TL;DR
FabCon 2026 produced two kinds of news in Atlanta: features you can use in production now and previews you should only pilot. Data agents, Direct Lake on OneLake, materialized lake views, Maps and Oracle and SAP Datasphere mirroring were generally available in March 2026. Runtime 2.0, extended mirroring and the Copilot Studio link for data agents then reached GA in August and September 2026. Operations agents, Graph and planning in Fabric IQ (July 2026) are GA too as of October 30, 2026, while Database Hub and the Fabric IQ ontology remain previews. If your team runs Power BI on Fabric, adopt the GA items now; if it runs many Azure databases, pilot Database Hub with one named owner. Before any agent goes live, fix OneLake permissions and semantic models, because every agent inherits both.
Key Takeaways
Database Hub launched in early access at Atlanta in mid-March 2026 and reached preview at FabCon Europe in September 2026.
Fabric data agents were GA in March, and their Copilot Studio integration went GA in August 2026.
Fabric IQ adds an ontology and planning on top of semantic models; planning went GA in July 2026, while the ontology is still in preview.
Runtime 2.0 moved from preview to GA in August 2026 with Spark 4.1, Delta Lake 4.2 and Python 3.13.
OneLake mirroring for Oracle and SAP Datasphere was GA in March, and extended mirroring went GA in September 2026.
Migration assistants for Data Factory, Synapse and Azure SQL opened in public preview; the SQL database assistant is GA as of October 30, 2026.
Watch on YouTube
Why Did FabCon 2026 Matter for Kanerika? | Recap & Key Takeaways
Kanerika’s team recaps FabCon 2026 first-hand: the Atlanta announcements that matter for data teams and what they change for Fabric projects.
The CFO’s Question After Atlanta
Picture a data platform lead, back from Atlanta in mid-March 2026, presenting a slide of about 30 announcements. They run from a new Database Hub to remote MCP servers.
The CFO reads it once and asks one question: “Which of these can we use this quarter?” The slide cannot answer that. Some items were generally available on day one, and others were previews, early access programs or promises for “the coming weeks.”
Seven months on, that question has a clear answer. This article sets out what Microsoft announced at FabCon 2026 and checks each item against Microsoft Learn on October 30, 2026. It closes with a plain call for every item: adopt it now, pilot it or watch it.
What Was FabCon 2026?
FabCon 2026 was Microsoft’s third annual Microsoft Fabric conference, held in Atlanta in mid-March 2026 alongside SQLCon. In his keynote post of March 18, 2026 , Arun Ulag wrote to “our 8,000 attendees” across the two events.
Putting a database conference next to an analytics conference was, in effect, the message. Microsoft wants operational databases and analytics on one platform, so both can be managed one way. That idea sits behind the biggest launch of the week, the Database Hub. It also explains why SQL teams now have a reason to watch Fabric news.
If you are new to the platform, Microsoft Fabric is a software-as-a-service analytics suite. It puts data engineering, warehousing, real-time analytics and Power BI on one shared data lake called OneLake, and our Microsoft Fabric guide covers the basics.
The platform was not new to buyers by then. Microsoft said more than 31,000 customers used Fabric about 2.5 years after general availability, and it named The Coca-Cola Company among them.
A second edition followed in Barcelona in late September 2026, and we cover what changed there further down.
From Data Platform to Agent-Ready Foundation
The big idea at Atlanta was simple: AI agents give weak answers without business context. Fabric’s new job is to supply that context. Microsoft groups this under one name, Microsoft IQ, which combines Work IQ, Foundry IQ and Fabric IQ. Fabric IQ is the data part, and our Fabric IQ vs Foundry IQ comparison shows where the line between them falls.
It helps to picture the platform as a stack of five layers. Operational databases sit at the bottom, while OneLake holds the shared data above them. The engines (Spark, SQL, Real-Time Intelligence and Power BI) then work on that data. Fabric IQ adds business meaning on top, and agents use all of it to answer questions or take action.
Each of Atlanta’s four big shifts lands on one of those layers. At the database layer, Database Hub pulled operational databases closer to Fabric. At the OneLake layer, mirroring gained more sources, and Azure Databricks gained the ability to read OneLake data through Unity Catalog. Fabric IQ gave business meaning a home in its ontology, and at the top, agents became real products once data agents reached general availability.
A sales question shows why the layers matter. Suppose a regional manager asks an agent why margin fell in the Southeast. The agent needs Fabric IQ to know what “margin” means and the semantic model to compute it. It also needs OneLake to hold the numbers and the source database to send them. If any layer is wrong, the answer is wrong, however good the model is.
FabCon 2026 Announcements at a Glance: Then and Now
A conference announcement is a promise with a status attached, so this table follows each promise from March to the end of October. The March column comes from Microsoft’s Atlanta keynote post. We then checked the right-hand column against Microsoft Learn’s What’s new in Fabric page on October 30, 2026, and against Microsoft’s Barcelona post.
The table covers Atlanta items only. For releases that had nothing to do with the event, see our tracker of every Microsoft Fabric update released in 2026 .
Status Scorecard: March to October 2026
Announcement Status in March 2026 Status on October 30, 2026 Source
Database Hub Early access Preview (from Barcelona) Atlanta and Barcelona posts, Learn Oracle and SAP Datasphere mirroring GA GA Atlanta post Extended mirroring (change data feed, views) Announced, Snowflake first, paid option GA, September 2026 Atlanta post, Learn Unity Catalog reads from OneLake Public preview OneLake catalog federation in Azure Databricks listed as preview Atlanta post, Learn OneLake security GA “in the coming weeks” Members and data experience GA, September 2026 Atlanta post, Learn Fabric Runtime 2.0 Preview GA, August 2026 Atlanta post, Learn Materialized lake views Announced as GA GA entry on Learn, September 2026 Atlanta post, Learn Fabric IQ ontology Announced; MCP server “soon” in preview Ontology in preview; Fabric IQ MCP Server GA, September 2026 Atlanta post, Learn Planning in Fabric IQ Announced GA, July 2026 Atlanta post, Learn Direct Lake on OneLake GA GA Atlanta post Graph in Fabric GA “in the coming weeks” GA; no preview label on Learn Atlanta post, Learn Fabric data agents GA GA; Copilot Studio integration GA, August 2026 Atlanta post, Learn Operations agents Announced GA; some add-ons still in preview Atlanta post, Learn Fabric MCP servers Local GA; remote in public preview Fabric Core MCP Server GA, September 2026 Atlanta post, Learn Migration assistants (Data Factory, Synapse, Azure SQL) Public preview SQL database assistant GA, August 2026; Data Factory still preview Atlanta post, Learn
Read the table from its right-hand column, since that is where the news is. GA means Microsoft supports the feature for production use. Preview means the feature can still change and carries no GA support commitment, so treat it as something to test.
The Major FabCon 2026 Announcements, Explained
The table shows where each item stands, and this section explains what each one does, with an example and the team that should care most.
Database Hub and the SQLCon Announcements
Database Hub is one place to manage operational databases across Azure and beyond. When it launched, it covered six engines, listed below. Agents help with routine operations, while a person stays in the loop to approve what they suggest.
Engine in Database Hub Typical use
Azure SQL Relational application databases Azure Cosmos DB NoSQL and globally distributed apps PostgreSQL Open-source relational workloads MySQL Web and application back ends Arc-enabled SQL Server SQL Server running on premises or in other clouds Databases in Fabric SQL databases that live inside Fabric itself
Think of a team that runs dozens of Azure SQL databases, two Cosmos DB accounts and an on-premises SQL Server estate. Today each one has its own portal and its own alerts, so problems surface in many places. Database Hub aims to put them in one view, with agents that flag problems and suggest fixes.
Database Hub started in early access in March. Then, at Barcelona in September, Microsoft moved it to preview, “building on its introduction” at Atlanta. Database administrators and platform teams should care most, especially those with mixed estates.
Microsoft also announced a savings plan for databases. It promises “up to 35%” against pay-as-you-go pricing on select services, but the footnote says savings are estimated at between 0% and 35%. Until your bills prove otherwise, treat 35% as a ceiling, not a forecast.
OneLake: More Mirroring, Shortcut Transformations and Open Interop
OneLake gets outside data in three ways, and Atlanta improved all of them. Mirroring copies a source into OneLake and then keeps the copy in sync. Shortcuts point at data where it already lives, without copying it, as our guide to OneLake shortcuts explains. Interoperability lets another platform’s engine read the same tables.
On mirroring, Oracle and SAP Datasphere sources became generally available. SharePoint lists and Dremio arrived in preview, and SharePoint list mirroring reached GA in September 2026. Azure Monitor was named as coming. Extended mirroring adds change data feed and views, starting with Snowflake as a paid option, and Learn lists it as GA in September 2026. The SAP Datasphere source belongs to SAP Business Data Cloud , which our explainer covers.
Shortcut transformations went GA, which turns files reached through a shortcut into Delta tables without a separate pipeline. An Excel-to-Delta option also arrived in preview.
Interop also moved in two directions, toward Databricks and toward Snowflake. As of October 30, 2026, Azure Databricks can read OneLake data through Unity Catalog, though Learn still lists that feature as preview. A Databricks team could then query the same OneLake tables a Power BI model uses, without making a copy.
Microsoft also described its Snowflake interoperability as generally available. For a team sharing tables between Snowflake and Fabric, GA means that link is supported in production.
Security came last but matters most once agents arrive. In March, Microsoft said OneLake security would be GA “in the coming weeks.” Learn now shows its members and data experience as GA in September 2026. Our post on OneLake security in Microsoft Fabric covers how its roles work.
Fabric Analytics: Runtime 2.0, Materialized Lake Views and Maps
For Fabric data engineering teams, Runtime 2.0 was the biggest change of the week. Its March preview ran on Spark 4.x, Delta Lake 4.x, Scala 2.13 and Azure Linux 3.0 as the base. Later, in August 2026, it went GA with Apache Spark 4.1, Delta Lake 4.2, Python 3.13 and Java 21.
That is a major version jump, so jobs written for Spark 3 may behave differently. As of October 2026, Runtime 2.0 is the default for new workspaces, and older ones can switch under Spark settings in Workspace settings. Per Microsoft’s Runtime 2.0 guide , an Environment item can also pin Runtime 2.0 for only the notebooks attached to it. Run your heaviest notebooks that way in a test workspace before you switch production pools.
Materialized lake views let you define a table as a SQL query, and Fabric keeps the result up to date. They suit medallion pipelines in a Fabric lakehouse , where bronze data is cleaned into silver and then summarized into gold. Microsoft called them generally available in March, and Learn added a GA entry in September 2026.
Maps in Fabric was announced as GA for location analytics in March. ArcGIS Maps for Fabric followed to GA in September 2026, which helps teams that already rely on Esri layers. Our Fabric and Esri ArcGIS webinar covers asset intelligence in more detail.
Talk to Kanerika
Map FabCon 2026 to Your Fabric Platform
Walk through your Fabric workspaces, OneLake roles and semantic models with Kanerika and get a short list of what to adopt, pilot and fix first.
Schedule a Demo → Fabric IQ: Ontology, Planning and Semantic Models
Fabric IQ is the layer that tells an agent what your data means. Its ontology defines business entities, such as customer, product or plant, and how they relate, then maps them to tables in OneLake. An agent that reads the ontology knows that “active customer” has one agreed definition.
Atlanta also announced planning in Fabric IQ, for budgets, forecasts and scenarios built on the same governed data. Microsoft said an MCP server for Fabric IQ would “soon” enter preview.
Two technologies power Fabric IQ, according to Microsoft’s keynote post: Power BI semantic models and Graph in Fabric. Semantic models reach more than 35 million active users and hold the business definitions agents need. So Direct Lake on OneLake , GA since the event, matters: those models now read OneLake tables directly. Graph in Fabric, promised as GA “in the coming weeks,” carries no preview label on Microsoft Learn’s graph overview as of October 2026. Only its use as a data agent source is still marked preview.
As of October 2026, the status is mixed. Planning went GA in July 2026, while the ontology is still in preview. The Fabric IQ MCP Server and Fabric IQ in Microsoft 365 Copilot, by contrast, both reached GA in September 2026. Our Microsoft Fabric IQ guide walks through the ontology in more detail.
Case Study
70% Less Investigation Time with One Governed Ontology on Fabric
A global pharma enterprise moved KPI logic out of separate Power BI dashboards into one governed Fabric IQ ontology on OneLake, standardizing 85+ KPIs and cutting investigation and reconciliation time by 70%.
Read the Case Study → Fabric Data Agents and Operations Agents
Fabric now has two kinds of agents, and they do different jobs. A data agent answers questions in plain language over data you choose, such as a lakehouse, a warehouse or a semantic model. Teams use them as domain analysts, especially for questions a dashboard was never built for.
Data agents were GA at Atlanta, and their integration with Copilot Studio went GA in August 2026. So a data agent can now sit inside a wider agent built in Copilot Studio . Our guide to Microsoft Fabric data agents covers setup and limits.
An operations agent, though, works the other way round. It watches real-time data and acts when a condition is met, without waiting for a question. Per Microsoft Learn’s operations agent guide , it checks each rule every 5 minutes and sends a Teams message by default. As of October 2026, that guide carries no preview label, while newer add-ons such as investigator insights remain in preview.
A logistics team shows the difference. A planner asks the data agent which carriers missed delivery windows last week and gets a table back. The operations agent, by contrast, watches the live shipment stream and alerts the hub manager the moment late arrivals cross a threshold.
Reports can act too, through translytical task flows . Per Microsoft’s translytical task flow documentation , a Power BI report can write data back, start a Teams workflow or call Azure OpenAI. A sales manager could update a discount and send an approval request to Teams without leaving the report.
Developer Tools: MCP Servers, Git and Open Source
The Model Context Protocol (MCP) is an open standard that lets AI assistants call tools and data sources. At Atlanta, the local Fabric MCP server went GA. It is an open-source server that connects coding assistants such as GitHub Copilot to Fabric. A developer can then ask the assistant to inspect a workspace from the editor.
Remote MCP, which entered public preview, is different: a cloud-hosted engine that lets AI agents take authenticated actions in Fabric. Since those agents act, not just read, scope each remote connection as governance check 2 describes. For example, an agent that drafts pipeline fixes could write to one development workspace and only read everywhere else. Separately, the Fabric Core MCP Server reached GA in September 2026, according to Learn.
The other developer news was smaller but useful:
Selective branching in Git integration, which lets a developer branch only the items they need. Branch workspaces and selective branching went GA in September 2026.
Agent Skills for Fabric and Fabric Jumpstart, released as open source starting points.
The Fabric Extensibility Toolkit, announced as GA, for partners and teams building their own workloads.
Migration Assistants for Data Factory, Synapse and Azure SQL
Microsoft opened migration assistants for Azure Data Factory, Azure Synapse Analytics and Azure SQL in public preview at Atlanta. The Data Factory and Synapse assistant moves pipelines and artifacts like Spark pools and notebooks into Fabric. The SQL assistant imports schemas through DACPACs and uses AI to find and fix compatibility issues.
Seven months later, the two halves have moved at different speeds. The Migration Assistant to SQL database went GA in August 2026. Data Factory assessment and migration is still in preview, so plan a manual check of each migrated pipeline.
Picture a team with a few hundred pipelines. The assistant can move them, but each should still match the old pipeline’s output on last month’s data before the old one is retired.
These tools matter most when a team has years of Azure data work behind it. Our guides to Azure to Fabric migration and moving SSIS packages to Fabric cover the parts the assistants do not.
What Changed After Atlanta: FabCon Europe in Barcelona
Barcelona was the follow-up, and Microsoft’s September 28, 2026 post reported nearly 5,000 attendees and more than 130 sessions. Fabric’s customer count had grown from more than 31,000 in March to more than 40,000, and Power BI passed 425,000 customers. Database Hub moved to preview there, and Fabric IQ in Microsoft 365 Copilot reached GA. Our FabCon Europe 2026 recap covers the full Barcelona list.
Which FabCon 2026 Announcements to Adopt Now, Pilot or Watch
Not every announcement deserves the same response, so sort each one by two questions: is it GA, and how much damage could a mistake do? GA items with low risk are ready now, because Microsoft already supports them in production. Previews with clear value need a pilot and an owner, and vague timelines go on a watch list.
Call Announcement Why
Adopt now Direct Lake on OneLake GA since March; Power BI models read OneLake tables without import copies Adopt now Data agents on a governed semantic model GA, with Copilot Studio integration GA since August 2026 Adopt now Oracle and SAP Datasphere mirroring GA since March; replaces hand-built extract jobs Adopt after tests Runtime 2.0 GA since August 2026, but a major Spark upgrade needs regression tests Pilot Database Hub Preview; start with one database team and one named owner Pilot Migration assistants SQL assistant GA, Data Factory still preview; pilot one workload first Pilot Fabric IQ ontology Workload in preview; model one business domain first Pilot Planning in Fabric IQ GA since July 2026; trial one budget cycle first Pilot Graph in Fabric GA; new to most teams, so test one use case first Watch Unity Catalog reads of OneLake Still preview seven months on; wait for GA
Picture the CFO from the opening again. The honest answer to “what can we use this quarter?” is the four rows marked adopt. Then add one or two pilots that have an owner and an end date.
Should You Invest in Fabric Because of FabCon 2026?
FabCon is a reason to sequence Fabric work, not on its own a reason to buy Fabric. If you already run Power BI on Fabric capacity, most of the adopt list costs little beyond setup time. Teams not yet on Fabric gain more from a phased plan than from chasing announcements, and our Microsoft Fabric adoption guide sets one out.
Governance and Cost Checks Before You Turn On Agents
An agent can only see what its user can see, so a permission mistake quickly turns into a wrong or leaked answer.
The order matters more than the speed. When a team fixes roles first, switching on data agents is a short project. When it skips that step, it can spend weeks explaining why an agent showed salary data to the wrong person. Run these checks before the first agent reaches business users:
Fix OneLake security roles first. Agents inherit permissions, so clean up broad workspace access before anyone asks an agent a question. An operations agent, for instance, acts with its creator’s permissions, per Microsoft Learn.Scope every MCP connection. Decide which workspaces each MCP server can reach and whether it may write or only read.Treat previews as previews. A preview carries no GA support commitment, so keep it away from processes that cannot tolerate change.Check who needs a license. On F64 or larger capacity, users with a free license can view Power BI content with the viewer role. Creating Power BI items in shared workspaces still needs Pro, per Microsoft’s Fabric licensing page .Model the savings plan at 0%, not 35%. Microsoft’s own footnote gives a range of 0% to 35%, so budget for the low end.
Most of these steps are ordinary governance work, although agents raise the stakes. Our Microsoft Fabric governance guide covers workspace and role design. The guide to Microsoft Fabric pricing explains capacity sizes such as F64, which matches a P1.
How Kanerika Helps Teams Act on FabCon 2026
Kanerika is a Microsoft Fabric Featured Partner and a Solutions Partner for Data and AI (Analytics). Our team also covered the Atlanta event first-hand in the recap video near the top of this page. Our client work now centers on what Atlanta put on stage: governed semantic models, Fabric IQ ontologies and agents that answer from them.
Case Study: One Governed Ontology for a Global Pharma Company
One global pharma case study shows what that looks like when it runs in production. The company’s KPI logic lived inside separate Power BI dashboards, so the same metric could show different numbers in different reports. Analysts had to reconcile figures before anyone could act on them.
Kanerika built one governed Fabric IQ ontology on top of the company’s OneLake data. It modeled more than 40 entities and relationships and standardized more than 85 KPIs across Power BI. The ontology was then extended to Fabric AI and Copilot, so questions in plain language draw on the same definitions as the dashboards.
The published result was 70% less investigation and reconciliation time. It is the Fabric IQ ontology announced at FabCon, used for real work rather than a demo.
Life sciences teams face this KPI problem often, because one metric can feed both a regulatory report and a sales review. Our post on Microsoft Fabric in healthcare and life sciences covers that context in more depth.
For teams sorting their own adopt, pilot and watch list, we start with a short review of permissions and semantic models. We then pick one agent use case that can prove value within a quarter, such as a finance data agent over a governed model. The rest of the FabCon list waits until that first use case works.
Kanerika Service
Microsoft Fabric Services
Kanerika, a Microsoft Fabric Featured Partner, helps teams plan, migrate to and govern Fabric, from OneLake and semantic models to Fabric IQ and data agents.
Explore Microsoft Fabric Services The Bottom Line
Atlanta set the direction for Fabric: databases, OneLake, business meaning and agents on one platform. By October 30, 2026, the data foundation, data agents, Graph, operations agents and planning in Fabric IQ were GA. Database Hub and the Fabric IQ ontology were still previews. The next step is to pick your four adopt items and one pilot, then fix permissions before any agent goes live. FabCon 2026 announced one platform for databases, data, meaning and agents, and most of it has shipped since.
Frequently Asked Questions
What is FabCon, and what is it for?
FabCon is Microsoft’s annual conference for Microsoft Fabric, its data and analytics platform. It is where Microsoft announces new Fabric features, shares its roadmap and runs sessions for data engineers, analysts and architects. The 2026 Atlanta event was the third FabCon and ran alongside SQLCon, which brought Microsoft’s database teams into the same week.
When and where was FabCon 2026 held?
FabCon 2026 took place in Atlanta in mid-March 2026, alongside SQLCon. Arun Ulag’s keynote post, dated March 18, 2026, addressed 8,000 attendees across both events. A second edition, FabCon Europe, followed in Barcelona in late September 2026, with nearly 5,000 attendees and more than 130 sessions, according to Microsoft’s September 28 post.
What changed in Fabric Runtime 2.0?
Runtime 2.0 is a major upgrade of the Spark environment in Microsoft Fabric. It entered preview in March 2026 on Azure Linux 3.0 and went GA in August 2026 with Apache Spark 4.1, Delta Lake 4.2, Python 3.13 and Java 21. Spark jobs written for older runtimes should pass regression tests before teams switch production pools.
How do the new migration assistants help Azure Data Factory and Synapse users?
The Fabric migration assistant for Azure Data Factory and Synapse Analytics moves existing pipelines and artifacts like Spark pools and notebooks into Fabric. It opened in public preview at Atlanta, and Data Factory assessment and migration is still in preview. A separate assistant for SQL databases imports schemas through DACPACs and went GA in August 2026.
How much can the savings plan for databases cut costs?
Microsoft says the savings plan for databases can save up to 35% against pay-as-you-go pricing on select services. Its own footnote estimates savings at between 0% and 35%, depending on the service and usage. Budget for the low end of that range, and treat 35% as a ceiling until your own bills confirm a higher figure.
Does Microsoft Fabric use AI?
Yes. Fabric includes data agents that answer questions in plain language and operations agents that act on real-time data. Fabric IQ gives these tools business context through an ontology, it now works inside Microsoft 365 Copilot, and MCP servers let outside AI assistants work with Fabric. Data agents and operations agents are both generally available.
Is Microsoft Fabric in demand?
Yes, adoption has grown quickly. Microsoft reported more than 31,000 Fabric customers in March 2026 and more than 40,000 by September 2026, with The Coca-Cola Company named among them. Power BI has more than 425,000 customers, and Microsoft says Fabric semantic models reach more than 35 million active users, which keeps Fabric skills relevant for data teams.
Are FabCon 2026 preview features safe for production?
Preview features can change and carry no GA support commitment, so keep them out of processes that cannot tolerate change. Pilot them on one non-critical workload with a named owner and an end date. Use GA items such as data agents and Direct Lake on OneLake in production, and revisit previews when Learn lists them as GA.
What were the biggest announcements at FabCon 2026?
The biggest announcements were Database Hub for managing six database engines in one place, Fabric IQ with an ontology and planning, and data agents reaching general availability. Microsoft also added mirroring for Oracle and SAP Datasphere, previewed Runtime 2.0, released Fabric MCP servers and opened migration assistants for Data Factory, Synapse and Azure SQL.
Which FabCon 2026 announcements are generally available now?
Data agents, Direct Lake on OneLake, materialized lake views, Maps and Oracle and SAP Datasphere mirroring were GA in March 2026. Runtime 2.0 and the Copilot Studio integration for data agents reached GA in August 2026. Extended mirroring, the Fabric IQ MCP Server and the Fabric Core MCP Server followed in September 2026.
How was FabCon Europe 2026 different from FabCon Atlanta?
Atlanta in March 2026 was the launch event, where Microsoft introduced Database Hub, Fabric IQ planning and GA data agents. Barcelona in September 2026 was the follow-up. It moved Database Hub to preview, made Fabric IQ in Microsoft 365 Copilot generally available and reported more than 40,000 Fabric customers, up from 31,000 in March.
What is the Database Hub in Microsoft Fabric, and is it generally available?
Database Hub is one place to manage Azure SQL, Azure Cosmos DB, PostgreSQL, MySQL, Arc-enabled SQL Server and databases in Fabric. Agents help with routine operations while a person approves their suggestions. It is not generally available yet. It launched in early access at Atlanta in March 2026 and moved to preview in September 2026.
What is Fabric IQ, and why does it matter for AI agents?
Fabric IQ is the part of Microsoft Fabric that gives data its business meaning. Its ontology defines entities such as customers, how they relate and which tables hold them. Agents that read it use one agreed definition for each metric, so their answers match the dashboards. The ontology is still in preview; planning went GA in July 2026.
What is planning in Fabric IQ?
Planning in Fabric IQ is a capability announced at FabCon 2026 for budgets, forecasts and scenario models. It works on the same governed semantic models and business definitions that Fabric IQ already uses, so plans and actuals share one source. It reached general availability in July 2026, so pilot it on one budget or forecast cycle first.
What is the Model Context Protocol (MCP) in Microsoft Fabric?
The Model Context Protocol is an open standard that lets AI assistants call tools and data sources in a consistent way. In Fabric, MCP servers let an assistant or agent work with Fabric items. The local Fabric MCP server went GA at Atlanta, a remote version entered preview, and the Fabric Core MCP Server reached GA in September 2026.
How do Fabric data agents differ from BI tools and operations agents?
A BI report shows answers to questions someone planned in advance. A Fabric data agent answers new questions in plain language over chosen data, such as a semantic model or a lakehouse, and it is generally available. An operations agent watches real-time data and acts when a condition is met. Both are now GA.
