Browse All Posts (106)

Stephen Toub breaks down how GitHub rewrote the Copilot agent runtime from a TypeScript/Node.js/V8 stack into a production Rust runtime, largely using Copilot agents. He explains the architectural drivers (in-process hosting, lower memory, better startup), the incremental port strategy, interop layers, regression patterns, and the measured performance gains.

Bruno Capuano introduces MafClaw, a four-part live series that builds a production-ready AI agent in C# using the Microsoft Agent Framework harness—starting from a minimal IChatClient wrapper and progressively adding tools, planning, safe file access, approvals, memory, skills, code execution, observability, governance, evaluations, and deployment to Microsoft Foundry.

Aashna Garg and Julia Kasper explain Project HydraFusion in GitHub Copilot, focusing on how orchestrating multiple models per request (instead of selecting a single model) can reduce frontier-model costs, and how the approach is evaluated with demos and benchmarking.

Allison announces a GitHub Enterprise Cloud update that lets enterprise admins automate SSO authorization for existing classic PATs and SSH keys across multiple SSO-protected organizations using an enterprise-installed GitHub App and a new bulk-authorization REST API.

Allison announces an additive SCIM change in GitHub: user responses now include the standard profileUrl attribute (RFC 7643), making it easier for identity providers and provisioning integrations to map SCIM users to the linked GitHub account without extra lookups.

Allison announces general availability of Copilot budget increase requests, letting members request additional AI credits when they hit their limit and enabling org/enterprise owners and billing managers to approve, adjust, or deny requests directly from GitHub settings under usage-based billing.

Tommaso Stocchi compares two multi-agent architectures in Microsoft Agent Framework: delegating work to specialist agents over A2A versus moving specialist instructions into distributed skills served over MCP, with concrete migration steps, code snippets, and measured differences in model calls, latency, and token usage.

GitHub hosts Cassidy and GPS discuss common AI “hot takes” aimed at developers, including whether reading code still matters, how hiring expectations are changing, and whether ideas like MCP and RAG are being misunderstood or prematurely declared “dead.”

fiveisprime previews a guided GitHub Copilot experience in VS Code that aims to take developers from an app idea to a deployed Azure application using a structured workflow with explicit checkpoints for scaffolding, local setup, and deployment.

Allison announces an update to GitHub code scanning: AI Scan for pull requests can now run even when CodeQL default setup isn’t enabled, expanding where AI-powered vulnerability detections can be applied for GitHub Advanced Security customers.

Wiliam_Rosa explains a public preview update to Azure Databricks Genie in Slack: when Genie answers include charts, Slack now shows the visualization as an image (not just text). The post also covers the required Slack app reinstall to grant files:write and a SQL query to audit Slack-originated Genie requests.

Waldek Mastykarz explains why AI coding agent eval pass rates can be misleading when the sandbox leaks information, and how to design evaluations around information boundaries rather than individual tool deny-lists.

The Visual Studio Code Team shares the 1.139 Insiders changelog, highlighting updates across the editor, terminal, extensions, and authentication APIs, plus ongoing Copilot-related work like agent-driven PR flows and improved model selection in multi-turn conversations.

Pieter Vandenheede breaks down four practical BizTalk migration paths—do nothing (short-term), coexistence, domain-by-domain migration, and big-bang replacement—and provides a decision matrix to help teams choose an approach that fits their constraints and risk tolerance.

SurbhiA walks through an image-based deployment approach for Azure Virtual Machine Scale Sets, showing how to build custom VM images, version them in Azure Compute Gallery, and roll out new versions via CI/CD with Azure DevOps pipelines, including rolling upgrades, rollback, and practical security considerations.

Wiliam_Rosa breaks down how Azure Databricks’ internal “AI SRE” agent speeds up incident investigation by gathering context automatically before engineers start debugging. The post focuses on an operational pattern: deterministic health checks and version-controlled runbooks first, then LLM-based synthesis—plus the guardrails needed to keep observability access safe.

Julia Gong, Ben Liggett, and Ulugbek Abdullaev explain how GitHub Copilot’s inline suggestions in VS Code moved from three separate models (completions, next edit suggestions, and long-distance edits) to a unified “3-in-1” approach, including the diff-patch output format, training strategy, and evaluation methods used to ship it.

Gaurav Bhardwaj proposes a Microsoft Foundry + Azure AI Search architecture that preserves meaning when turning job descriptions into recruiter-reviewable Boolean searches, with structured extraction, terminology provenance, an explicit approval step, and evaluation/observability to catch errors like “preferred” skills becoming mandatory.

Allison announces an update to GitHub Advanced Security that lets enterprise admins enforce security configurations across organizations, preventing org and repo admins from overriding enterprise-level settings and helping security and compliance teams apply consistent policies at scale.

Allison announces a public preview feature where GitHub Copilot suggests allowed values while admins define repository custom properties, helping organizations build consistent governance metadata that can be used to scope rulesets across large fleets of repositories.

Sharon Xu announces Foundry Dev Pack, an all-in-one installer that sets up a Microsoft Foundry development environment across the terminal, VS Code, and coding agents, including Azure CLI/azd tooling and starter workflows for creating hosted agents.

Michaela Isaacs introduces pq-adbc-advisor, a self-serve notebook-based scanner that inventories Microsoft Fabric workspace items still using legacy ODBC drivers and helps teams plan the ODBC-to-ADBC cutover. It produces a per-item impact report, classifies cutover risk, and recommends concrete fixes to validate ahead of connector-specific deadlines.

stevenbucher announces public preview support for versioning custom Azure Policy definitions and initiatives, including how to create new versions, pin assignments to major/minor versions, and validate which version is effective via REST API, portal, CLI, and PowerShell.

Nivas Narayanasamy lays out a practical architecture for making private Azure VMware Solution workloads reachable by Azure AI Foundry agents without moving data. The post shows how to use MCP “bridge” servers on VNet-integrated Azure Container Apps, with ExpressRoute, private endpoints, Entra managed identity, and Key Vault-backed secrets to keep access private and read-only.

Mayuri Gupta announces the public preview of Azure Red Hat OpenShift hosted control planes, explaining how the control plane can run as a Microsoft-managed service while worker nodes stay in the customer subscription. The post covers operational benefits, Azure integrations (identity, monitoring, IaC), security controls, supported regions, and preview pricing.

Allison announces that GitHub has completed the planned shutdown of SHA-1 support in HTTPS for github.com and partner CDNs, with notes on which GitHub Enterprise offerings are affected and where to find the earlier deprecation details.

Pamela_Fox shares the full set of MCP Live session recordings and resources, covering the Model Context Protocol’s latest spec updates, MCP server/client development in VS Code and GitHub Copilot, enterprise governance with Microsoft Foundry Toolboxes, and evolving MCP authentication patterns.

Aaron Powell shows how to baseline and then raise unit test coverage in the Interview Coach app using Visual Studio’s code coverage tooling and GitHub Copilot’s Test Agent, moving from targeted tests for a single class to solution-wide analysis and test generation.

Justin Yoo walks through a sample “Interview Coach” app that uses the GitHub Copilot SDK as an agent runtime (outside the editor) to run an interview workflow: ingest a resume, ask behavioral/technical questions, store session context, and generate end-of-interview feedback using MCP tools and Agent Framework handoffs.

GitHub shows how to hand off an active GitHub Copilot app session into VS Code so you can keep working in your editor without losing context, including the task plan, conversation history, diffs, and an isolated git worktree, then push changes back to a pull request.

Microsoft Developer shows how to use GitHub Copilot with a Microsoft Foundry skill to review and refactor an existing “Sparkle Cupcakes” agent, aligning it with current Foundry patterns and SDK guidance while improving production readiness.

Satoshi Kubota proposes using a Data Center WordCode Layer (DCWL)—a fixed-length 32-bit word encoding originally aimed at Azure text-processing workloads—to speed up Copilot for Windows and other Windows client text-heavy pipelines such as local RAG, semantic search, and log analysis.

Stephen Toub walks through a large set of runtime and library performance changes in .NET 11, with runnable micro-benchmarks and disassembly examples. It covers JIT optimizations (devirtualization, escape analysis, bounds-check removal), a new runtime-driven async implementation, SIMD/codegen improvements, and faster paths across core libraries like networking, JSON, regex, and collections.

Andrew Lock introduces Device Bound Session Credentials (DBSC), a browser/server protocol that makes stolen authentication cookies much harder to reuse on another machine. He explains the threat model (cookie theft and session hijacking), how DBSC uses TPM-backed keys and signed challenges, and what endpoints a server needs to add to support it.

Pieter Vandenheede explains how to decide whether to keep BizTalk after Microsoft confirmed BizTalk Server 2020 is the final release, and what conditions make “staying” a deliberate (and limited) choice. He also outlines practical signals that it’s time to move and frames the shift from “if we migrate” to “how we migrate safely.”

Thomas Maurer explains how Adaptive Apps and the open-source Radius control plane help teams engineer, test, and prove application portability across Azure, Azure Local, and other Kubernetes environments by separating app intent from environment-specific implementations.

GitHub shares three practical workflow changes to reduce review fatigue and keep pull request backlogs under control when teams adopt AI coding tools, focusing on better prompt inputs, smaller changes via stacked PRs, and shifting reviews toward assessment rather than line-by-line editing.

nzthiago explains the new GA support for site-scoped certificates and end-to-end TLS in Azure Functions Flex Consumption, including certificate sources (Key Vault, managed certs, uploads), how to expose certs to code on Linux workers, and how to implement inbound and outbound mutual TLS safely.

Wuyi Weng explains how Azure SRE Agent can use an Azure Managed Grafana dashboard (via the MCP endpoint) as executable context to investigate GitHub Copilot telemetry in Application Insights, turning panel queries, variables, and descriptions into fast, accurate triage and per-session diagnoses.

Gaurav Bhardwaj walks through a practical “refund agent” example to show why AI agent observability needs more than green HTTP checks, and how to use Microsoft Foundry plus Azure Monitor (Application Insights and Log Analytics) to trace runs, evaluate tool-call correctness, set alerts, and bake in security and privacy controls.