Browse All Posts (109)

Waldek Mastykarz explains why a model’s published knowledge cutoff date doesn’t reliably predict whether it can do real work with a specific developer product, and shows how workload-based evaluations give a more accurate signal when building and testing AI coding agents and extensions.

Welcome to this week's .NET roundup, where the big themes are speed, agent architecture, and safer defaults. .NET 11 performance work landed across the JIT, GC, async/runtime plumbing, and core libraries, with BenchmarkDotNet-backed results that make it easier to validate wins (and explain them) in your own workloads. On the AI side, the focus shifted from isolated demos to end-to-end agent patterns using the GitHub Copilot SDK, Microsoft Agent Framework, and MCP, with practical guidance on tool scoping, orchestration, and observability. We close with a timely ASP.NET Core reminder about protecting sensitive runtime-generated files that can accidentally become public under wwwroot.

This week's AI roundup is about turning agent work into something teams can run reliably: connected Copilot sessions across app, IDE, terminal, and PRs, plus clearer patterns for repo readiness, skills, and handoffs. On the platform side, GitHub added cost and policy levers (Auto model tiers, budget escalation) while warning about mid-October model deprecations and improving Copilot code review iteration tracking to reduce reviewer noise. Under the hood, we got a practical look at production agent engineering, from the Copilot runtime's move to Rust and the shift to diff-first inline suggestions, to the growing focus on MCP authorization, evaluation, and observability for data and ops agents.

This week's Security roundup ties together three threads: tighter controls in CI/CD and package publishing, stronger enterprise identity enforcement in GitHub, and more practical guidance for hardening AI agents and their tool access. On the supply chain side, GitHub Actions workflow execution protections reached GA, npm introduced stage-only tokens for safer automation, and GitHub completed its SHA-1 HTTPS sunset while warning teams to audit risky `pull_request_target` usage ahead of enforcement. We also look at agent security patterns (MCP governance, identity-aware tool authorization, and outcome-based observability) plus Azure hardening work across TLS, private networking DNS automation, hosted control planes, and repeatable image and configuration delivery.

This week's ML roundup focuses on how data agents are getting closer to production standards, from Fabric IQ and MCP-grounded Copilot Studio experiences to Databricks Genie evaluations that run as a continuous quality loop. On the platform side, Fabric shipped updates that reduce migration risk (BigQuery mirroring GA and an ODBC inventory tool for the ADBC cutover) and improve day-2 operations with bounded, read-only natural-language diagnostics for Fabric Data Warehouse. We also look at Azure's MLPerf Inference v6.1 results for DeepSeek-R1 on NVIDIA GB200 and GB300 systems, plus smaller governance and integration updates across Databricks and Fabric.

This week's Azure updates focus on making platforms easier to run, easier to govern, and safer to change. Plan for an action-required Azure Monitor Workspace shift to spec-compliant PromQL regex matching, and take a look at Azure Policy custom policy versioning to roll out guardrails with less risk. On the build-and-ship side, guided Copilot flows in VS Code, GA TLS improvements for Functions Flex Consumption, and Durable Functions-based dynamic workflows show a move toward repeatable delivery and long-running orchestration. The agent story gets more concrete too, with MCP hosting and tool-call governance patterns, Foundry setup and observability guidance, and Databricks pushing continuous evaluation and retention controls as standard production hygiene.

This week's GitHub Copilot updates tightened the feedback loop from pull request review to merge, with a refreshed review experience that tracks findings across pushes and makes resolution behavior easier to understand. On the governance side, admins got new levers for model lifecycle planning (including mid-October deprecations), Auto model selection tiers, and a cleaner workflow for budget increase requests. Agent workflows continued to expand across the Copilot app, VS Code, and the CLI, with more emphasis on portable sessions, safer tool calling via MCP, and stronger measurement through new usage and engagement breakdowns. Under the hood, deep dives on the Rust runtime migration and the new Inline Suggestions model show how GitHub is making agentic changes faster to run and easier to review.

This week in DevOps, GitHub Copilot moved from "chat in the IDE" toward agent-first delivery workflows that span Slack, the Copilot app, VS Code, and the terminal, with PR handling and issue triage featured heavily. Product updates added the kinds of controls teams need at scale, including improved code review, auto model selection tiers, and Business and Enterprise reporting and policies, plus a Sentry canvas for incident context. On the platform side, Microsoft previewed a guided Copilot flow for building Azure apps that uses structured stages, `az` and `azd`, and cost-aware resource summaries, while the repository readiness checklist grounded the conversation in repeatable automation, CI enforcement, and least-privilege controls.

Authorised Territory shows how to stop passwords, API keys, and other secrets from being sent to an LLM when building AI agents with the Microsoft Agent Framework in .NET and Ollama, by adding middleware that detects and redacts sensitive values before model calls.

Matt Pocock shares 25 practical “agent skills” for getting more consistent results from GitHub Copilot, including techniques for aligning on requirements, generating more useful pull requests, and running code review and architecture checks earlier in the implementation process.

DevClass reports on Microsoft’s migration of the GitHub Copilot runtime from TypeScript/Node.js to Rust using AI agents, including token costs, release cadence, performance benchmarks, and the practical risks of compiler-passing regressions when LLMs generate Rust at scale.

GitHub discusses how organizations can respond when developers adopt unapproved AI coding tools, focusing on practical governance approaches that keep teams moving while reducing risk.

James Montemagno and Pierce Boggan demonstrate an end-to-end “agent-native” workflow with GitHub Copilot, moving between Slack, the Copilot app, VS Code, and the terminal to build, test, and ship an application. The session highlights connected agent sessions, live previews, automated issue triage, PR creation and resolution with agent merge, and HydraFusion for model/workflow selection.

Allison announces generally available improvements to GitHub Copilot code review, including a clearer pull request overview that tracks findings across review iterations, smarter auto-resolution of Copilot comments, and generated commit titles/descriptions when accepting eligible batches of suggestions.

Allison announces that GitHub’s REST API can now manage the “Restrict code coverage” repository ruleset condition, enabling teams to enforce minimum coverage or limit coverage drops on pull requests programmatically across many repositories.

Allison summarizes the September 14 weekly GitHub Copilot updates, including improvements to Copilot code review, new auto model selection tiers, admin controls and usage metrics for Business/Enterprise, a Sentry canvas in the Copilot app, and new agent workflow features in VS Code.

Allison announces that several AI models will be deprecated across GitHub Copilot experiences on October 19, 2026, and outlines which replacement models to use plus how Copilot Business and Enterprise admins can enable alternatives via model policies.

javedeqbal explains how to extend an Azure Landing Zone for FinOps at scale by layering application-centric monitoring, consolidated alerting, automated remediation, and richer cost allocation on top of Azure-native governance, observability, identity, and cost-management services.

Allison announces stage-only npm granular access tokens, which let CI/CD workflows stage package versions for review while preventing direct publishing to the npm registry. The post explains how staged publishing works with 2FA approval, what permissions stage-only tokens still have, and how to prepare for npm’s January 2027 token transition.

John Savill runs through a quick Azure weekly update covering AKS Azure Linux with OS Guard retirement, Azure Red Hat OpenShift hosted control planes, Application Gateway HTTP/3 over QUIC, AVNM high-scale mesh, storage migration over SMB, PostgreSQL skills and MCP, Azure SQL DB soft delete, AVD endpoints, and Azure Payments HSM v2.

Amanda Silver reflects on Microsoft’s shift toward building developer products in the open, touching on early ASP.NET support for jQuery and the later open-sourcing of TypeScript, Chakra, .NET, and Visual Studio Code.

GitHub shares a short clip from the GitHub Podcast discussing whether needing to fine-tune an AI model just to work effectively with a codebase is a sign of deeper codebase structure and maintainability issues.

GPS unpacks a set of common “AI hot takes” and reframes them into practical guidance for day-to-day software work, covering how to review AI-generated code responsibly, where judgment matters more than tool choice, and how MCP, skills, and RAG can fit together in real agent workflows.

Visual Studio Code shows how to customize the Agents window using new background support, focusing on personalizing the AI/agent experience inside the editor.

Wiliam_Rosa explains why Azure Databricks Genie Agents need the same ongoing evaluation discipline as production ML models, and walks through Genie Agent Benchmarks, including SQL-based “Chat mode” scoring, LLM-judged “Agent mode”, and a practical workflow for monitoring and improving agent accuracy over time.

Prakhar Sharma outlines a practical pattern for treating Azure VM images like versioned build artifacts: bake once with Packer, publish to Azure Compute Gallery, validate the published version via an Azure DevOps pipeline, capture evidence, and promote only approved image versions.

Fokko at Work demos the GitHub Copilot and Visual Studio Code 1.138 updates, focusing on sharing Copilot automations, running agent sessions inside local Dev Containers, and improvements to Voice Mode session awareness, with notes on how enterprise policies and pricing plans can affect feature availability.

Sayanika walks through building a conversational analytics agent that queries data in OneLake using Microsoft Fabric semantic models and Fabric IQ Ontology, with a Copilot Studio setup that normalizes messy location inputs (e.g., abbreviations and misspellings) before running ontology-backed queries.

Tyler Kight explains an upcoming Azure Monitor Workspace change that makes PromQL regex matchers fully anchored (spec-compliant), which can cause existing queries, alerts, and dashboards to return fewer series. The post shows how to update matchers for exact/starts-with/contains intent and how to move an AMW to query engine version 2.0.

Jesse Houwing breaks down a nasty PSScriptAnalyzer failure in GitHub Actions where Invoke-ScriptAnalyzer intermittently crashed with a missing Get-Command, then shows how Copilot-assisted investigation, benchmarking, and targeted fixes turned it into a deterministic, faster analyzer with evidence reviewers can rerun.

Allison announces updates to GitHub Copilot’s impact dashboard and reporting APIs, adding 28-day feature engagement metrics so enterprise and organization admins can see which Copilot experiences are being used regularly and where enablement may be needed.

Allison announces expanded GitHub Copilot CLI reporting in the Copilot usage metrics API, adding agentic activity metrics for skills, custom agents, MCP servers, slash commands, and plugins, plus guidance on how counts are calculated and who can access the reports.

Madhura Amlekar explains a common Azure API Management failure mode where custom-domain certificate renewals break when APIM sits behind Front Door, Traffic Manager, or Application Gateway. The post shows why validation passes at deployment but fails months later, and how to redesign the edge so renewals stay reliable.

Bruno Capuano and Justin Yoo walk through updates to Interview Coach, a .NET sample that uses Microsoft Agent Framework and the GitHub Copilot SDK to build an agent-powered interview practice app with specialist agents, shared MCP tools, and a Blazor UI.

Gwyneth Peña-Siguenza demonstrates agent-first development in Visual Studio Code, focusing on practical workflows for running AI agents, controlling what they can do, understanding where sessions execute, and using built-in debugging views while building an app in agent mode.

goupadhy announces a preview soft-delete lifecycle for standard-mode on-premises data gateways in Microsoft Fabric, letting tenant and gateway admins restore deleted gateway clusters and members within a 30-day retention window instead of having to recreate them after accidental deletion.

Ron Pessner explains how Microsoft’s Secure Now experience in Microsoft Security Exposure Management turns common security guidance into prioritized actions, focusing on the fundamentals that reduce exposure as AI-driven and cross-domain attack paths become faster and harder to triage.

Microsoft Developer shows how to build and customize a hosted agent directly from the GitHub Copilot Chat window using Foundry Canvas, focusing on moving from an initial agent to a more production-ready setup with best-practice checks, generated fixes, observability upgrades, and testing with Agent Inspector.

Allison announces that the Ubuntu 26.04 GitHub Actions runner image is now generally available, and explains how the ubuntu-latest label will migrate from Ubuntu 24.04 to 26.04 with a staged rollout window and guidance to avoid broken builds.

Michaela Isaacs announces general availability of Mirroring for Google BigQuery in Microsoft Fabric, positioning it as a supported way to bring BigQuery data into Fabric for broader analytics work without building or maintaining ingestion pipelines.