Fabric Weekly

Issue 15

20th October 2026

🤖 Welcome to issue #15 of Fabric Weekly!

Data agents get smarter this week, with advanced query generation now GA, while new previews land across governance, ingestion and Fabric Apps. There's also loads of great community content, including a lot centred around governance, which is becoming increasingly important to think about as more people (and agents...) start querying, and acting on, your data.

On the official side:

From the community:

Finally, some UK events in the next couple of weeks:

  • Tuesday 20 October (tonight), online - Devon & Cornwall Power BI with Kevin Chant on end-to-end CI/CD for Fabric, followed on Tuesday 27 October by Marjolein Opsteegh on working with AI to build dashboards people understand.
  • Thursday 22 October, online - Microsoft Fabric UK with Sandeep Pawar and Virginia Roman on what's next for semantic models in data agents.
  • Thursday 22 October, London - London Fabric & Power BI with a FabCon recap and Dhaval Sadaiya from Zebra BI.
  • Thursday 29 October, Manchester - PBIMCR with Martin Brierley and Ian Pike on their FabCon Europe takeaways.

🤖 Copilot, AI & Agents

  • Fabric Operations Agent: Who Approves, and Whose Permissions Run by Matthias Falland Matthias Falland looks at the operations agent in Fabric Real-Time Intelligence (generally available since June 2026), which watches eventhouse or ontology data, reasons over it and recommends actions in Microsoft Teams for a person to approve. He covers when it fits versus a simple Activator rule, why it shouldn't run under an admin identity, the known issue with ontologies bound to semantic models, and the limits under workspace outbound access protection.
  • Retiring the Fabric data agent custom tool in Microsoft Foundry: move to Fabric IQ by Amir Jafari · 5 min read The Fabric data agent custom-tool integration in Microsoft Foundry is deprecated and stops being supported on 6 November 2026. Foundry agents using it must move to the Fabric IQ (OneLake Catalog) tool, currently in preview, which lets you pick published data agents from the catalog, connect more than one per Foundry agent, and calls each through an MCP endpoint. The post gives migration steps for both the Foundry portal and the SDK/REST API.
  • Advanced NL2SQL and advanced DAX generation in Fabric data agents (Generally Available) by Misha Desai · 7 min read Advanced NL2SQL and advanced DAX generation for Fabric data agents replace one-shot query generation with an agent that explores the schema, tests joins and looks up real filter values before writing SQL or DAX. Microsoft's internal benchmarks show smart-answer accuracy rising from 78.8% to 90.9% for SQL and from 89.15% to 91.86% for DAX, with lower latency. The improvements, plus a move to GPT-5.6 Terra, roll out to the standard runtime by default between 2 and 10 November; schema descriptions and topics for SQL sources stay in preview. Microsoft recommends re-testing existing agents.
  • Ontology Agent - Ep.572 by Mike Carlo, Tommy Puglia · 1 min read Tommy Puglia and Mike Carlo try the Fabric IQ ontology agent (preview), which can draft an ontology from a lakehouse but still can't finish the job it offers to do, and explain why they're keeping it out of any workspace a client depends on.
  • Fabric IQ MCP Server: What AI Agents Can Read From Power BI by Matthias Falland Matthias Falland on the Fabric IQ MCP Server (generally available since September 2026), a remote server that lets AI agents and MCP clients such as GitHub Copilot query Power BI reports and semantic models as the signed-in user. He covers when it fits (business definitions already live in governed semantic models) and its limits: no service principal support for unattended jobs, no joins across models, answers only as fresh as the last refresh, and no availability in Power BI-only home regions or sovereign clouds.

📊 Reporting & Insights

  • Autoselect the Latest Month in a Slicer (Another Attempt) by Dane Christian Belarmino Dane Christian Belarmino tries another way to automatically select the latest month in a Power BI slicer while still showing the actual month and year, after explaining the limits of the usual workarounds: a calculated column that relabels the latest month as 'Latest Month', and bookmarks that don't update when a new month arrives.
  • Continuous axis for any Date table in Power BI by Marco Russo · 6 min read Marco Russo shows how to get the continuous X-axis that Power BI's auto date/time hierarchy offers at month level on your own Date table: build a user hierarchy from a Years-categorised top level down to a date column marked PaddedDateTableDates (an undocumented data category), set in TMDL view, Tabular Editor or via the Power BI Modeling MCP server. He also covers the tooltip quirks and which visuals are affected.
  • The new Gantt chart in Power BI by Valerie Junk · 1 min read Valerie Junk walks through Power BI's new native Gantt visual (preview): how to turn it on, which data it needs, how to build and format it, and how to highlight overdue tasks in red.
  • Monitoring The Impact Of Full-Text Indexing On Semantic Model Refresh In Power BI by Chris Webb · 2 min read Chris Webb looks at the refresh cost of Power BI's new per-column full-text indexing (which enables the TextContains() and TextSimilarity() DAX functions), using a Profiler trace to time the index build. On a street-name column with 73,206 distinct values it took about 1.4 seconds, but he warns longer strings and more columns could slow refreshes significantly. Reports can't use the indexes yet, but he thinks they're worth exploring for 'chat with your data' scenarios.
  • From API to Data Agent: Germany’s power grid with Real-Time Intelligence in Microsoft Fabric by Rene Fuerstenberg · 1 min read René Fürstenberg builds an end-to-end Real-Time Intelligence project on German power grid data from Fraunhofer ISE's free Energy-Charts API: Azure Functions push events into an Eventstream and Eventhouse (with materialized views to handle late corrections), feeding a three-page Real-Time Dashboard, seven Activator email alerts and a data agent over dedicated KQL functions. He shares the pitfalls (time zones, a missing Boolean type, agents giving inconsistent answers until analyses moved into KQL functions), notes that the data agent needs a paid F2+ capacity, and links the full code on GitHub.
  • Is Power BI Dead? Absolutely Not. by Aleksi Partanen A Short clipped from Aleksi Partanen's September 2026 feature summary panel, in which Oskari Heikkinen answers 'Is Power BI dead?' with 'Absolutely not' and picks out the new Power BI Desktop experience where you describe a report in natural language and an agent builds it, which he expects to change how many people create reports.
  • Fabric App functions to use secure server-side custom logic (Preview) by Sunitha Muthukrishna · 3 min read Fabric App functions (preview) add TypeScript serverless functions, built on Fabric User Data Functions, to Fabric Apps for logic that shouldn't run in the browser: business rules, calls to external APIs, secrets management, and access to downstream resources under the app's own identity (application authentication). The post walks through scaffolding with the Rayfin CLI, building with GitHub Copilot, debugging locally and deploying.

⚙️ Data Engineering

  • Metadata-Driven Pipeline in Fabric: One Pipeline, Many Tables by Matthias Falland A short film from Matthias Falland's Fabric Friday series making the case for metadata-driven pipelines: instead of one near-identical pipeline per source table, a Lookup, ForEach and Copy activity driven by a config table handle them all, so a new source is a new row and a fix is made once. It pays off when the sources look alike.
  • You Probably Don't Need So Many Master Pipelines by Aleksi Partanen A Short clipped from Aleksi Partanen's September 2026 feature summary panel: Fabric pipelines are getting pipeline-level dependencies with run conditions (preview), so a pipeline can wait for others to finish without everything being wrapped in one master pipeline, something Aleksi says he's wanted since the Azure Data Factory days.
  • Snowflake CDC in Fabric Copy Job: Changes From a Read-Only Source by Matthias Falland Matthias Falland on the new public-preview CDC support for Snowflake in Fabric Copy job, which ingests inserts, updates and deletes from read-only Snowflake databases without needing a stream grant. He covers when it fits (consuming a Snowflake account you don't administer), when it doesn't (change tracking off, schedules outside the retention period), and when Snowflake mirroring is the better tool.
  • File Data Ingestion with Materialized Lake Views in Microsoft Fabric (Preview) by Haresh Gowri Sankar · 4 min read File data ingestion for materialized lake views is now in preview: a CREATE MATERIALIZED LAKE VIEW ... USING OneLake_Files statement ingests CSV or Parquet files (including nested folders) straight into a bronze Delta table, so the bronze layer joins the same declarative pipeline, lineage, scheduling and data-quality checks as silver and gold. It supports DYNAMIC or FIXED schema modes and adds a __filepath__ column for tracing each row back to its source file.
  • Fabric Notebook Toolkit - Ep.571 by Mike Carlo, Tommy Puglia · 1 min read Tommy Puglia and Mike Carlo discuss Microsoft's Fabric Notebook Toolkit (preview), which lets an AI agent you already use drive Fabric notebooks from a CLI, and treat it as a tool to learn rather than a pattern to roll out to every workspace.

🔐 Governance & Security

  • Even a Global Admin Couldn't See All the Connections by Aleksi Partanen A Short clipped from Aleksi Partanen's September 2026 feature summary panel on the new admin APIs for connections, which will let tenant admins inventory every connection in the tenant, something Teemu Multanen notes even a global admin couldn't see before. Heini Ilmarinen calls it 'what we've all been waiting for'.
  • Fabric Policies: Granular governance at scale (Preview) by Daniel Marcovich · 4 min read Fabric Policies, now in preview, let admins define what actions are allowed, for whom and where, at tenant or capacity scope. Policy Sets can restrict item creation for any Fabric item type per capacity (Power BI items aren't enforced yet) and limit security-sensitive workspace settings such as inbound/outbound access protection and customer-managed keys to a named group, with external data sharing controls coming soon. Advanced policies are allow-based, so anything no rule matches is blocked; policies never grant roles or access, and are managed through the Policies Center, the audit log and REST APIs.
  • Microsoft Fabric Policies: What Works and What Doesn't by Aleksi Partanen Aleksi Partanen and Teemu Multanen set up Fabric Policies (preview) from scratch and test them with a second account: creating a policy set at tenant or capacity scope, building allow rules from security groups, item types and workspaces, and why your first allow rule blocks everyone else. They note that policies aren't yet available in every region, hit a few preview issues along the way, and recommend locking down the workspace your policies live in and trying them in a test tenant before production.
  • Workspace-level private link for SQL database in Microsoft Fabric (Preview) by Suk Kaur · 1 min read Workspace-level private link is now in preview for SQL database in Fabric, letting you restrict inbound access to specific workspaces via Azure Private Link rather than across the whole tenant, so connections must come through an approved private endpoint. The post outlines setup through the workspace's Inbound network settings and suggests it for teams handling sensitive data or with differing compliance needs.
  • Why OneLake Security Cannot Restrict a Fabric Contributor by Matthias Falland Matthias Falland explains why giving a team Contributor and fencing off a sensitive folder with OneLake security doesn't work: roles combine as a union and OneLake security has no deny, so it can't take back the write access Contributor already grants. Start people as Viewer and grant upward instead. A clip from his episode on workspace roles, OneLake security and capacity.
  • #8 Assessing Your Organizational Team’s Fabric Maturity by Paul Turley · 1 min read Part 8 of Paul Turley's Fabric & Power BI Essentials series: how to assess whether your team is ready for Fabric before switching features on. He suggests assessing admins, engineers, analysts and business users separately across skills, tooling and data culture, using Microsoft's adoption-roadmap maturity levels and aiming first for a documented, repeatable middle stage. Ask rather than assume (governance workarounds reveal where process is too slow), then turn the gaps into a focused, prioritised training plan.
  • My workspace governance improvements (Generally Available) by Sergei G. · 1 min read My workspace governance features are now generally available, including in government clouds. Two new admin REST APIs, callable by a Fabric admin or a service principal, grant and revoke an admin's 24-hour temporary access to a user's My workspace, so this can be automated rather than done by hand in the Admin portal. The post also reminds admins that orphaned My workspaces can be restored within 30 days via the Restore Workspace API, with a new name and a new admin.
  • Managing Power BI gateways in a large tenant: the jobs the portal has no button for by Lukas Novacek · 1 min read Lukas Novacek introduces GatewayLens, a free, open-source (MIT) Windows console for Power BI and Fabric data gateway admins in large tenants. It covers the jobs the portal can't do in bulk: a status wall with offline alerts (including to Teams), searching connections by server, account or user, updating credentials across many on-premises connections at once, handing a leaver's access to a successor, and flagging unused or duplicate connections for cleanup before a cluster hits the 1,000 data source limit. It uses the Fabric REST API, signs in as you, stores no passwords and backs up before every delete; bulk credential updates are on-premises only for now.
  • Fabric Viewer Role: Why SQL Works and Spark Does Not by Matthias Falland Matthias Falland explains why the Fabric workspace Viewer role isn't simply 'read-only': it grants read through the SQL analytics endpoint, where row and object security are applied, but Spark reads the underlying files and never passes that gate. SQL read (ReadData) gives a filtered view; file read (ReadAll) gives the files themselves, so Viewer alone isn't enough for Spark users. A clip from his episode on workspace roles, OneLake security and capacity.

💰 Management & Cost

  • Asking the right Fabric CI/CD architecture questions by Kevin Chant · 1 min read Kevin Chant sets out the questions to ask before designing a Fabric CI/CD architecture: whether to use Git integration and which provider (Azure DevOps or GitHub, and any planned migration), which items are in scope and whether to deploy natively with Deployment Pipelines (which have no built-in approvals) or orchestrate externally, who will own and maintain the process long-term, how deployments will authenticate and handle secrets (service principal secrets vs workload identity federation or OIDC, Key Vault network restrictions), and whether there are ambitions for agentic workflows.
  • API for GraphQL – The Dialog You See Once by Matthias Falland A Fabric Friday deep dive from Matthias Falland on Fabric API for GraphQL, starting with the single sign-on vs saved credentials choice, which applies to every source on the API and decides whether row-level security still holds. He covers the request limits, multi-source fan-out queries, and the trap that a deployed API using saved credentials can keep reading development data without an error, with four demos and community Q&A.
  • Dipping my toes into Azure DevOps Deployments by Reitse Eskens · 5 min read Reitse Eskens shares what he ran into deploying Fabric code from development to production with Azure DevOps pipelines rather than Fabric deployment pipelines (whose deployed items can still point at development): the prerequisites (Key Vault, an app registration with Fabric permissions, a service connection, a secret variable group), why the service connection needed workload identity federation to read Key Vault secrets, and a warning that the app registration ends up owning every item, while some items such as the lakehouse still need it to sign in and refresh its token.
  • Microsoft Fabric Capacity Throttling and Smoothing Explained by Matthias Falland A Fabric Friday deep dive from Matthias Falland on why a capacity can be throttled when nothing appears to be running. He explains smoothing across 2,880 thirty-second timepoints a day (interactive work over 5 to 64 minutes, background over 24 hours), follows one heavy notebook through the Capacity Metrics app, and covers the throttling stages, why surge protection isn't a sizing fix and why halving the SKU doubles the percentage, before community Q&A.
  • Microsoft Fabric: Throttled with nothing running? #Shorts by Matthias Falland A Short from Matthias Falland: a capacity throttled when nothing is running is paying for work already done, spread out by smoothing, so size from the smoothed line rather than the spike. A clip from his capacity throttling and smoothing episode.
  • Who Can Move Your Fabric Workspace to Their Capacity? by Matthias Falland Matthias Falland explains why capacity admins don't all have the same power over your workspace: admins of P and EM (Power BI Premium/Embedded) capacities can pull any workspace onto their capacity, while F and A capacities also require workspace admin rights. He also notes that a move cancels running jobs and that only seven item types survive a cross-region move. A clip from his episode on workspace roles, OneLake security and capacity.
  • What Is the New F0 Capacity in Fabric? by Aleksi Partanen A Short clipped from Aleksi Partanen's September 2026 feature summary panel, in which Heini Ilmarinen explains the zero-provisioned F0 capacity and on-demand billing, listed as coming soon: no upfront compute cost, but a new need to watch governance and costs closely. Admins will be able to set a spending limit in CU-hours over a rolling 24-hour window, after which new operations pause.

📅 Community & Events

Get Fabric Weekly in your inbox every Tuesday

Sign up to receive the email every week and keep on top of all the announcements.

Don't forget to confirm your subscription. We have a double opt-in sign up process to prevent spam.

We will never provide your details to any third parties. We don't spam.

Fabric Weekly Logo