Browse Artificial Intelligence Blogs (44)
Hidde de Smet's Blog breaks down the difference between AGENTS.md (repo-wide, always-on instructions for coding agents) and .agent.md (custom agent profiles for GitHub Copilot), including where to place each file, what fields matter, and how to use roles, tool restrictions, and handoffs safely.
John Edward explains why event-driven architecture is a strong fit for agentic AI systems, and breaks down the core patterns (pub/sub, event sourcing, sagas) plus practical concerns like ordering, observability, and infrastructure overhead.
Hidde de Smet's Blog explains how GitHub Copilot “skills” work via SKILL.md folders, why the YAML description is the key to discovery, and how this approach keeps context lightweight compared to a giant copilot-instructions.md. It includes a practical Azure Monitor/Application Insights KQL skill you can copy into a repo.
DevClass.com reports on Visual Studio 18.5 (Visual Studio 2026), covering new Copilot-driven “agentic” debugging, changes to how IntelliSense/Copilot suggestions are prioritized, and ongoing developer complaints about theme contrast and forced auto-updates.
Hidde de Smet compares three AI coding setups—single-agent, agent-with-tools, and multi-agent—using a realistic .NET Aspire + ASP.NET Core rate-limiting task to show trade-offs in fit, cost, latency, and common failure modes.
John Edward explains when to use single-agent vs multi-agent AI architectures in a Microsoft context, mapping common designs to Semantic Kernel, AutoGen, and Azure services like Azure OpenAI, Azure AI Search, Functions, Service Bus, and AKS.
DevClass.com reports on GitHub’s private preview of Stacked PRs, a workflow for breaking large changes into smaller, independently reviewable pull requests that can still depend on each other, with an optional gh stack CLI that’s also intended to work well with AI agents.
Jesse Houwing summarizes GitHub’s update that GitHub Copilot can now keep inference processing and associated data within US or EU data residency regions, and shows the enterprise/org policy you must enable to restrict Copilot to data-resident models.
Rob Bos walks through running GitHub Copilot CLI against local OpenAI-compatible inference servers (Ollama, LM Studio, Foundry Local, vLLM/TGI), focusing on the practical constraints (32k context, tool calling, VRAM/KV-cache) and sharing concrete Windows/PowerShell setup and throughput numbers.
Emanuele Bartolesi shows how to point GitHub Copilot CLI at an Azure AI Foundry (Azure OpenAI) deployment using a BYOK-style setup, including how to deploy a model, build the correct endpoint URL, set the required environment variables, and validate the connection.
Emanuele Bartolesi explains how to run GitHub Copilot CLI against a local LLM via LM Studio’s OpenAI-compatible API, including the exact PowerShell environment variables needed to avoid cloud fallback and when this offline setup is (and isn’t) worth using.
Hidde de Smet explains how Spec-Kit’s extension system works, highlights useful community extensions, and walks through the Ralph Loop extension, which runs a GitHub Copilot agent in iterations to implement tasks from `tasks.md`, commit changes, and track context in `progress.md`.
Harald Binkle explains the latest Visuals MCP update, adding a chart tool that lets AI agents render single charts and full dashboards directly inside GitHub Copilot Chat in VS Code, with Storybook examples and export options for turning analysis into shareable visuals.
Thomas Maurer introduces Azure Local Disconnected Operations and explains how to run Azure-style infrastructure—and selected AI workloads—inside fully disconnected or air-gapped environments for sovereignty and compliance needs.
John Edward outlines a practical security checklist for running Microsoft AI agents in production, covering Entra ID identity controls, least-privilege access, data boundaries and DLP, audit logging with Azure Monitor/SIEM, and concrete defenses against prompt injection and unsafe agent behavior.
Randy Pagels explains a simple GitHub Copilot workflow: before asking for an implementation, prompt Copilot to ask clarifying questions so you uncover assumptions, edge cases, and missing requirements early—leading to better prompts and better code changes.
Jesse Houwing clarifies GitHub Copilot’s April 24 interaction-data policy change, explaining which subscription tiers may have interactions used for training, what is and isn’t included (like private repos), and practical ways enterprises can enforce license tiers and lock down developer environments.
Emanuele Bartolesi explains how to make GitHub Copilot less “agreeable” and more useful by adding a repo-level voice instructions file that pushes Copilot to be direct, critical, and focused on correctness and maintainability.
Zure summarizes recent Microsoft Fabric and Purview capabilities for metadata management and governance, covering OneLake catalog search, workspace tagging, bulk definition APIs, and how AI agents/copilots intersect with lineage, compliance, and risk controls.
John Edward shares practical ways to control Azure-based copilot and AI agent spend, focusing on token discipline, caching, model selection, and ongoing governance so LLM solutions scale without surprise bills.