Browse All Posts (139)

sbaynes introduces SkillOpt, a method for improving AI agent reliability by “training” a natural-language skill file through controlled text edits, validation gating, and feedback from execution trajectories—without updating model weights.

Matthew Leibowitz, David Ortinau, and Gerald Versluis walk through what’s new in SkiaSharp 4 and what the update means for .NET MAUI developers, then share highlights from the .NET MAUI community including MAUI UI July, recent releases, and notable ecosystem contributions.

Allison announces general availability of browser tools for GitHub Copilot in VS Code, enabling Copilot agents to drive a real browser to navigate and test live web apps, capture page content and console errors, and run scripted flows, with privacy defaults and enterprise controls.

Arvind Shyamsundar explains how to keep costs down without sacrificing performance by using Azure SQL features that help applications scale efficiently as they grow.

Natalie Guevara outlines six quick GitHub repository settings that materially reduce common attack paths for maintainers, focusing on vulnerability intake, secret leakage prevention, dependency risk, static analysis, and safer merges to the default branch.

John Maeda and Ross Heise wrap up the Cozy AI Kitchen series with a reflective finale on making AI concepts more approachable for developers, revisiting recurring ideas like Semantic Kernel, embeddings, tokens, and “agents,” and sharing what they learned from blending education, storytelling, and hands-on metaphors.

Samantha Kubota reports on the Theodore Roosevelt Presidential Library’s “living library” experience, where AI is used to organize and reconstruct archival documents, enable natural-language search over Roosevelt’s writings, and power an interactive Roosevelt avatar with safety controls for public use.

Allison announces an update to Copilot CLI where Auto model selection routes each request to an appropriate model based on task needs and real-time reliability signals, with billing tied to the selected model’s multiplier and a discount for paid subscribers using Auto.

GitHub highlights Arnis, an open source project that turns real-world map and elevation data into detailed Minecraft worlds, letting you generate accurate terrain and buildings and tweak settings like scale and spawn points.

Waldek Mastykarz explains why public AI coding benchmarks like SWE-bench often fail to predict how a model will perform in your real developer environment, especially when proprietary code, instruction files, and a complex extension stack are involved. He outlines what benchmarks are good for and how to run small, repeatable evals that match your actual workflows.

Carlotta Castelluccio demonstrates how to build a low-code AI agent in VS Code using the Foundry Toolkit’s Agent Builder, connect it to the Microsoft Learn MCP Server, configure instructions and tools, and then validate the agent’s behavior using GitHub Copilot evaluation.

xujx summarizes recent Microsoft Fabric Eventstream connector updates for Real-Time Intelligence, including GA private network support, GA Kafka and Azure Service Bus connectors, and GA custom CA/mTLS. The post also covers preview features like workspace identity for Event Hubs, richer IoT Hub metadata, Oracle CDC ingestion, and HTTP pagination.

John Edward outlines an end-to-end approach for building a ServiceNow-focused AI assistant in Microsoft Copilot Studio, aimed at ticket auditing, SLA risk monitoring, and engineer performance reporting. The guide covers a layered architecture, ServiceNow REST API integration via Power Platform custom connectors, and practical security considerations for enterprise deployment.

The Visual Studio Code Team highlights early changes in VS Code 1.128 (Insiders), including enterprise policy support for OpenTelemetry settings and updates to AI model configuration and credentials when using Copilot-related agent features.

j_folberth shows how to add fast smoke tests to a GitHub Actions deployment pipeline for Azure AI Foundry hosted agents, using a JSON prompt catalog and a Python runner to validate basic agent behavior (reachability, prompt alignment, threading, refusals, and hallucination resistance) right after deployment.

Lee Stott walks through a full “Multi‑Agent Dev Canvas” scenario for GitHub Copilot Canvas, showing how to decompose work, execute agent flows, validate with in-surface tests, inject failures, and evolve the design live (including GDPR/PII redaction) until the system meets its acceptance criteria.

Kent Weare demonstrates a Microsoft Dataverse plugin that gives coding agents guardrailed, schema-aware access to Dataverse so they can provision solutions, import data, and validate results from natural-language requests without producing brittle, hallucinated implementations.

GitHub announces that GitHub Copilot Agent can now be added natively in JetBrains IDEs through JetBrains AI Assistant, using agent client protocol support. The video highlights a quick enablement flow from the AI agents tab and examples like summarizing projects, reviewing README files, and running tests with code coverage checks.

hetvip shares measured performance results for Azure Container Apps Express, comparing it to a standard Consumption environment for cold starts and first-time provisioning. The post explains how MicroVM-based startup and disk/memory state restore reduce wake-up time, and links to docs for getting started in preview.

DevClass reports on renewed community criticism of Microsoft’s .NET support lifecycle, arguing that the three-year LTS window is too short for enterprise upgrade and adoption cycles and can leave only a one-year window to move between LTS releases.

Mark Russinovich explains why Microsoft is moving its post-quantum cryptography work forward, and what “quantum-safe readiness” looks like in practice across network protocols, stored data, and trust chains like signing and certificates.

YuliaTurchin announces the general availability of Item Recovery in Microsoft Fabric, covering how soft-deleted items can be restored via the Workspace Recycle bin or REST API, and how admins can set tenant-wide retention windows (up to 90 days) and update operational runbooks and governance workflows.

sbaynes (Microsoft Incident Response) explains how agentic AI systems can be compromised when Model Context Protocol (MCP) tool metadata is poisoned, causing an agent to take unintended actions. The post maps a realistic finance workflow attack chain and gives concrete detection and mitigation steps using Microsoft security controls.

Natalie Guevara explains how GitHub’s Open Source Program Office uses GitHub License Compliance (part of GitHub Advanced Security) to review dependency licenses in pull requests, manage exceptions, and enforce organization-wide policies without blocking teams during rollout.

Allison announces general availability of Anthropic’s Claude Sonnet 5 in GitHub Copilot, including where it can be selected (IDE, CLI, and GitHub surfaces), how it’s billed under usage-based pricing, and how Business/Enterprise admins can enable it via model policy settings with Zero Data Retention.

jiang_jenny1 introduces a preview, AI-assisted command-line approach for migrating Azure Synapse Spark artifacts and Synapse pipelines into Microsoft Fabric. It covers the end-to-end phases, two migration strategies (lift-and-shift vs. migrate-and-modernize), automatic refactoring and dependency handling, and how the companion pipeline skill rewires orchestration to Fabric Data Factory.

Divakar Kumar introduces Archaios, an AI-powered exploration platform for archaeologists that processes large LiDAR datasets and related geospatial signals. He shows how .NET and Azure Durable Functions can orchestrate event-driven pipelines, and how Semantic Kernel-based multi-agent workflows can simulate expert collaboration to help identify potential historical sites.

Jan Krivanek and Yuliia Kovalova show how to run MCP-powered MSBuild binlog diagnostics inside GitHub Actions so failed PR builds get an automated root-cause comment (with inline suggestions) without anyone downloading logs. They also catalog new Binlog MCP tools and share evaluation data on quality, time, and token usage.

Allison announces a public preview feature in GitHub Code Quality that lets teams enforce code coverage thresholds via branch rulesets, blocking pull request merges when coverage drops below a minimum or falls too far from the default branch.

Allison announces updates to GitHub Releases that make long release pages easier to navigate via a sidebar table of contents, and adds per-asset download counts directly in the Releases UI for users with write access.

Anthony Nocentino and Anna Hoffman show how to give GitHub Copilot controlled, real-time visibility into SQL Server using MCP (Model Context Protocol) servers. They demo a custom DMV-based MCP server for DBA workflows and a zero-code Data API Builder approach for app data access, with guardrails and permissions defined by you.

Mark Downie covers the June Visual Studio update, focusing on GitHub Copilot usage tracking (token-based metering with alerts), new trust validation for MCP servers, and the general availability of the Copilot modernization agent for C++ MSVC upgrades.

Allison announces a deeper integration between JetBrains AI Assistant and GitHub Copilot, making Copilot a first-class selectable agent inside JetBrains’ agent picker. The update highlights model selection and reasoning-depth controls, plus agentic workflows where Copilot can propose changes, run commands, and iterate on multi-step coding tasks.

Allison announces a public preview feature in GitHub Enterprise Cloud that enforces enterprise-wide open source license policies on dependency changes, so teams can block noncompliant licenses before code is merged and reaches production.

Alym Rayani rounds up June 2026 updates across Microsoft Security, including an agentic vulnerability scanning system (codename MDASH), new Microsoft Defender protections for local AI agents and MCP servers, Entra Backup and Recovery GA, and expanded Defender for Cloud and Purview capabilities for multicloud and data security.

Allison announces a Dependabot change for npm private registries: instead of trying to infer .npmrc from lockfiles (which often broke auth), you can now set a registry scope in dependabot.yml so Dependabot generates the correct .npmrc reliably.

Allison announces a new GitHub cloud data retention policy for closed Dependabot security alerts, including what stays available in the UI/API, what moves to archival storage after two years, and what admins need to change in any REST API-based reporting before August 25, 2026.

Allison announces upcoming GitHub access restrictions for several public REST API endpoints and related UI views that expose stargazer and watcher lists, aimed at reducing data scraping used for spam and improving platform security.

Andrew Lock explains the new C# “closed class hierarchies” feature in .NET 11 preview 5, showing how it restricts inheritance to a single assembly and why that matters for safer switch expressions. He also covers how to enable the preview feature and what limitations to expect.

Allison announces a new GitHub Enterprise Cloud billing control for GitHub Copilot: per-user AI credit budgets scoped to cost centers, so admins can set consistent per-user limits that automatically follow users as they join or leave teams.