Browse All Posts (131)

Christine Caggiano, Joshua Bradley, Steven Truitt, and William Chappell describe how Microsoft Discovery’s Discovery Engine adds CLIO (a self-adaptive “cognitive loop”) to improve agentic scientific problem-solving, and share benchmark results from Agent’s Last Exam showing higher scores and better consistency through multi-model, evidence-driven exploration.

Scirqueira shares practical operating rules for keeping a Microsoft Fabric Data Warehouse medallion architecture healthy in production, focusing on batch ingestion, rerunnable Silver transformations, consumption-ready Gold modeling, using Fabric DW defaults effectively, and monitoring performance and failures by layer.

Allison announces generally available updates to npm trusted publishing, including multiple OIDC configurations per package, staged approvals that wait for malware scanning to finish, and improved version history visibility for maintainers on npmjs.com.

Allison summarizes three GitHub Actions improvements: a new REST API to track runner version deprecation timelines, a least-privilege `vulnerability-alerts` permission for `GITHUB_TOKEN` to read Dependabot alerts, and new `job` context properties that help reusable workflows identify their source at runtime.

Fokko at Work recaps what’s new across Visual Studio Code 1.133–1.136 with a focus on GitHub Copilot features, showing practical demos like Agent Plugins support, prompt UX improvements, usage details per chat turn, and agent merge workflows.

Andre Terceros and Bodhisatva Gautam announce a preview of Workspace Outbound Access Protection (OAP) in Microsoft Fabric for Operations Agent and Fabric Maps, explaining how workspace admins can enforce a default-deny model for outbound connections, allow only approved endpoints, and get clearer visibility into blocked actions.

ryanjadams explains how Fabric’s “connection recency” signals help admins govern connections at enterprise scale, and how to use Fabric REST APIs plus a companion Python notebook to find stale, duplicate, and single-owner connections before taking cleanup actions.

Allison announces the deprecation of several AI models used across GitHub Copilot experiences on October 2, 2026, and outlines the replacement models plus the admin steps needed to enable them via Copilot model policies.

Allison announces a required PGP signing key update for the GitHub CLI (gh) Linux APT and RPM repositories ahead of the September 5, 2026 key expiration, with guidance on who needs to update their setup and which installation methods are unaffected.

Allison announces that Gemini 3.8 Flash is now available as a selectable model in GitHub Copilot, including where it can be used (IDEs, CLI, and cloud agent), who gets access across Copilot plans, and how admins can control enablement via model policy.

wolfgangdesalvador introduces a public preview optimization for Azure Blob Storage List Blobs that returns results in Apache Arrow format, cutting response size and client parsing cost. The post explains the required REST headers and SDK versions, shows a Python decoding example, and shares rclone benchmarks demonstrating large speedups with parallel listing.

Amaury Levé explains how MSTest 4.4 uses source generation to make test projects publishable and runnable as Native AOT executables, so your tests exercise the same trimming and reflection constraints as the app you deploy.

pranabpaul walks through the first part of building a retail assistant demo using Microsoft Fabric and Azure AI Foundry, focusing on setting up the Azure environment, provisioning Fabric capacity, loading sample data into a Lakehouse, creating an ontology, and wiring a Fabric Data Agent to query that data.

pranabpaul continues the 3IQ Retail Assistant demo by wiring unstructured PDF content into an Azure AI Foundry agent using Azure Blob Storage and Azure AI Search, including the RBAC permissions needed for managed identities and the networking trade-offs involved when publishing the agent to Microsoft 365 Copilot.

Stefan Kingham’s documentary traces how Visual Studio Code evolved from early web-based editor experiments into a widely used, open-source developer tool, covering key technical and product shifts like Electron, extensions, LSP, remote development, and the editor’s adaptation to the AI era including GitHub Copilot.

GitHub explains the “AI harness” concept for guiding LLM execution, and how harness engineering adds the runtime, tools, and guardrails that let AI agents work safely against a codebase in agentic workflows.

sbaynes introduces MAI-Transcribe-2, a speech-to-text model positioned for low-latency, multilingual transcription with features like diarization, word-level timestamps, keyword biasing, and configurable output styles, alongside benchmark results and launch pricing aimed at high-throughput transcription workloads.

ryanjadams introduces Microsoft Fabric Connection Recency (preview) and the related REST APIs for auditing and governing connections at enterprise scale, including how to identify stale and duplicate connections and reduce ownership risk using role assignments and Entra group ownership.

pankajarora announces a preview of Capacity Operation Events in Microsoft Fabric Real-Time Hub, adding near real-time, operation-level telemetry to help capacity admins pinpoint which operations, workspaces, and artifacts drive consumption, throttling, and performance issues, and to build custom monitoring and alerting workflows.

Mohammed Barqawi introduces Logic App Storage Inspector, a read-only Kudu site extension that helps troubleshoot Azure Logic Apps Standard by inspecting the underlying storage for workflow history, versions, and queue health, with guidance on installation, configuration settings, and managed identity permissions.

Bob Ward and Anna Hoffman discuss five lesser-known aspects of Azure SQL Database Hyperscale, focusing on what the “Hyperscale” tier really means, how its architecture differs from other tiers, and the practical implications for replicas, backups, and pricing.

Kayla Cinnamon explains how the GitHub Copilot app can run multiple agent sessions in parallel on the same project by isolating each session with its own Git worktree, so you can queue up tasks (like feature work, accessibility review, and tests) and review results as they complete.

Microsoft Security Research (Noam Kochavi and Sarah Wolstencroft) explains how “ASCII smuggling” with invisible Unicode tag characters moved from AI prompt-injection research into real-world phishing evasion, and shares practical detection and mitigation guidance for email defense pipelines.

Allison announces that GitHub is gradually reopening sign-ups for Copilot Business and Copilot Enterprise paid by credit card or PayPal, alongside billing and account-vetting changes that affect how seats are activated, charged, and how overage usage may require additional payment.

Dalibor Kovacevic explains how to use Azure SRE Agent’s advanced ServiceNow filtering to route the right incidents to the right response plan, then control how much autonomy the responder has (Autonomous vs Review) while you preview and tune matching before enabling it.

Allison summarizes what’s new in CodeQL 2.26.4 for GitHub code scanning, including language support updates and query accuracy improvements that affect security detections across C#, Java/Kotlin, and GitHub Actions workflows.

dotnet covers the production-readiness pieces for an AI agent, focusing on observability with OpenTelemetry, governance and data protection with Microsoft Purview, and quality evaluation as the SDK evolves, then shows how to deploy the agent to Foundry Hosted Agents and maintain it with a refresh-prompt workflow.

dotnet demonstrates how to make an AI agent interact with real user data more safely by adding file access, explicit human approvals for sensitive actions, and durable memory that persists across restarts.

dotnet explains four additions that make an AI agent more practically useful: skills for loading domain knowledge on demand (including centrally managed skills in Azure AI Foundry), constrained shell access for file work, a CodeAct sandbox for writing and running code, and background agents for parallel research.

dotnet demonstrates how to stand up an agent “harness” in C# using a single AsHarnessAgent call, then incrementally add tool calling, web search, and planning so a vague request becomes a tracked, step-by-step execution loop.

yairgil explains how the Azure Copilot Observability Agent can use Azure Monitor Log Analytics Basic and Auxiliary table plans as investigation evidence, helping teams reduce log costs while still correlating alerts, metrics, traces, topology, and recent changes during incident response—especially in high-volume AKS environments.

Microsoft Developer demonstrates building the “Sparkles” AI agent in VS Code using the Microsoft Foundry Toolkit, including deploying a model, wiring up an MCP server, and validating the agent with traces and evaluation runs for repeatable testing.

kinfey explains why “AI agent infrastructure” needs stronger security boundaries than traditional model hosting, using kars (an open-source Agent Reference Stack for Kubernetes) to show how to isolate agent runtimes, broker identity, govern tools/MCP, control egress, enforce token budgets, and produce auditable evidence for incident response.

Visual Studio Code shows how to use the Command Palette to navigate faster in VS Code, including opening recent files, jumping to symbols in the current file, grouping symbols by type, and searching for symbols across the entire workspace.

Microsoft Security Research, Sagar Patil, Arlette Umuhire Sangwa, Jesse Birch and Ravikant Tiwari break down a hands-on-keyboard intrusion that starts with Microsoft Teams helpdesk impersonation, escalates through remote-control sessions and a malicious MSI, and ends with Node.js-based persistence and WinRM lateral movement, with concrete mitigations and Defender XDR hunting queries.

Cassidy Williams breaks down the new AI vocabulary showing up in developer conversations—loops, harnesses, squads, hill climbing, and model “openness”—and explains how these concepts map to practical ways teams build and improve agent-driven workflows (with GitHub Copilot as a concrete example of a “harness”).

reshmarahim announces the public preview of Radius Canvas for the GitHub Copilot app, a Canvas extension that builds a versioned application model from a repository so developers can visualize architecture, review PR impact via an application-level diff, and deploy across environments using generated GitHub Actions workflows with OIDC-based credentials.

ryanjadams explains how to keep Fabric Data Factory pipelines and Copy Jobs fully private when moving data to or from Snowflake, using Workspace-level Private Link, a private Azure Storage staging account, and Snowflake’s Azure Private Link storage integration.

Allison announces an update to GitHub Copilot enterprise-managed settings that lets organizations set any Copilot model as the default for new conversations, with optional per-team overrides via managed settings files.

Allison announces general availability of content exclusion support in the GitHub Copilot app and Copilot CLI, so enterprise, organization, and repository admins can prevent Copilot from using specified files as context and better protect sensitive code during agentic workflows.