senior son, cross country for his youngest, and kindergarten starting for his wife (a teacher).

John enjoyed his long civic holiday weekend in Canada (variously named by municipality—John Galt Day in Guelph), while Jason’s family learned valuable lessons about jalapeño capsaicin oils during salsa-making from their garden harvest.

July delivered focused updates: governance enhancements through domain tags, breakthrough AI agent integration with Copilot Studio, and Cosmos DB preview completing Fabric’s database story. As Jason summarized: “That was a decent amount of stuff though” for mid-summer, with the platform “constantly moving forward.”

---

Links:

• Microsoft Fabric July 2025 Feature Summary
• Fabric Roadmap
• Episode 298 – Fabric June 2025: Notebooks Get Variable Libraries & Event Streams Level Up

Subscribe: SoundCloud | iTunes | Spotify | TuneIn | Amazon Music