Enterprise AI Agents on Microsoft Fabric: From Demo to Production
Sixty seconds of demo. Six months of approvals. This workshop is about the six months.
One day on the architecture, security and cost decisions that take an agent on Microsoft Fabric, Copilot Studio and Microsoft Foundry from a working demo to production.
Most training stops at building an agent. This day is about getting one through security review: identity, isolation and cost.
Wednesday 18 November 2026Live virtual workshop 08:30 to 16:30 UK time, lunch 12:00 to 12:45
Most agent pilots do not stall on the model. They stall when security, risk and finance ask three questions: who does the agent see the data as, what happens when it is wrong, and where can the data go.
This workshop works through those questions layer by layer, in the product, using one business question traced across Fabric data agents, Copilot Studio and Microsoft Foundry.
You'll leave with the decisions, test evidence and configuration you need to take your own agent through review.
Why this day
One business question traced across Fabric data agents, Copilot Studio and Microsoft Foundry.
No tenant needed on the day: watch every test run live, then take the dataset and runbooks home to repeat them on your own Fabric capacity.
Where MCP fits: the Fabric IQ Data MCP tool, its authentication mode, and why User versus Maker decides who the agent sees the data as.
Demonstration-led
What you'll learn
Every layer is built and tested live on a synthetic insurance dataset, then handed to you as a lab pack.
01
You'll see built and tested
Identity and data access, tested with two users
Row-level and column-level security under an agent
Just-in-time admin access
Checks that catch a wrong answer
Tenant isolation settings
The cost and capacity screens
02
You'll take away
Synthetic dataset and expected answers
Step-by-step demo runbooks
Tenant isolation configuration pack
Cost paper and calculator
The five governance decisions checklist
One focused day
Workshop agenda
From the questions a security reviewer asks to the decisions that answer them.
The three questions every reviewer asks, and why a working demo does not answer them.
02
The Stack and Its Boundaries
Fabric data agents, Fabric IQ, Copilot Studio and Microsoft Foundry: what each one does, where they meet and which route fits which use case.
03
Identity and Data Access
User and maker authentication, row-level and column-level security, and a live test in which two users ask the same question.
04
Privileged Access
Running without standing admin rights: just-in-time elevation and the audit trail it leaves.
05
Accuracy as Architecture
A five-gateway framework for catching a wrong answer before a user sees it.
06
Tenant Isolation and Observability
The isolation decision matrix, the configuration pack, and what you can monitor across Foundry, Copilot Studio and Fabric.
07
Cost
Where agent costs appear, how to read the capacity and billing screens, and how to model your own workload.
08
Five Governance Decisions and Next Steps
The decisions that determine whether a pilot reaches production, a checklist for your first week back, and open Q&A.
Learning outcomes
Leave ready for your own review
Answer the three security review questions
Choose between Fabric, Copilot Studio and Foundry routes
Decide how the agent authenticates to data
Test identity with two users
Apply row-level and column-level security
Remove standing admin access
Design checks for wrong answers
Choose a tenant isolation model
Know what to monitor
Estimate agent cost for your workload
Who should attend?
For the people taking agents past the pilot
Working knowledge of Microsoft Fabric or Power BI is assumed.
Ideal for
Platform Engineers · Data and Solution Architects · Data Engineers · Fabric and Power BI Developers · AI Engineers · Consultants · Technical Leads
Also suitable for
Heads of Data · Data Platform Managers · Security and Risk Leads · Product Owners for AI initiatives
Live instruction Real experience
Meet your trainer
Leon Gordon
Microsoft Fabric Consultant, Trainer & Founder of Onyx Data
Leon works with organisations implementing Microsoft Fabric, with a focus on practical delivery, governance and operational visibility.
This workshop extends the session he presented at FabCon Europe 2026 in Barcelona, Building Enterprise AI Agents on Microsoft Fabric: Architecture, Security and Production Lessons, from a conference hour into a full day.
Launch cohort pricing
Choose the right booking
Every booking includes a VAT invoice, the complete live workshop and the lab pack.
Platform engineers, architects, data engineers and technical leads who have seen an agent demo and now need to get one through security, risk and finance review. It assumes working knowledge of Microsoft Fabric or Power BI. It is not an introduction to Fabric.
Is the workshop hands-on?
The day is demonstration-led. Leon builds and tests each layer live and you can question every step. You receive a lab pack with a synthetic dataset, the demo runbooks and a tenant isolation configuration pack so you can repeat the work in your own tenant afterwards.
Do I need a Microsoft Fabric tenant to attend?
No. You only need to join the live session. To repeat the labs afterwards you will need Fabric capacity and Copilot Studio licensing; the lab pack lists the prerequisites.
Does it cover Copilot Studio and Microsoft Foundry as well as Fabric?
Yes. The workshop follows one question through Fabric data agents, Copilot Studio and Microsoft Foundry, and focuses on the boundaries between them: identity, data access, isolation, observability and cost.
Is this an official Microsoft course?
This workshop is delivered independently by Onyx Data, a Microsoft Solutions Partner.
Can my company send multiple people?
Yes. A discounted booking is available for three delegates.
Is the workshop recorded?
The workshop is delivered live to encourage interaction and questions. Any supporting resources will be communicated after the session where applicable.
Value beyond the day
After the workshop
For every delegate
The lab pack
The dataset, runbooks and configuration pack are yours to keep, so you can repeat every test in your own tenant.
