QUICK ANSWER
Microsoft Build Day 2 (June 3) shipped the engineering detail behind the Day 1 platform announcements. Key releases: GitHub Copilot agents for debugging, profiling, and testing in Visual Studio - not just code generation; multi-agent support in VS Code with the GitHub Copilot extension; the Surface Laptop Ultra (15-inch, NVIDIA RTX Spark SoC) unveiled at Computex; and Phi-4 Mini Reasoning on device confirmed for Copilot+ PCs. The GitHub, Copilot, VS Code Live session is available on-demand at build.microsoft.com.
GitHub Copilot in Visual Studio: Agents That Debug, Profile, and Test
The Build 2026 Day 2 session on Visual Studio (BRK207, presented by Mads Kristensen and Nik Karpinsky) was the demo-heavy follow-up to the Day 1 agent announcements. The central idea - articulated in Microsoft's pre-Build Visual Studio Blog - is that code is an asset, not just an artifact. The tooling around it should keep it healthy, correct, and easy to evolve. Every Day 2 announcement for Visual Studio was an expression of that idea.
Three specific GitHub Copilot agent capabilities were demonstrated in Visual Studio:
Debugging agent - Root-cause analysis
When a breakpoint is hit or an exception is thrown, the Copilot debugging agent can be invoked to perform root-cause analysis. It examines the call stack, variable state, and recent code changes, then proposes a fix with an explanation. The demo showed the agent correctly identifying a null reference exception caused by a race condition in an async initialization path - a class of bug that typically requires significant manual inspection of thread execution order.
Profiler agent - Performance suggestions from real data
The profiler agent reads the Visual Studio Performance Profiler's output - CPU usage data, hot paths, memory allocations - and generates concrete refactoring suggestions. Rather than generic "reduce allocations" advice, the agent proposes specific code changes targeting the actual hot paths in the developer's profile run. This connects existing profiler data to actionable code changes without requiring the developer to manually interpret what the profiler output means.
Test generation from live code paths
The test agent generates unit tests from actual executed code paths rather than static code analysis. By observing which paths are covered and which are not during a test run, the agent generates tests targeting the uncovered branches. The goal is tests that reflect real runtime behavior rather than the developer's assumptions about what paths exist.
VS Code: Multi-Agent Support and GitHub Integration
The GitHub, Copilot, VS Code, and More live session (LIVE104, June 3 9AM PT) was described by the Visual Studio Blog as "the closest thing Build has to a hallway conversation with the engineers shipping the work." It covered multi-agent support in the VS Code GitHub Copilot extension - allowing multiple specialized agents to work on different aspects of a task simultaneously within the editor - and deeper integration between VS Code, GitHub Actions, and Azure Pipelines.
The multi-agent VS Code integration connects directly to the Windows Agent Framework 1.0 announced on Day 1: agents built for the Windows Agent Framework can now be invoked from within VS Code via the Copilot Chat interface. This means a developer can trigger a Windows agent (for example, a deployment agent that pushes to a staging environment and runs smoke tests) directly from their editor conversation, without switching context to a separate management interface.
Surface Laptop Ultra - NVIDIA RTX Spark on a 15-Inch Windows Machine
Microsoft unveiled the Surface Laptop Ultra ahead of the Build keynote at Computex. The 15-inch device uses NVIDIA's RTX Spark system-on-chip - a new SoC that combines CPU, GPU, and NPU into a single package, positioned as NVIDIA's answer to Apple's M-series integration. The device was described by Engadget as "a 15-inch MacBook Pro lookalike" - a deliberate positioning against Apple's dominant developer laptop.
The RTX Spark chip is significant for AI developers specifically. Unlike the current generation of Copilot+ PC chips (Snapdragon X Elite, AMD Ryzen AI), which use dedicated NPUs for AI acceleration, RTX Spark integrates NVIDIA's CUDA ecosystem directly on the chip. This means the full NVIDIA AI software stack - CUDA, TensorRT, cuDNN - runs natively on the device NPU rather than requiring a discrete GPU. For ML engineers and AI developers who currently use a Mac for portability and a Windows desktop with an NVIDIA GPU for actual ML work, RTX Spark is designed to collapse those two devices into one.
Pricing and availability have not been announced. The Computex reveal means the device is expected to ship before end of 2026, but no specific date was given at Build.
Phi-4 Mini Reasoning - On-Device AI for Copilot+ PCs
Microsoft confirmed at Build that Phi-4 Mini Reasoning - a small language model optimized for on-device inference on Copilot+ PC NPUs - is coming to Windows. The model is designed to run entirely on-device at the NPU level, enabling reasoning-capable AI features that work offline and without sending data to cloud endpoints. This is different from larger frontier models that require cloud inference: Phi-4 Mini Reasoning is small enough to run on the 40+ TOPS NPUs in current Copilot+ hardware (Snapdragon X Elite, AMD Ryzen AI 300 series).
Combined with the Unified Windows AI SDK announced at Build (which bundles ONNX Runtime, DirectML, and the Copilot Runtime into a single NuGet package), Phi-4 Mini Reasoning represents Microsoft's push to make on-device AI practical for Windows app developers. Previously, enabling on-device AI in a Windows app required navigating fragmented toolchains. The unified SDK and Phi-4 Mini Reasoning on the Windows AI Foundry simplify that path significantly.
Frequently Asked Questions
Where can I watch the Build Day 2 sessions on-demand?
All Build 2026 sessions are available on-demand at build.microsoft.com after the conference. The key Day 2 sessions for developers are LIVE104 (GitHub, Copilot, VS Code Live - June 3, 9AM PT) and BRK207 (GitHub Copilot in Visual Studio: Agents That Debug, Profile, and Test - June 3, 4PM PT).
When will the Visual Studio Copilot debugging and profiling agents be available?
Microsoft has not announced a specific GA date. The agents were demonstrated at Build as preview capabilities available to Visual Studio Insiders. Check the Visual Studio Blog at devblogs.microsoft.com/visualstudio for the rollout timeline.
What is NVIDIA RTX Spark and how is it different from current Copilot+ chips?
RTX Spark is NVIDIA's new system-on-chip that integrates CPU, GPU, and NPU in a single package - similar in concept to Apple's M-series but with NVIDIA's CUDA ecosystem. Current Copilot+ chips (Snapdragon X Elite, AMD Ryzen AI) use ARM or x86 CPUs paired with dedicated NPUs that support ONNX and DirectML but not CUDA. RTX Spark adds full CUDA support to the integrated NPU, giving Windows developers access to the full NVIDIA AI software stack on a laptop.