The AI Wire

Lead dispatchtechcommunity.microsoft.com ▸ azure· filed 2 Oct 2026

Federated Agent Factory: Microsoft's Reference Architecture for AI Agents Across Entra Tenants

Microsoft's reference architecture for multi-tenant AI agent governance uses a federated agent factory, keeping agent identities and data within business-unit tenants while a governing tenant sets standards and operational views. It specifies runtime enforcement via Azure API Management for pro-code agents and Agent 365/Purview for Copilot Studio agents, with policy-as-code for per-action authorization. Several components remain in preview, including the dedicated AI Gateway tier and multitenant agent management.

Read the dispatch →

No. 087

Latest on the wire75 dispatches · page 01 / 08

techcommunity.microsoft.com ▸ models· filed 2 Oct 2026

Microsoft adds streaming transcription and two text-to-speech models to Azure AI Foundry

Microsoft has introduced three speech models to Azure AI Foundry: one that streams partial transcripts before a speaker finishes, and two text-to-speech models that trade off latency and cost. The streaming model is priced at $0.54 per hour until the end of the year, while the voice models run $22 and $15 per million characters. Benchmark comparisons come from vendor-reported figures, and the pricing of transcription for the long term remains unspecified.

techcommunity.microsoft.com ▸ product· filed 27 Sept 2026

Azure Container Apps Sandboxes: per-agent microVMs with egress controls

Azure Container Apps Sandboxes runs untrusted agent workloads in hardware-isolated microVMs that start in under a second, making it practical to give each task its own machine. Outbound traffic is forced through a default-deny proxy that can enforce allowlists, inject authentication headers outside the sandbox, and send requests to a customer webhook for approval. Microsoft says preview usage passed one million sandboxes created per day; pricing is per vCPU-core-second and GiB-second.

github.blog ▸ product· filed 27 Sept 2026

GitHub's HydraFusion preview orchestrates coding models at runtime

GitHub has released Project HydraFusion as a research preview in Copilot, a runtime orchestrator that selects among single-model, cascade, and critique-revise workflows for each coding request. Offline evaluations against Claude Opus 5 on three benchmarks show mixed quality—one gain, two slight shortfalls—with estimated costs down 36–67%, based on GitHub's own assumptions rather than production measurements. The preview is limited to well-scoped first-turn tasks in autopilot mode, with multi-turn sessions listed as the next focus.

microsoft.com ▸ product· filed 27 Sept 2026

Microsoft launches Copilot Managed Runtime public preview with SDK and CLI

Microsoft has made Copilot Managed Runtime available in public preview, adding an SDK and CLI so apps built with external tools can run inside a Microsoft 365 tenant with Entra identity, organizational policies, and a centralized admin inventory. The runtime handles provisioning, deployment, versioning, and governance while keeping code Git-backed and editable. Pricing and production timelines are not yet specified, and consistency of policy enforcement across third-party integrations remains to be tested.

techcommunity.microsoft.com ▸ product· filed 25 Sept 2026

Microsoft Foundry Insights public preview automatically analyzes agent traces for unknown failure patterns

Microsoft Foundry Insights is now in public preview, automatically scanning agent traces in Azure Monitor Application Insights to surface recurring failure patterns teams did not anticipate. It flags findings with categories, severity, and likely causes, and complements rather than replaces monitoring and evaluation. Evidence quality and scan results depend on telemetry; an empty scan does not prove an agent is problem-free.

learn.microsoft.com ▸ product· filed 18 Sept 2026

Microsoft Foundry Agent Endpoints: Stable URLs, Version Pinning, and No Traffic Splitting

Microsoft Foundry agents receive stable endpoints at creation and route traffic by a version selector, but the default "always use latest" policy means any new version goes live immediately unless an administrator pins a specific version. There is no traffic splitting, so rollbacks are all-or-nothing, and production teams must rely on manual pinning. Entra ID-only authorization and an opt-in public endpoint for Microsoft 365 are among the operational constraints.

azure ▸ product· filed 10 Sept 2026

Fabric Data Agent Integration with Copilot Studio Reaches General Availability as a Tool-Based Experience

Microsoft has made its Fabric data agent integration with Copilot Studio generally available as a tool-based experience, replacing the earlier connected-agent approach. Copilot Studio agents can now select Fabric IQ Data MCP alongside other tools, with the orchestrator deciding when to call Fabric and execution and permission checks staying inside Fabric. Teams that built against the preview integration will need to adjust their agents accordingly.

azure ▸ tooling· filed 10 Sept 2026

Microsoft Introduces MCP Server for Azure Update Discovery, With Only Lifecycle Details

Microsoft has introduced the Release Communications (MRC) MCP server for AI-powered discovery of Azure product updates, but the announcement only supplies lifecycle-stage categories and omits technical and access details. The page does not say whether the server exposes the same update feed as the Azure Updates site, nor how it handles preview-versus-launched distinctions. As a result, the practical value for MCP-driven tooling remains unspecified.

azure ▸ product· filed 10 Sept 2026

Claude Fable 5.1 arrives in Microsoft Foundry alongside restricted Mythos 5.1

Anthropic has released Claude Fable 5.1 in the Microsoft Foundry model catalog and restricted Claude Mythos 5.1, describing both as the same underlying model with different safeguard configurations. The company says the main advance is sustained performance on multi-step, tool-use workflows, with fewer overconfident errors and better reporting of incomplete work. Anthropic recommends testing on realistic workloads and simplifying prompts built for earlier models, though these claims have not been independently verified.

azure ▸ azure· filed 10 Sept 2026

Azure API Management AI Gateway adds OpenTelemetry observability and cost limit policies

The AI Gateway tier of Azure API Management now emits OpenTelemetry logs, traces, token metrics, and estimated cost metrics, and supports calendar-based cost limit policies that can block requests when projected budgets are reached. The update also refines provider onboarding and expands Anthropic Messages compatibility. Because spend figures are estimates, teams using hard blocks should verify the gap between projections and final billing.