Microsoft Fabric set up so capacity, security and Power BI all line up.
We design Fabric workspaces and OneLake, build lakehouses, warehouses and pipelines, move Power BI and Synapse workloads in, and size capacity so one team's job doesn't slow everyone else.
Fabric IQ in Copilot, BigQuery mirroring (28 Sep 2026)
release watch
FabCon Europe 2026 announcements28 Sep 2026
DiskANN vector indexes28 Sep 2026
connects with
in short·20-second read
Cohash implements Microsoft Fabric for data and analytics teams.
We plan capacity and workspaces and design OneLake with lakehouses and warehouses. We build Data Factory pipelines and mirroring, and migrate Power BI and Azure Synapse workloads. Since FabCon Europe 2026 we also help teams use Fabric IQ in Microsoft Copilot and the new data engineering agent, and we check which features are generally available and which are still in preview.
you probably need this if
Reports slow to a crawl whenever a large pipeline runs on the same capacity.
Power BI is going onto Fabric capacity and nobody's sure which SKU you'll need.
The data sits in Azure SQL, Snowflake and BigQuery, and analysts copy it around by hand.
Capacity sizing report based on a measured trial week
Workspace and domain layout with roles mapped to Entra ID groups
Lakehouse design with a shortcut or mirroring decision per source
+ 2 more documents
who does it
Fabric data engineer (Lakehouse, Spark)Fabric analytics engineer (semantic models)Power BI developer
Microsoft Certified: Fabric Analytics Engineer Associate
Microsoft Certified: Fabric Data Engineer Associate
28 Sep 2026FabCon Europe 2026 announcements Now GA: Fabric IQ in Microsoft Copilot Chat and Cowork, and Google BigQuery mirroring. Still in preview: the data engineering agent, IQ sharing, observability and Database Hub. source ↗
28 Sep 2026DiskANN vector indexes Generally available for SQL database in Fabric, Azure SQL Database Hyperscale and Azure SQL Managed Instance. source ↗
today · Oct 22, 2026
the story · microsoft fabric estate
From the first call to a setup your own team can run.
Here's what usually prompts a call about your Microsoft Fabric estate, what we do about it and in which order, and where you end up.
01 · where it hurts02 · what we change03 · how it runs04 · where you end up
incident board · microsoft fabric estate3 open
openReports slow to a crawl whenever a large pipeline runs on the same capacity.
recurringPower BI is going onto Fabric capacity and nobody's sure which SKU you'll need.
blockedThe data sits in Azure SQL, Snowflake and BigQuery, and analysts copy it around by hand.
exposure
Capacity and workspace planningOneLake, lakehouse and warehouse…Pipelines and mirroringPower BI and Synapse migrationSecurity and governanceFabric IQ and Copilot setup
startrehearsehandover
Workload and capacity…
Workspace and OneLake design
Ingest and mirror
Security mapping
Migrate and compare
we are here
ready to hand over
We traced the Monday slowdown to specific refreshes, and those run off peak now.
Each Synapse pool gets a retirement date.
Capacity sizing report based on a measured trial week
Workspace and domain layout with roles mapped to Entra ID groups
chapter 01 · where it hurts
Why teams reach out
Reports slow to a crawl whenever a large pipeline runs on the same capacity.
Power BI is going onto Fabric capacity and nobody's sure which SKU you'll need.
The data sits in Azure SQL, Snowflake and BigQuery, and analysts copy it around by hand.
chapter 02 · what we change
How we solve it
Capacity and workspace planning
OneLake, lakehouse and warehouse design
Pipelines and mirroring
Power BI and Synapse migration
Security and governance
Fabric IQ and Copilot setup
chapter 03 · how it runs
Our approach
Workload and capacity assessment
Workspace and OneLake design
Ingest and mirror
Security mapping
Migrate and compare
chapter 04 · where you end up
What changes
We traced the Monday slowdown to specific refreshes, and those run off peak now.
Each Synapse pool gets a retirement date.
Capacity sizing report based on a measured trial week
Workspace and domain layout with roles mapped to Entra ID groups
what we do
Our Microsoft Fabric work
6 kinds of Microsoft Fabric work, each run by engineers who have done it on live tenants.
Capacity SKUs, workspace layout and domain structure sized from your real workloads, with monitoring so throttling is spotted before users complain.
Medallion layers in lakehouses, a Fabric warehouse where SQL teams need it, and shortcuts to data already in Azure, AWS or Google Cloud.
Data Factory pipelines, dataflows and database mirroring from sources such as Azure SQL, Snowflake and Google BigQuery.
Semantic models and reports moved onto Fabric capacity, and Synapse workloads rebuilt as Fabric items with tested results.
Workspace roles, OneLake security, sensitivity labels from Microsoft Purview and Entra ID groups mapped so access matches what people need.
Fabric data agents and Fabric IQ configured on curated semantic models, tested on real questions before they reach Copilot users.
capability 01 / 06in production
Capacity and workspace planning
Capacity SKUs, workspace layout and domain structure sized from your real workloads, with monitoring so throttling is spotted before users complain.
We pieced these together from Microsoft Fabric situations that come up a lot. They're patterns. You won't find a specific client in them.
case file 01
Power BI capacity throttling every Monday at 9 a.m.
composite · not a client story
Beforethe situation
Power BI capacity throttles each Monday when refreshes, dataflows and report traffic overlap. The data team wants to add lakehouses and notebooks on the same capacity, and finance wants a number before approving a larger SKU.
What we dothe work
We go through several weeks of Capacity Metrics app data and split interactive load from background load. Heavy refreshes move off peak and the worst dataflows become pipelines. Nobody picks a SKU until a trial F capacity has run a representative week with the new workloads.
Afterwhere you end up
We traced the Monday slowdown to specific refreshes, and those run off peak now. Finance gets a SKU figure taken from a measured week. The new notebooks don't slow the 9 a.m. reports.
case file 02
Synapse pools, Spark jobs and a Copilot pilot due next quarter
composite · not a client story
Beforethe situation
An analytics team runs Azure Synapse dedicated SQL pools and Spark jobs that feed Power BI. Leadership wants Fabric IQ in Copilot for the sales team next quarter, and the semantic models it would read have overlapping measures.
What we dothe work
We inventory pipelines, pools and Spark jobs, map each to a Fabric warehouse, lakehouse or notebook, and run old and new side by side until results match. The sales semantic model is cleaned up first, and Fabric IQ is tested on real questions from sales before anyone outside the pilot sees it.
Afterwhere you end up
Each Synapse pool gets a retirement date. Sales puts its Copilot questions to one cleaned-up semantic model, and nothing in preview goes to production until Microsoft calls it GA.
We don't start a step until the last one has produced something you can check.
step 01
Workload and capacity assessment
We measure current Power BI, Synapse and pipeline workloads, test a representative week on trial capacity, and read the Capacity Metrics app before picking a SKU.
step 02
Workspace and OneLake design
We lay out domains, workspaces, lakehouses and warehouses by team and environment. For data that already lives in Azure, AWS or Google Cloud, we use shortcuts.
step 03
Ingest and mirror
We mirror supported databases and use Data Factory pipelines for the rest. Each gets monitoring and alerts.
step 04
Security mapping
Before business users get access, we map workspace roles, OneLake security and Purview sensitivity labels to Entra ID groups.
step 05
Migrate and compare
We move Power BI and Synapse workloads in batches and run them next to the originals until the numbers match.
ai on microsoft fabric
Where AI fits on Microsoft Fabric, and where it needs guardrails.
Most of what Fabric does with AI goes through Fabric IQ, generally available now in Microsoft Copilot Chat and Cowork. Fabric data agents and Copilot in Power BI sit next to it, there's a data engineering agent in preview, and Microsoft ships several Fabric MCP servers as well. They all answer from semantic models and OneLake data. Get model quality and workspace permissions right first.
AI agentasks for data or an action
Scoped identityonly the access it needs
MCP serverthe one door into the platform
Logged, approveda person signs off on writes
Microsoft Fabricdoes the work
Microsoft publishes Fabric MCP servers: Fabric IQ MCP, the Fabric Core MCP Server and a local open-source server. The remote ones use Entra ID OAuth, and live operations run with the identity's Fabric permissions. We use dedicated identities with workspace roles, keep client approval on for writes and check Fabric audit logs.
Fabric IQ in Microsoft Copilot
Fabric IQ went generally available in Copilot Chat and Cowork at FabCon Europe 2026. It answers from your semantic models. If two measures overlap, you get conflicting answers. So we clean up the models it reads and test with real questions before Copilot users see it.
Data engineering agent
The data engineering agent is in preview and works with GitHub Copilot, VS Code, Codex and Claude Code. Until Microsoft calls it generally available, we only use it in development workspaces.
Fabric MCP servers
Fabric IQ MCP is remote and read-only, for Power BI reports and semantic models. Fabric Core MCP Server manages workspaces, items and permissions. The local, open-source Fabric MCP Server covers development and OneLake files. We pick per task. Core doesn't get production admin from us.
Operations agents and observability
Operations agents with pre-approved actions and root-cause analysis, and observability in Fabric, are in preview. We'd pilot them on non-critical pipelines, with pre-approved actions limited to reruns and alerts.
How data reaches OneLake and flows on to Power BI and Copilot
Databases and files come into OneLake through mirroring and Data Factory, while shortcuts point at data you keep elsewhere. Entra ID groups and Purview labels decide access. Power BI and Fabric IQ read from the same curated tables.
These are working documents, written against your configuration, so your team can run them without us.
report01
Capacity sizing report based on a measured trial week
yours to keep
document02
Workspace and domain layout with roles mapped to Entra ID groups
yours to keep
design03
Lakehouse design with a shortcut or mirroring decision per source
yours to keep
inventory04
Synapse migration inventory with side-by-side result comparisons
yours to keep
inventory05
Feature status register marking GA and preview items, reviewed each quarter
yours to keep
talent and delivery · microsoft fabric
Certified engineers, wherever you need them.
Some clients want a project team, some want a couple of engineers on theirs. Either works. Onsite, nearshore or offshore, with handovers across time zones that keep things moving.
Put these to anyone bidding on your Microsoft Fabric work. The ones who've done it answer differently from the ones who've read about it.
Q1
How will you size our Fabric capacity, and what data will you base it on?
Capacity is shared, so one team's notebook can throttle everyone's reports. Sizing from the Capacity Metrics app and a test week beats a guess from user counts.
Q2
Which features in your design are generally available today?
Fabric IQ in Copilot and BigQuery mirroring reached GA at FabCon Europe 2026, but the data engineering agent and Database Hub stayed in preview. Any production design should spell out which is which.
Q3
How will OneLake security, workspace roles and Purview labels line up?
When three permission layers disagree you get over-sharing or a pile of support tickets. Design the mapping to Entra ID groups once, then test it.
Q4
What will you rebuild from Synapse rather than lift across?
Dedicated SQL pool code, Spark configuration and pipelines don't all map one to one. Knowing which pieces need rework changes the timeline more than data volume does.
Q5
When would you use shortcuts or mirroring instead of copying data in?
Shortcuts to existing storage and database mirroring save extra pipelines, but they change latency, cost and who governs the source. We decide each source on its own.
01How do we size Fabric capacity?
Start from your current Power BI and data engineering workloads, then test a representative week on a trial or small capacity. We watch the Capacity Metrics app for throttling and background usage before recommending a SKU.
02Should we use Fabric or Databricks?
It depends on your team and what they run. Fabric fits organizations built around Power BI and Microsoft 365. Databricks fits heavy Spark, streaming and ML engineering. Some clients run both, with OneLake shortcuts pointing at Delta tables.
03Are the FabCon Europe 2026 features ready for production?
Some of it. Microsoft lists Fabric IQ in Copilot and BigQuery mirroring as generally available. The data engineering agent and Database Hub are still preview. We use preview features in development only, never in production.
04Do you migrate from Azure Synapse?
Yes. We inventory pipelines, SQL pools and Spark jobs, decide what maps to Fabric directly and what needs rework, and run old and new side by side until results match.
05What drives Microsoft Fabric cost?
Capacity size, mostly, plus OneLake storage. Overlapping refreshes, inefficient dataflows and notebooks with oversized Spark settings use capacity fastest. We read the Capacity Metrics app before recommending a SKU change.
services
Related services
Projects on Microsoft Fabric usually end up touching one of these, so they're worth a look.
