Building better platforms with Azure, GitHub, Terraform and AI
I have been building and using agent skills across Claude Code, Codex and GitHub Copilot, alongside plugins such as Ponytail and Superpowers. I am using Microsoft Agent Package Manager (APM) to manage that collection through a shared manifest. My agent-setup repository holds the configuration and a copy of the global lockfile. The manifest declares what … Read more
Send GitHub Copilot CLI telemetry to Azure Application Insights using an authenticated OpenTelemetry Collector on Azure Container Apps, then query it with KQL.
Retrieve Azure Log Analytics workspace keys with Azure CLI
AI code review becomes noisy quickly on busy pull requests. These are the patterns I would use for managed comments, finding identity, severity, evidence and merge gates.
Update one pull request comment from GitHub Actions instead of creating duplicates after every push, using a stable marker, ownership checks and a simple upsert pattern.
Part 1 covered the consumption side of an AI gateway: rate limits, token quotas, usage attribution and the telemetry needed to understand them. The next runtime decision is whether a request should be allowed through at all, followed by whether that consumer is entitled to use the model it asked for. Those are separate controls, … Read more
How I would use Azure API Management to control AI request rates and token consumption, attribute usage, emit useful telemetry and handle backend throttling.
Azure Policy and API Management solve different parts of AI governance in Azure. This post goes into the actual policies and APIM XML behind that two-layer model, covering network controls, approved models, token quotas, content safety and observability.
An Azure AI Landing Zone should make the governed route the easiest route for delivery teams. This post covers how Azure API Management, Azure Policy, identity, networking, quotas, telemetry and clear ownership boundaries work together to control AI consumption without turning the platform team into an approval bottleneck.
Measuring AI-assisted engineering by activity alone misses the point. Active users, token usage, generated lines of code, and agent sessions are useful signals, but they do not tell you whether the work was reviewable, trusted, safe, or worth the cost.
This post looks at what platform teams should measure instead: workflow success, review effort, guardrail failures, context reuse, cost per useful outcome, and developer confidence.