Skip to content
Contoso Pharmacy Copilot Lab

Roadmap

This roadmap lists useful next work. These items are not in this lab. The current lab stays tenant-free, synthetic, and local-first.

Fabric workspaces and deployment

Section titled “Fabric workspaces and deployment”

Not in this lab:

  • Create Fabric workspaces for dev, test, and prod.
  • Connect workspaces to Git integration.
  • Use Fabric deployment pipelines for staged promotion.
  • Evaluate fabric-cicd or Fabric REST API automation for deployment.
  • Add GitHub environments, approvals, and OIDC for deployment workflows.

Microsoft Learn describes Fabric deployment pipelines and Fabric Git automation as ALM options. Product support and preview status can change, so this work needs a dated design when phase two starts.

Power BI Copilot and Fabric data agents

Section titled “Power BI Copilot and Fabric data agents”

Not in this lab:

  • Publish the semantic model to a Fabric workspace.
  • Enable Fabric Copilot for Power BI in a tenant and capacity.
  • Test Power BI Copilot against the published model.
  • Create a Fabric data agent over the semantic model.
  • Compare local question-tester answers with Fabric data agent answers.

As of the docs checked on 2026-09-23, Fabric data agents can use Power BI semantic models as sources and require paid Fabric or Premium capacity. This lab avoids that requirement.

Fabric ontology

Section titled “Fabric ontology”
Preview feature

Not in this lab:

  • Model Contoso Pharmacy concepts in Fabric ontology, part of Fabric IQ preview.
  • Bind ontology entities to OneLake or semantic model sources.
  • Test whether ontology context improves data agent grounding.

This is a phase two research item because ontology is a governed business context layer, not a repo-only TMDL change.

Monitoring and evaluation at scale

Section titled “Monitoring and evaluation at scale”

Not in this lab:

  • Store repeated agent answer runs in a central evaluation store.
  • Track regressions over time by model version, prompt version, and agent version.
  • Add dashboards for answer quality, validation failures, PR cycle time, and Copilot review findings.
  • Add human adjudication workflows for ambiguous business questions.

The current lab uses local JSON run files and deterministic grading so learners can understand the pattern first.

Governance

Section titled “Governance”

Not in this lab:

  • Sensitivity labels and data loss prevention policies.
  • Endorsement, certification, and ownership workflows.
  • Lineage across workspaces and downstream reports.
  • Tenant settings for Copilot, cross-geo AI processing, and data agents.
  • Environment-specific secrets, approvals, and audit logging.

These topics need an organization’s real governance model and should not be simulated with fake tenant IDs.

Lab-site enhancements

Section titled “Lab-site enhancements”

Planned improvements for the public site:

  • Add screenshots captured on Windows 11 with synthetic data only.
  • Add short recordings for the presenter path.
  • Add accessibility tests for keyboard navigation, contrast, headings, and screen-reader labels.
  • Run a weekly link check against GitHub, VS Code, Microsoft Learn, and APM docs.
  • Add a release note when product UI labels change.
  • Add a Windows full-track lastVerified date after a clean end-to-end run.

Repo and agent improvements

Section titled “Repo and agent improvements”

Possible future repo work:

  • More model defects for advanced labs, such as calculation groups and role-playing dates.
  • A richer measure-review agent that checks DAX style and performance.
  • A package-ready APM example for one skill, one agent, and one prompt.
  • More sample cloud-agent PRs for facilitator fallback.
  • Better diff annotations for PBIR visual rebinding.