Microsoft’s June 2, 2026 Build announcement wave is aimed squarely at developers. If your engineering teams could not clear their calendars for the keynotes, the session catalog is available at build.microsoft.com/sessions. This recap walks through the announcements that matter most for state and local government technology teams, with enough technical detail to help you decide which sessions to prioritize.
A quick caveat before we dive in: several of these capabilities are in public preview, planned for general availability, or likely to appear first in commercial cloud environments. Government developers running in Microsoft 365 Government or Azure Government should treat this as architecture education, not a deployment checklist, and confirm availability in their cloud, tenant, and region before planning production work.
Microsoft Foundry: the production agent platform fills in
The biggest theme this year is the maturing of Microsoft Foundry from a model and prototyping platform into a production agent platform. The June 2, 2026 Build Edition post is the best index of what shipped, what is in preview, and what is still planned.
Agent Framework reached stable building blocks. Microsoft Agent Framework now offers a stable agent harness with skills, memory, and middleware; stable multi-agent orchestration patterns including Magentic-One; and stable integrations with the GitHub Copilot SDK and the Claude Agent SDK. File system tools, memory tools, and a deep research agent landed in public preview. This matters because it gives teams orchestration primitives they no longer have to hand-roll.
Hosted agents are nearly GA. Hosted agents in Foundry Agent Service provide a managed, framework-agnostic runtime where every session runs in its own sandbox with dedicated compute, memory, and filesystem access. Per the Build Edition post, hosted agents are expected to reach general availability by early July 2026. They support two protocols: the Responses API for OpenAI-compatible stateful interactions, and an Invocations protocol for schema-free pass-through scenarios. Agents built with Agent Framework, the GitHub Copilot SDK, LangGraph, or other SDKs can deploy without rewrites. The new routines capability, now in public preview, lets an agent run on a timer or schedule, which is a natural fit for overnight case triage or daily compliance reporting.
Toolboxes and Voice Live. Toolboxes in Foundry, now in public preview, give an agent a single managed endpoint for tools, skills, and Model Context Protocol clients, with Foundry handling auth, lifecycle, and governance. Voice Live unifies speech recognition, text-to-speech, turn detection, interruption handling, avatars, and other real-time conversational features into one API. Voice Live is generally available for prompt agents, while hosted voice agents are in public preview for teams that want their own runtime or orchestration framework.
Memory and grounding. Memory in Foundry Agent Service, in public preview, now spans procedural, user, and session memory. Procedural memory helps agents learn how to do work across runs, and Microsoft cites +7 to +14 percent absolute success-rate gains on Tau-bench at near-baseline cost. On the grounding side, Foundry IQ consolidates retrieval behind one SLA-backed endpoint. Foundry IQ knowledge bases are generally available, Foundry IQ Serverless is in public preview, and Microsoft Web IQ provides sub-200 ms web grounding with zero data retention under limited access. For teams that have been hand-building retrieval-augmented generation pipelines, this removes a lot of chunking and indexing plumbing.
Publishing where people work. Foundry agents can publish directly to Microsoft Teams and Microsoft 365 Copilot, planned for general availability in June 2026, with identity, permissions, and policy flowing through automatically.
A practical first step for your team: the Foundry Toolkit for VS Code is now generally available. You can scaffold an agent from a template, debug runs locally with trace visualization, connect to Toolboxes, and deploy to Foundry Agent Service without leaving the editor.
GitHub Copilot: SDK goes GA, CLI gets a major refresh
The GitHub Changelog for June 2, 2026 is packed. Two items stand out for engineering teams.
The GitHub Copilot SDK is now generally available. Per the GA announcement, the SDK lets you embed Copilot’s agentic engine, including planning, tool invocation, file edits, streaming, and multi-turn sessions, into your own applications. It now ships in six languages:
| |
Key capabilities include custom tools and MCP server registration, fine-grained system prompt customization, OpenTelemetry tracing with W3C trace context propagation, and flexible authentication including GitHub OAuth, GitHub Apps, environment tokens, and bring-your-own-key for OpenAI, Microsoft Foundry, Anthropic, and other providers. The SDK is available to all existing Copilot subscribers, including Copilot Free for personal use, and to non-Copilot users through bring-your-own-key.
Copilot CLI got a substantial upgrade. The June 2, 2026 CLI release makes three features generally available:
- Rubber duck, a built-in critic agent the CLI can hand its current plan, design, implementation, or tests to for a structured second opinion before continuing.
- Prompt scheduling via
/everyand/after, for example/every 30m run the frontend testsor/after 2h create a summary of recent changes. Running either command with no arguments opens a schedule manager. - Voice input, which runs locally so recorded audio stays on your machine. GitHub Docs currently list English and Spanish dictation support.
An experimental redesigned terminal interface adds tabs for Issues, Pull requests, and Gists, plus accessibility-focused color modes such as high-contrast and colorblind. Screen reader support is on by default when a screen reader is detected. Opt in with /experimental on and update with copilot update.
Other notable Copilot changelog items from the same day: cloud and local sandboxes for Copilot entered public preview, Copilot code review for Azure Repos entered technical preview, the cloud agent gained scheduled and automated tasks, a new /chronicle view surfaces insights across agent sessions, Gemini models joined Copilot CLI, the cloud agent, the Copilot app, and the Copilot SDK, and MAI-Code-1-Flash began rolling out in GitHub Copilot starting with VS Code.
Azure infrastructure: silicon built for agents
On the infrastructure side, Microsoft announced early access preview for Azure Cobalt 200 Arm-based VMs on June 2, 2026. Cobalt 200 is built on the Arm Neoverse V3 compute subsystem, fabricated on TSMC’s 3nm process, and scales to 128 vCPUs with 3 MB of L2 cache per core and 192 MB of system-level L3 cache. Microsoft reports up to 50 percent better CPU performance over Cobalt 100, plus higher NVMe storage IOPS and network bandwidth through Azure Boost. Memory encryption is enabled by default through a custom memory controller, raising the baseline security posture with negligible performance impact. For agentic workloads, the design lets teams pack more agent sandboxes per VM while meeting latency and throughput targets.
Microsoft also announced general availability of Microsoft Discovery, a platform for building and governing agentic AI workflows across scientific and engineering disciplines, for organizations that have research or engineering simulation needs.
Why This Matters for Government
For state and local government IT leaders, three threads from Build 2026 are worth tracking closely.
First, agent governance is now a first-class platform concern, not an afterthought. Hosted agents run each session in an isolated sandbox, Toolboxes centralize auth and lifecycle, and Foundry IQ puts retrieval behind a single SLA-backed endpoint. Foundry IQ security updates, including permission sync and sensitivity-label governance, are in public preview. That centralization maps well to the audit, least-privilege, CJIS, FedRAMP, IRS 1075, and HIPAA expectations many public-sector agencies must plan around. When agents publish into Teams or Microsoft 365 Copilot, identity and policy flow through automatically, which can simplify the access reviews security teams already perform.
Second, secure-by-design infrastructure can reduce compensating-control burden. Cobalt 200’s always-on memory encryption is the kind of platform baseline that can help agencies document stronger default protections. Even if your agency runs primarily on x86 today, Arm-based VMs are increasingly viable for web tiers, caches, and data pipelines. For budget-constrained teams, the right evaluation is price, performance, and operational fit in your own workload rather than a generic cost claim.
Third, availability still needs tenant-by-tenant validation. Microsoft Foundry capabilities, GitHub Copilot sandboxes, and Cobalt 200 include a mix of GA, preview, expected-GA, and early-access preview statuses. Azure Government documentation also emphasizes that services and features available in global Azure might not be available, or might be configured differently, in Azure Government. Agencies should validate which Foundry, IQ, Copilot, and Microsoft 365 features are in scope for their environment before committing to a design. Treat the session catalog as architecture education and proof-of-concept fuel, and pair any pilot with a Responsible AI review and a sovereignty check against your specific cloud.
What to watch on demand
If your team has limited time, prioritize these topics from the Build session catalog: the hosted-agent runtime story, Foundry IQ and agentic retrieval, and the Toolboxes and tool-connection patterns. Developers building internal tooling should pair those with the Copilot SDK getting-started materials. The sessions are practical, and they give your architects the depth that an announcement post cannot.
The consistent message across Foundry, Copilot, and Azure infrastructure this year is that the platform is converging on production-grade agents with governance baked in. For government teams, that is the right direction, provided you verify availability in your cloud before you build.
Sources: Microsoft Foundry Build Edition recap (June 2, 2026), GitHub Changelog (June 2, 2026), Copilot SDK GA (June 2, 2026), Copilot CLI release (June 2, 2026), Azure Cobalt 200 announcement (June 2, 2026), Microsoft Discovery GA announcement (June 2, 2026), and Azure Government overview.