[Initiative] 🧠 Improve azd foundation for Copilot CLI + Azure scenarios

[kristenwomack]

Problem statement

azd already handles the full lifecycle from init to deploy, and developers rely on it daily. This pillar is about making that experience even better: faster provisioning, broader test coverage, measurable quality for AI-assisted flows, and new Copilot-powered scenarios that save developers real time.

There's meaningful room to improve. Provisioning and deployment can be parallelized to reduce wait times. Test coverage is at 58% and climbing, but we want to push past 70% with better PR-level tooling. As AI-assisted features like agentic init grow, we need systematic ways to measure and improve their quality. And there are clear opportunities to expand what Copilot can do inside the CLI.

If the foundation isn't solid and consistently improving, every feature built on top of it (extensions, hosted agents, the developer toolkit) inherits that fragility. This pillar is about making azd faster, more reliable, and smarter.

Vision

azd is the fastest, most reliable way to go from code to cloud on Azure, and Copilot makes every step smarter. Developers experience measurably faster deployments, never hit regressions that should have been caught, and discover AI-assisted workflows that save them real time on real tasks.

Who this helps

• Developers shipping daily get faster azd up through parallel provisioning and deployment. Example: a multi-service app that currently provisions resources serially gets DAG-based parallelism, cutting wait time significantly.
• Contributors to azd get immediate, automated feedback on whether their PR maintains quality. Example: open a PR and see a coverage diff showing exactly what your change affects, with playback tests gating merge.
• Developers using AI-assisted flows (agentic init, Copilot commands) get eval-backed quality signals so those experiences improve with every release. Example: CI runs eval scenarios for agentic init and reports pass/fail rates, catching regressions before they ship.
• Anyone getting started with Azure benefits from a reliable first experience. Example: azd up on a template project completes predictably fast, building trust from the very first run.

Goals (in scope)

• Performance: Cut azd up end-to-end provisioning and deployment with parallelism. Establish baseline latency in April, then measure improvements as we optimize.
• Quality: Raise code coverage from 58% to 70%+. Ship playback tests as PR preflight gates. Introduce coverage diff tooling on PRs.
• AI quality measurement: Define eval scenarios for agentic flows and integrate eval tooling into CI to report scenario success metrics.
• AI experience: Define and ship 1-2 new Copilot-powered commands inside azd beyond existing agentic flows.

Non-goals (out of scope)

• Rewriting the provisioning engine from scratch; this is optimization, not replacement
• Building or owning an eval platform; azd is a consumer of eval tooling
• "Making everything AI": Copilot expansion is scoped to 1-2 high-value scenarios
• Full E2E integration test suite rewrite

Success criteria

• azd up latency reduced by a measurable percentage (baseline established in April with recently introduced improvements)
• Code coverage reaches 70%+ with coverage diff tooling active on PRs
• Playback tests run as PR preflight gates in CI
• At least one eval scenario runs in CI and reports scenario success metrics
• At least one new Copilot-powered command ships beyond existing agentic flows