‘Fragmentation is poison’: How Microsoft is targeting disparate data to boost AI adoption

Microsoft is consolidating data management and data ontology to support the expansion of agentic AI, with new tools intended to make data analytics, labelling, and context sharing easier than ever.

At FabCon and SQLCon 2026, held in Atlanta, the tech giant unveiled two new expansions for its SaaS analytics platform Microsoft Fabric.

Database Hub is a new unified database management plane for Microsoft Fabric, with built-in AI agents and Microsoft Copilot allowing for natural language exploration of enterprise data across Azure SQL, Azure Cosmos DB, Azure Database for PostgreSQL, and others.

Microsoft is also expanding the extent to which enterprises can ground AI in business context through Fabric IQ, with new semantic and ontological controls supported through a new model context protocol (MCP) server in preview.

ITPro spoke to Amir Netz, CTO of Microsoft Fabric, to understand the context for the announcements and the problems Microsoft is looking to solve at the data layer.

Netz is one of the original creators of the data analytics and visualization suite Power BI, which started life as Power Pivot and Power Query within Microsoft Excel.

He told ITPro he is “incredibly happy” with the uptake in the years since, and that his team has now turned its attention to further consolidating and easing the process of data retrieval and analytics.

The goal, he noted, is to do for data what Microsoft Office did for productivity, by rolling multiple products into one.

“You would never think about buying Google Sheets and using it with Word, that seems kind of ludicrous, because the power of integration is amazing and people really appreciate integration over almost any other feature you can have in a suite.

“And the world of data is so complex, and so fragmented, that when we came with Fabric we said, ‘we want to really do what Office did for productivity, we want to do with data’.”

This involved taking approximately 20 distinct data services at Microsoft and integrating them into a single product to meet enterprise data workload needs, Netz explained. It includes batch data, streaming data, data warehousing, business intelligence, machine learning, Internet of Things (IoT), and data lakes.

Netz added that 90% of the Fortune 500 now uses Fabric, with revenue and users on the rise.

Fabric, IQ, and OneLake

In 2023, Microsoft launched Fabric in preview along with OneLake, a new unified SaaS data lake that Netz dubs “the OneDrive for data”.

Within OneLake, organizations can use ‘shortcuts’ to point to external data without duplicating it. Another solution, Mirroring in Fabric, allows users to replicate data sources such as Azure SQL, Google BigQuery, Oracle, SAP, or Snowflake directly within OneLake with low latency.

“It doesn't really matter where it's physically sitting, and that's super, super important, because fragmentation is poison, it's so hard to work with data that is fragmented,” Netz said.

Data is often named and labelled poorly, Netz said, which can lead even the most qualified experts to struggle to use it to its full potential.

“If I took the smartest person in the world, IQ 180, just graduated from MIT, and I brought them in front of the data lake with 100,000 tables, and I gave them a business question, they will struggle,” Netz told ITPro.

Microsoft is looking to Power BI to solve this problem with semantic models, which help to describe, define, and act as a “reference manual” for dense data, Netz explained.

On top of this foundation, AI can begin to process and reason through institutional knowledge, beyond basic functions such as analytics or reporting.

At launch, the natural language function within Database Hub is powered by Microsoft’s default OpenAI models, but Netz added that in the near future customers will be able to pick their own including those by Anthropic or others.

“I think the world of BI is moving to the same place,” Netz said.

“Instead of the human users going report by report, page by page, visual by visual, we are going to see the humans going to the Copilot and asking the question, and letting the AI be the one who is visiting the reports, reading what’s on the pages, the visualization, synthesizing the answers, and giving the answer back.”

Using Ontology for reliable agentic AI

Throughout his conversation with ITPro, Netz stressed the importance of institutional knowledge for making sure AI agents act in a similar manner to human workers.

He gave the example of an airline, which has data like any other organization but will also have unique organizational processes that need insider knowledge to run properly.

“I can imagine how British Airways works, but I don't really know how British Airways works,” he said.

“When a British Airways employee is joining British Airways, they're getting orientation, somebody explains how the organization is working. And there's a lot to know, like how do you assign flight crews to the plane? How do you make sure that the baggage is offloaded online and getting the right carousel? There is a process.”

This is where ontology comes in, providing machine-readable mapping of an organization’s data, its properties, and the relationships between entities within databases in the context of the organization.

Microsoft uses Ontology in Fabric IQ for this functionality, as Netz explained.

“The ontology is describing your business in the most operational aspect of the business. How are things related? What are the rules? What are the policies? What are the actions you can take? What can I do with a plane? I can assign a crew to a plane. I can divert the plane. I can ground a plane. I can load the plane with luggage.”

An understanding of organizational ontology is essential for agentic AI, Netz added. Microsoft expects 1.3 billion agents to go live by 2028 and leaders will need to implement organizational guardrails to prevent them from misbehaving.

Open source standards such as MCP and agent to agent (A2A) can impose specific behavioral limits for agents, as well as set ethical constraints such as how tools can be used and who is allowed to access specific data.

But because these rules are invoked at the execution stage, they aren’t a comprehensive method for keeping AI agents predictable.

Netz told ITPro that Microsoft thinks agents will have human managers and lower-level work carried out by database administrators will be automated to an increasing extent.

As organizations deploy AI agents such as OpenClaw, they will need to clearly map out business goals and values through ontology to ensure tasks are completed in line with all expectations.

“Think back to British Airways: they want to be profitable, but they also want to have customer satisfaction and they also want to take off and land the plane safely.

“Which one takes precedence over which? How much are you willing to sacrifice in safety to get more profitability? These are things that MCP tools don't tell you, because these are the things that we expect employees to understand… you don’t do these things, you don’t take risks with planes not landing safely just to make a few more dollars. But AI may not know that.”

FOLLOW US ON SOCIAL MEDIA

Follow ITPro on Google News and add us as a preferred source to keep tabs on all our latest news, analysis, views, and reviews.

You can also follow ITPro on LinkedIn, X, Facebook, and BlueSky.