The setup

• Best practices live in the repo
  We maintain a docs/best-practices.md file describing ALM rules, governance decisions, and severity levels (INFO, WARNING, CRITICAL, BLOCKER).
• An export pipeline creates the pull request
  An Azure DevOps pipeline exports and unpacks the Power Platform solution, commits the changes to an export branch, and creates a pull request into main.
• Power Automate reacts to the PR
  When the PR is created, a Power Automate flow is triggered automatically.
• Secrets are handled properly
  The Azure AI Foundry API key is stored in Azure Key Vault and referenced securely from the flow

What happens in the flow

When the pull request is created, the flow:

1. Reads the pull request context (title, description, branches)
2. Fetches the list of changed files
3. Fetches the best-practices.md document from the repo
4. Sends this information to Microsoft AI Foundry
5. Posts the AI’s review back as a comment on the pull request

The review is grouped by severity and meant to be read by developers, not enforced blindly.

If the AI flags anything as CRITICAL or BLOCKER, the flow also creates a Bug in Azure DevOps automatically — making the issue visible and traceable without blocking the pipeline.

Thinking ahead: multiple reviewers, not one AI

In this demo we use a single AI reviewer, but the design intentionally leaves room for more.

A natural next step is to introduce multiple specialized agents, for example:

• One agent focused on best practices and ALM rules
• One focused on security
• One reviewing changes from a user or product perspective
• One comparing the implementation to the linked user story or acceptance criteria

Power Automate remains the orchestrator, while Foundry provides the intelligence. Each agent has a clear responsibility, just like human reviewers do today.

Why this earns the badges

ACDC Craftsman
This solution demonstrates disciplined ALM practices: automated exports, pull requests, documented governance, secure secret handling, and traceability from change to review to bug.

Hipster
Using Microsoft AI Foundry as a first-class part of the PR workflow is about as current as it gets, not as a gimmick, but as a practical reviewer.

Power User Love
Power Automate orchestrates the flow, Azure DevOps provides the backbone, and pro-code HTTP calls fill the gaps. It’s low-code where it makes sense, and code where it matters.