Azure Developer CLI Ships AI Agent Extension: Local Run, Debug, Monitor, and Deploy AI Agents End-to-End from Terminal to Microsoft Foundry

On March 31, 2026, Microsoft published the Azure Developer CLI (azd) March 2026 roundup covering versions 1.23.7 through 1.23.13 — seven releases in a single month. The headline feature is a new azure.ai.agents extension that creates a full lifecycle for AI agents directly from the command line.

New AI Agent Extension Commands:

1. azd ai agent run — Starts an agent locally for development and testing
2. azd ai agent invoke — Sends messages to an agent, whether running locally or deployed to Microsoft Foundry
3. azd ai agent show — Displays container status and health of deployed agents
4. azd ai agent monitor — Streams container logs in real time for debugging

This creates a complete inner-loop experience: developers can build, test, and iterate on AI agents locally before deploying them to Microsoft Foundry with a single command. The extension supports the full development lifecycle from code to cloud.

GitHub Copilot Integration: azd init now offers a Set up with GitHub Copilot (Preview) option that uses a GitHub Copilot agent to scaffold projects. Key details: it checks for dirty working directories before modifying files, and prompts for Model Context Protocol (MCP) server tool consent upfront — showing that MCP is now deeply integrated into Microsoft developer tooling.

AI-assisted error troubleshooting is also new: when a command fails, azd offers a multi-step flow where a Copilot agent can explain the error, suggest guidance, apply a fix, and retry the failed command — all without leaving the terminal.

Container App Jobs for Agent Workloads: azd now deploys Azure Container App Jobs (Microsoft.App/jobs) through existing containerapp configuration. The Bicep template determines whether the target is a Container App or Container App Job automatically. This is significant for AI agents that run as background jobs rather than persistent services.

Additional Infrastructure Improvements:

• Remote build fallback: when ACR remote build fails, azd automatically falls back to local Docker/Podman build
• Local preflight validation catches Bicep parameter issues before deploying to Azure
• Configurable per-service deployment timeouts via --timeout flag
• Automatic pnpm and yarn detection for JS/TS agent projects
• Azure SRE Agent resource type now correctly displayed in provisioning output

Market Context: This positions Azure as the first major cloud to offer a dedicated CLI-based AI agent development and deployment workflow. While AWS launched its Agent Plugin for Serverless (also this month), Azure azd approach is more comprehensive — covering the full lifecycle from local development through production monitoring.