At Ignite 2026 Microsoft announced Microsoft IQ. The question that follows is "what do we need to do to switch this on?"
Frontier models are smart, but they don't know your business. Think about how you onboard a new employee: you train them, give them the documents and policies, and show them the reports the business runs on. An agent without that context fills the gaps with assumptions, so the same question asked twice can get two different answers.
>> Microsoft: What's new in Azure at Ignite 2026
Most organisations already have that context, in documents, conversations and the semantic models behind their Power BI reports. The trouble, as the same post puts it, is that "it's scattered, so every new agent starts from zero".
Microsoft IQ is that onboarding for agents. Microsoft describes it as an enterprise context management platform, there "so you don't have to explain how your business works every time you build an agent". It brings together four IQs, alongside more than 350 third-party connectors.
|
How people work: documents, conversations, collaboration and past decisions |
Curated institutional knowledge: policies and authoritative documents |
|
Timely context from the web: news, images and videos |
The state of the business: governed data, business entities, relationships and rules |
Fabric IQ is the data behind your business. It answers "how much", "how many" and "compared to last year", which makes Fabric Microsoft's route for governed business data to reach an agent.
Fabric IQ is made from things you already build in Fabric. It helps to think of three layers: the data, the meaning on top of it, and the people and agents that consume it.
This is where the data lives. OneLake is the single store underneath everything in Fabric: lakehouses and warehouses shape data for analysis, and eventhouses hold real-time data. Data held elsewhere can be brought in with shortcuts and mirroring, without a migration project.
An agent pointed straight at this layer sees tables and columns. It can see a column called fct_sales.amt_net, but it has to guess whether that is what your finance team calls revenue.
This layer gives the data business meaning, and it is the layer Fabric IQ hands to an agent.
Microsoft has had this layer for more than a decade, in the semantic models behind Power BI reports. A semantic model holds the tables, relationships and measures that define what revenue or churn means. The Power BI semantic model is mature and well established. Other platforms arrived recently: Snowflake's semantic views became generally available in June 2025, and Databricks put metric views into public preview in May 2025.
The ontology is the next step. A semantic model typically describes one subject area. An ontology describes the business itself (customers, products, orders and how they relate) across your models and data stores.
>> Microsoft Docs: What is ontology (preview)?
What if your data sits with another vendor, such as Snowflake? The data itself can already reach Fabric through shortcuts and mirroring. For the definitions, Microsoft is working with Snowflake on Apache Ossie (incubating), an open standard for sharing business definitions between platforms. In Fabric it shows up as semantic views (preview), metrics defined directly on data in OneLake and built to be reused on other platforms. It is early days: the open-source converter for Power BI semantic models does not carry everything across, so for now expect to rebuild some definitions by hand.
For years the semantic layer had one consumer, the Power BI report. Now the same definitions reach people and agents wherever they work:
Sources: Ignite 2026 Book of News, Introducing the new Microsoft IQ and Fabric IQ at Ignite 2026
Fabric IQ gives people and agents answers they can trust, in the tools they already use. An agent that answers from your semantic model gives the same revenue figure as your reports, and gives it again the next time someone asks. People can get that answer in Copilot, Excel or Claude without opening a report or waiting for one to be built.
The context is also built once and reused. Definitions agreed in Fabric serve every agent that follows, so each new one starts from what the business already knows.
The preparation in the next section pays off before any agent is switched on: owned data, agreed definitions and tidy access make your existing reports more trustworthy too.
None of this asks you to start again. Power BI semantic models are already part of Fabric IQ, and the ontology binds to data where it sits: semantic models, lakehouses, warehouses, eventhouses, SQL databases and mirrored databases.
The old rule still applies though: rubbish in, rubbish out. As Arun Ulag puts it, "every agent you build is only as good as what it knows about your business". Good context reduces wrong answers but does not remove them, and an agent will repeat whatever your estate tells it with complete confidence. So before switching anything on, ask:
If the answer to any of those is "sort of", that is where the effort goes. Three areas need attention first.
Clean, catalogued data. An agent takes your data at face value. Duplicates, gaps and tables nobody owns turn straight into wrong answers. Every dataset an agent can reach needs a data owner who is accountable for its accuracy, and a data steward who keeps it clean and described day to day. Data with no owner is the first thing to fix or to keep away from agents. Record what you hold, who owns it and what state it is in where people and agents can find it.
Business definitions. The assistant can only answer with what the model defines. If there are three measures called Sales, Sales v2 and Sales (new), the answer will be as ambiguous as the model. Agree one definition for each measure with the business, and hold it in the semantic model with a clear name and description. Prep for AI in Power BI, itself in preview, is the place to start.
Security. When an agent acts on someone's behalf, as Copilot and the Claude connector do, it only sees what that person can already access, and row-level and object-level security (the rules that limit which rows and columns a person can see) still apply. That is the right design, but a chat box makes it far easier to find everything you have access to than browsing workspaces ever did. Review who has access to what before you switch anything on, and remove what people no longer need.
Zure is a Microsoft Fabric Featured Partner and has delivered one of the largest Fabric implementations globally. If you are asking what it takes to switch on Fabric IQ, there are two places to start depending on where your estate is today.
Fabric Readiness Assessment. For organisations that already run Power BI or Fabric. Five workshops cover your business domains, data sources, compliance, team structure and skills. Our report then scores your maturity, shows where your estate falls short on the three areas above and sets out how Fabric would be implemented. Most assessments take three to four weeks, depending on how much there is to cover.
Fabric IQ Proof of Concept. For organisations that are not on Fabric yet, or want to prove the idea on one domain first. Two or three of our people take one of your datasets from source to a working proof of concept you can look at and talk to. It ends at a semantic model, which is the point where Copilot, Claude or data agents have something to answer from. We agree the timeline once we have seen the data, and for one dataset it is typically around a month.
Microsoft IQ, announced at Ignite 2026, is Microsoft's enterprise context management platform for agents. It brings together Work IQ, Foundry IQ, Web IQ and Fabric IQ, alongside more than 350 third-party connectors, so every new agent doesn't start from zero.
Fabric IQ is the part of Microsoft IQ that gives agents governed business data: semantic models, ontology and semantic views on top of data in OneLake. It answers questions like "how much", "how many" and "compared to last year".
Not necessarily. Power BI semantic models are already part of Fabric IQ, and data held elsewhere can be brought in with shortcuts and mirroring without a migration project.
Three areas: clean, catalogued data with a named owner, one agreed definition for each measure in the semantic model, and a review of who has access to what.
The data can reach Fabric through shortcuts and mirroring. For business definitions, Microsoft is working with Snowflake on Apache Ossie, an open standard that shows up in Fabric as semantic views (preview). The converter doesn't carry everything across yet, so expect to rebuild some definitions by hand.