Browse All Artificial Intelligence Content (812)

Waldek Mastykarz introduces the Agent Experience (AX) Practitioner Playbook, a practical method for evaluating how AI coding agents use your SDKs, docs, tools, and extensions—and for tracing failures back to their root causes so you can ship fixes to the sources agents rely on.

Fokko at Work demos what’s new in Visual Studio Code 1.141, focusing on workflow and UI updates plus several GitHub Copilot improvements, including continuing external Copilot sessions, sandboxing in the Copilot agent host, persistent chat progress, and visibility into discovered MCP servers.

Patrick Nikoletich explains how GitHub Copilot agent sessions can move between devices using GitHub’s shared agent runtime, the Copilot SDK, and the Agent Host Protocol, including a live demo that hosts an agent on a home Linux machine and connects to it from the Copilot CLI.

Gareth Bland explains why AI agent costs can scale much faster than the work being done, and how context caching helps control token spend as agents take more turns.

Tsuyoshi Ushio introduces Dynamic Workflows for Azure Functions Hosted Skills, a preview feature that lets an LLM generate a typed DAG plan once and then run it durably via Durable Functions. The post explains the runtime model, safety checks, and when this approach reduces token usage and latency for event-driven apps.

Microsoft Developer explains how to debug an AI agent that’s slow or confidently wrong by turning on server-side tracing for an Azure AI Foundry project and inspecting each run step-by-step in Application Insights, including model calls, tool calls, tokens, and latency, then querying patterns with KQL.

sbaynes explains how GitHub Copilot is rolling out local, on-device AI coding on new Windows PCs, including automatic routing between local and cloud models, a new MAI Code 1.1 Flash local model build, and sandboxed tool execution using Microsoft Execution Containers (MXC).

DivSwa announces public preview pricing and monthly free grants for Azure Logic Apps Automation, including what’s covered (managed environment, workflow runtime, retention, connectors, AI usage, knowledge storage, and isolated code execution) and how to track and keep eligibility for the free environment.

Allison announces that Claude Haiku 5.5 is now generally available in GitHub Copilot, positioned as a lightweight model for fast, high-volume coding tasks like subagents, quick edits, and terminal workflows, with access controlled via Copilot plan entitlements and enterprise model policies.

Chu Lahlou compares Azure Document Intelligence (ADI) and Azure Content Understanding (ACU) for real-world document processing, focusing on when to use purpose-trained document models versus generative, schema-driven extraction. The guide covers OCR/layout overlap, custom extraction trade-offs, RAG-ready output, multimodal analysis, and a practical evaluation checklist.

GitHub explains an upcoming GitHub Copilot feature that intelligently routes some requests to local, on-device models (including Ollama and Microsoft Foundry Local) to reduce cloud usage, enable offline workflows, and keep sensitive data on the machine.

Erin Havens shares GitHub data showing secret leaks are scaling with the volume of code being pushed, then explains how GitHub Secret Protection is using a fast ModernBERT-based classifier to improve push protection for unstructured secrets and reduce the human cost of remediation.

bobmital announces the general availability of Anyscale on Azure, a managed Ray runtime that runs on Azure Kubernetes Service in the customer’s subscription. The post explains the control-plane/data-plane split, Azure-native identity and governance integration, private connectivity options, and the kinds of AI/ML workloads teams can run end-to-end.

Allison announces a new fine-tuned GitHub model for context-aware leaked secret detection across secret scanning alerts, push protection, and GitHub Copilot /security-review. The post explains what’s changing, where it’s available (including GHES 3.23), and how AI Credits and budgets apply to the new opt-in checks.

Taesoo Kim shares what Microsoft Security’s FORGE Lab learned while scaling AI-assisted vulnerability research across Windows and major open-source projects, including how to turn agent-discovered bugs into reproducible evidence, validated fixes, and shipped security updates.

Allison announces general availability of local sandboxing for GitHub Copilot, adding a secure execution boundary for agent-run tools and commands on developer machines across Copilot CLI, the Copilot app, and VS Code Agent Host sessions.

Allison announces an update to GitHub Copilot CLI that lets you discover and select supported local models from a running Ollama instance using the new /model picker, without restarting your session. The post also clarifies how adding local models relates to offline mode, telemetry, and remote providers.

Apoorv Gupta explains how GitHub Spec Kit can be extended for enterprise use with presets, extensions, bundles, workflows, and governed catalogs, so teams can standardize spec-driven development without forking the core toolkit.

Rani Borkar explains how Microsoft applies agentic AI across Azure’s hardware and infrastructure lifecycle—from demand planning and supply chain through datacenter fleet operations—to shorten decision cycles, improve reliability, and build a closed-loop learning system that keeps human judgment in control.

Visual Studio Code introduces “Blobby” as the new VS Code Pet and shows how to enable it from GitHub Copilot Chat using the /vscode-pet command.

Gareth Bland explains TokenOps: a practical way to control AI agent spend by focusing token budget where it creates real value. He breaks down why multi-turn agents get expensive fast, how cached input changes the cost curve, and which efficiency ratios help teams measure whether they’re getting leverage or just burning tokens.

John Savill gives a quick, practical overview of MCP (Model Context Protocol), explaining why it matters for integrating AI with external tools and services. He frames MCP as a standard way for models to discover capabilities and interact with “target services” through local or remote MCP servers.

Allison explains why GitHub Copilot usage metrics may undercount agent activity after some IDEs moved agent sessions to the Copilot SDK, and what to do to restore accurate reporting. The post lists the fixed IDE versions, what report gaps to expect, and why missing data can’t be backfilled.

sbaynes outlines how “physical AI” shifts factory automation from pre-programmed instructions to systems that perceive, reason, and act in the real world. The article focuses on agent-assisted operations, scaling patterns across plants and fleets, and the governance and security controls needed when AI decisions have physical consequences.

Rory Preddy demonstrates an end-to-end workflow in VS Code using GitHub Copilot to build and debug a Java Spring Boot app, then expose app capabilities as MCP tools and validate behavior with Playwright tests.

Microsoft Developer explains why an AI agent’s output quality depends on the context you can retrieve at runtime, and shows how Foundry IQ-style capabilities connect agents to enterprise data, ground responses, and enable actions across business workflows.

Mahmoud A. Atallah shows how he uses GitHub Copilot agents in VS Code to turn meeting transcripts and other customer inputs into structured requirements and execution-ready delivery artifacts, then enriches them with Azure guidance via MCP servers and the Well-Architected Framework.

Alex Weininger introduces the Canvas authoring plugin, which guides GitHub Copilot through the native create-canvas workflow to generate an Azure canvas app based on Microsoft Canvas Toolkit, including a read-only starter and patterns for safe, explicitly scoped Azure operations.

Freddy Dezeure and Sesha Mani explain how frontier AI models are changing vulnerability management, and what CISOs can do to keep patching both fast and correct. The post highlights Microsoft’s use of AI “harnesses” for scanning and triage, and points to Baseline Security Mode and secure-by-default controls to reduce blast radius when patches can’t land in time.

Ayush Shekhar introduces ScreenMind, an open source project that runs a single local AI model to understand what’s on your screen, transcribe audio (like meetings), and answer questions about what it has seen—aimed at running on a GPU with as little as 4GB of VRAM.

Emanuele Bartolesi shares a practical fix for high CPU/disk usage on Windows 11 when running Claude, Codex, and GitHub Copilot tooling: adding targeted Microsoft Defender Antivirus folder exclusions. The post also calls out the security tradeoff and walks through the exact Windows Security steps to apply exclusions safely.

Victoria Hall introduces Azure Functions Agent bindings (preview) for Python, showing how to add bounded agentic reasoning via Microsoft Agent Framework while keeping deterministic function code in control. The post walks through project structure, package setup, local settings, and examples for both direct HTTP calls and Durable Functions orchestrations.

Waldek Mastykarz introduces Agent Experience (AX): how AI agents discover, choose, and correctly use your SDKs, APIs, and CLIs. He breaks AX down into propensity and efficacy, shows why you must measure both quality and task cost, and shares practical examples from GitHub Copilot Chat evaluations.

Samu Niemelä shares practical takeaways from FabCon 2026 on how Microsoft Fabric is maturing, from Fabric IQ grounding Copilot and agents in governed business semantics to new on-demand compute and stronger monitoring and governance features for running Fabric at scale.

Rob Bos shares a short pointer to his Azure Heroes session on how he works with GitHub Copilot, with a link to download the slides.

Lily Ma introduces Azure Functions managed connectors (public preview) and shows how connector triggers and typed clients let functions react to events and call actions across services like SharePoint and Teams without hand-rolling webhook registration or OAuth token handling.

Michelle Zhou introduces ReviewBench, an open benchmark for evaluating AI code review agents using a representative set of GitHub pull requests, a multi-source ground truth process, and calibrated scoring. The post explains how to interpret grounded vs augmented metrics and how teams can submit their own agents to the public leaderboard.

This week's AI roundup tracks a clear shift from chat-based assistance to action-taking systems, with GitHub Copilot adding dynamic workflows, desktop "computer use", multi-model orchestration (HydraFusion), and more automation across code review and PR merge. In parallel, Microsoft and partners focused on the controls teams need when agents can execute real changes, including identity and policy across Entra tenants, runtime enforcement via gateways, and isolation with sandboxes and microVMs. On the data side, Fabric IQ and OneLake updates reinforced that agent-ready context depends on governed semantics and permissions, while Azure SQL Hyperscale pushed vector search deeper into mainstream databases. We close with practical guidance on measuring cost and quality (routing eval toolkits, FinOps controls) and the security work needed to keep agent-driven development reviewable and accountable.

John Savill recaps major Microsoft AI announcements from September 2026, focusing on Azure AI Foundry updates (models, routing, agent features, and controls), Copilot platform changes, and developer-facing items like GitHub Copilot and VS Code agent workflows, plus data/AI additions such as Azure SQL vector capabilities.

GitHub introduces gh-secure, a one-command tool from the GitHub Security Lab aimed at improving basic repository security hygiene by helping prevent secrets from being leaked into public code, with usage via the Copilot app, Copilot CLI, or GitHub CLI.