Microsoft AI Updates: November 4, 2026

1. GitHub Copilot Joins JetBrains Air’s Agent Picker, With Cloud Tasks Billed to the Copilot Plan

GitHub made GitHub Copilot available in JetBrains Air through the Agent Client Protocol (ACP), so developers can pick Copilot in Air’s agent picker, delegate coding tasks on their existing Copilot subscription, track several agent sessions in parallel, and review changes next to their code, either locally in a JetBrains IDE or in the cloud. Cloud tasks run through Air Teams from the IDE or a browser and draw on the Copilot plan rather than JetBrains AI credits; each runs in a Docker container on a fresh virtual machine with all traffic passing through a proxy and no state carried between tasks. They require a JetBrains account in an Air Teams-enabled business organization, a Copilot subscription, and an Air administrator who has enabled Copilot. GitHub lists several limits: Air may verify subscription or credit availability only when the cloud environment starts, and the task fails with an error in the session if either is unavailable; users cannot choose per task whether their personal or organization Copilot plan pays; Air’s governance and analytics features do not include Copilot usage; and Air administrators can disable Copilot through a governance policy but cannot set Copilot spending limits inside Air. The integration sits alongside the existing GitHub Copilot plugin for JetBrains IDEs rather than replacing it. Source

2. Copilot Usage Metrics API Now Separates Bring-Your-Own-Key and Enterprise Custom Model Usage

GitHub extended the Copilot usage metrics API so administrators can tell Copilot-hosted model usage apart from bring-your-own-key (BYOK) usage, which previously blended into the totals because local and custom models often share names with Copilot models. Local BYOK, where a developer points Copilot CLI or the GitHub Copilot app at their own provider key or at local models served by tools such as Ollama, Foundry Local, LM Studio and llama-server, appears as a per-user used_local_byok flag, totals_by_local_byok session, request, prompt and token counts broken out by byok_kind (local_managed, local_user, local_auto, remote_user) and surface, and daily, weekly and monthly active local BYOK user counts. Admin-configured enterprise custom models, which route through GitHub to the customer’s provider, are flagged with a new is_custom_enterprise_model boolean on each totals_by_client[].totals_by_model[] entry and cover every client, including IDEs, because GitHub observes them server-side. Existing totals do not change, since BYOK usage was already counted; local BYOK figures depend on client telemetry and are a lower bound when telemetry is off or a local provider does not report tokens, local BYOK is not broken out by model name, and there is no backfill for earlier report days. The data appears in enterprise and organization reports at user and aggregate level in 1-day and 28-day reports, for roles with the View Copilot Metrics permission when the Copilot usage metrics policy is enabled. Source

3. Azure Application Gateway for Containers Previews an MCP Gateway That Merges Several MCP Servers Behind One Endpoint

Microsoft put Model Context Protocol routing for Application Gateway for Containers into public preview through a new MCPRoute Kubernetes custom resource, letting platform teams publish Kubernetes-hosted MCP servers behind a stable hostname and path while handling TLS, JWT validation, Web Application Firewall protection, backend health checks and routing policy at the gateway. MCPRoute can route directly to a single MCP server or aggregate several independent servers behind one logical endpoint; in the aggregated mode the gateway merges their tool, resource and prompt discovery responses, namespaces capability names to avoid collisions, and sends each invocation to the backend that owns the selected capability. It also supports session affinity for MCP servers that keep in-memory session state, plus backend TLS and cross-namespace references. Two related previews landed the same day: centralized JWT validation through an AuthPolicy custom resource that checks a token’s signature, issuer, lifetime and configured audience, and private frontends reachable only through approved Azure Private Link private endpoints. The announcements give no pricing or general availability date. Source Source

4. Dependabot Can Now Keep Agent Skills, Prompts and Instructions Up to Date Through Agent Package Manager

GitHub added Agent Package Manager (APM) to Dependabot version updates, so repositories that declare AI agent skills, prompts and instructions as dependencies in apm.yml get pull requests when supported git dependencies have newer versions, with apm.lock.yaml regenerated when present. Setup is an apm entry in .github/dependabot.yml that names the directory containing apm.yml and an update schedule. The support covers version updates only, not Dependabot security updates, and is generally available on github.com for all GitHub plans, with GitHub Enterprise Server support coming in version 3.24. Source