Microsoft Build 2026: What Developers Need to Know
June 3, 2026
Microsoft held Build 2026 on June 2–3 in San Francisco. The theme was agents. Every announcement, every demo, every product name came back to the same idea: Microsoft is giving agents a whole computer to live in.
Here's what actually matters if you build software.
Project Polaris: Copilot Is Cutting the OpenAI Cord
This was the biggest announcement, and it got buried under the product-name avalanche.
Microsoft unveiled Project Polaris — its own in-house AI coding model — as the future reasoning engine for GitHub Copilot. Polaris will replace GPT-4 Turbo as the default model for Copilot subscribers starting August 2026, with automatic migration and an optional three-month fallback period for teams that want to stay on GPT-4.
Polaris is a mixture-of-experts architecture with specialized sub-modules tuned for different programming languages and frameworks. According to Microsoft, it outperforms GPT-4 Turbo on HumanEval and MBPP benchmarks — with particular gains in low-resource languages like Rust and Haskell. Pro tier subscribers get multi-file context up to 100,000 lines and autonomous test generation.
The model runs on Microsoft's own Maia AI accelerators inside Azure — so Microsoft now controls the model, the inference stack, and the developer experience end to end. The OpenAI dependency in Copilot is done.
GitHub Copilot Gets a Standalone Desktop App
Microsoft announced a standalone GitHub Copilot desktop app — described as an agent-native experience where multiple agents can work on different projects simultaneously without stepping on each other.
It's the home base Copilot users have been missing. The app ships alongside a new Intelligent Terminal — an open-source experimental fork of Windows Terminal with native agent integration, an agent status bar, automatic error detection, and support for Agent Client Protocol-compatible agents. GitHub Copilot CLI is the default, but developers can plug in other compatible agents.
This matters if you've been bouncing between VS Code, the GitHub web interface, and the terminal to get things done. One app, multiple concurrent agent sessions.
I already covered the Copilot billing overhaul that went live on June 1 — a desktop app that burns AI Credits faster might be something to budget for before you install it.
Copilot Workspace Is Now Generally Available
Copilot Workspace exited beta and went generally available at Build. The GA release ships with Copilot Extensions for Jira, Datadog, and ServiceNow; Fleet mode, where Copilot CLI operates autonomously on narrowly scoped codebase tasks without per-step confirmation; and Autopilot mode, where Copilot acts on a bounded issue without a developer present.
Fleet and Autopilot are the meaningful upgrades here. This is where Copilot crosses from "AI-assisted" to "AI-delegated" — you define the task, Copilot executes it. Dependency updates, test gap coverage, documentation fixes can now run unattended if you set it up properly.
Windows Agent Framework Is Now MIT-Licensed
Windows Agent Framework v1.0 shipped April 2 and was MIT-licensed at Build. It lets developers define agents in YAML that run across local Windows machines, Windows 365 Cloud PCs, and Azure Arc-enabled edge devices — all using the same manifest. An agent can start as a local process on a developer's laptop, escalate to a Windows 365 GPU node when the task requires it, and publish to Azure as a service with no re-architecture.
The MIT license is the important part. Unlike Azure-specific SDKs, WAF can be forked and deployed outside Microsoft's cloud entirely. If you want on-premises Windows agents without Azure, this is now a viable starting point.
Surface RTX Spark Dev Box: The Local AI Hardware Pitch
Microsoft announced the Surface RTX Spark Dev Box — a compact desktop machine built on Nvidia's RTX Spark superchip, announced at Computex just a day earlier.
The Dev Box runs up to 20 ARM CPU cores, 6,144 Blackwell CUDA cores, and 128GB of unified RAM. It's aimed squarely at developers who want to run local AI models, handle software development, or manage data-intensive workloads — without touching a cloud meter.
RTX Spark is powerful enough to run 120-billion-parameter LLMs entirely locally with up to a 1 million token context window. The same CUDA stack as datacenter GPUs — no recompilation for PyTorch, llama.cpp, or TensorRT code. No pricing announced yet.
Jensen Huang literally said "no meter anxiety" onstage in Taipei. Given what GitHub Copilot just did to its billing model, that phrase is going to land differently for a lot of developers.
Azure AI Foundry: Multimodal and Expanded Models
Azure AI Foundry received native multimodal support — text, image, video, and audio in a single pipeline — a visual RAG designer for fine-tuning pipelines, and an expanded model catalog adding Cohere, Mistral, and Stability AI alongside existing OpenAI and Meta models. Per-project token budgets and consumption monitoring were also added.
If you're building multimodal agents and have been avoiding Foundry because of pipeline complexity, that excuse is gone.
The One-Line Summary
Microsoft is trying to move the industry from the app era into the agent era — and it's using Windows, GitHub, Azure, and its own hardware to make that case.
Project Polaris is the headline. Copilot Workspace GA and Fleet mode are the things to actually try this week. WAF's MIT license is the sleeper detail that enterprise teams with on-premises Windows infra should look at closely.
Sources: Microsoft Build 2026, ChatForest Build Recap, The Neuron