Browse Artificial Intelligence News (445)

Dylan Birtolo explains a Copilot CLI rollout that makes subagent delegation more selective, reducing unnecessary handoffs and improving reliability and wait times. The post breaks down the delegation failure modes they observed, the orchestration policy changes they shipped, and how they validated the impact with offline tests and production A/B experiments.

Allison announces new controls for GitHub Copilot code review, including organization-level runner configuration (GitHub-hosted, self-hosted, or large runners), support for Copilot content exclusions at repo/org/enterprise scope, and removal of the 4,000-character limit for repository custom instruction files.

Natalie Guevara summarizes GitHub’s May 2026 availability incidents and the reliability work underway, including moving parts of the monolith to Azure, isolating database domains, and hardening GitHub Actions and Copilot services against cascading failures.

MichalBar introduces a preview redesign of Microsoft Fabric Real-Time Dashboards tile editing, adding Copilot-assisted visual authoring alongside a more code-friendly KQL workflow. The post walks through creating a visual from a prompt or query, iterating with history, and testing parameterized queries directly in the editor.

Allison explains an update to GitHub AI usage reports so GitHub AI Credits usage is reflected in the standard report fields, including what to use going forward and what changed for data since June 1.

Waldek Mastykarz explains how AI coding agents can silently scaffold outdated Node.js projects when they run npx without pinning versions, due to npm’s engine-aware manifest selection. The post breaks down why this happens and gives practical steps to make agent-driven scaffolding more predictable.

Allison announces a new /settings command in GitHub Copilot CLI that centralizes configuration into a schema-driven interface, supporting an interactive full-screen dialog, inline one-liners, and reset-to-default workflows with tab-completed keys and validation.

Allison announces the public preview of GitHub Agentic Workflows, a GitHub Actions capability that lets teams define reasoning-based automations in natural-language Markdown and compile them into standard Actions YAML, with built-in controls aimed at keeping agent-driven changes safe to apply.

Natalie Guevara explains how GitHub improved secret scanning alert quality by adding LLM-based contextual verification, reducing false positives while keeping detection coverage. The post breaks down where verification fits in the pipeline, what “better context” means in practice, and the measured impact on customer-confirmed false positive alerts.

Allison announces that GitHub Agentic Workflows can now authenticate using GitHub Actions’ built-in GITHUB_TOKEN instead of a personal access token, reducing the risk of long-lived credentials and enabling organization-level billing for Copilot CLI usage in agentic workflows.

Laura Jiang announces Copilot Autofix in limited private preview for GitHub Advanced Security for Azure DevOps, which generates suggested fixes for supported CodeQL alerts and turns them into pull requests. The post explains what’s covered in preview, how the workflow fits into existing review gates, and how usage is billed via Azure.

Leah Tran introduces Visual Studio 18.7’s pull request review experience, which lets developers open PRs, browse diffs, discuss comments, and approve or merge changes from inside the IDE for both GitHub and Azure DevOps repos.

Allison announces an update to GitHub Copilot Chat on the web that improves handoff to Copilot cloud agent sessions and adds ways to query previous sessions. The release surfaces in-progress agent status in chat, enables pulling agent logs into the conversation, and adds session search for summarizing past work.

Waldek Mastykarz explains how to measure whether an AI coding agent extension actually improves generated code, using controlled comparisons, clear evaluation criteria, and repeatable scenario runs while tracking both quality and token cost.

Waldek Mastykarz explains why piling up dozens of agent “skills” can quietly burn your token budget and reduce response quality, and how to decide what should stay a skill versus what should become a manually-invoked prompt in tools like GitHub Copilot and VS Code.

Jeffrey Fritz announces the .NET Day on Agentic Modernization livestream (June 16, 2026), focused on practical ways to modernize existing .NET applications without a full rewrite. The agenda highlights GitHub Copilot-assisted modernization, Aspire-based approaches, migration of WinForms and line-of-business apps, and adding agentic/AI capabilities.

Apoorv Gupta explains Spec-Driven Development (SDD) as a spec-first workflow for AI-native engineering, where structured specs act as the shared source of truth across requirements, design, implementation, and validation. The post introduces GitHub Spec Kit and outlines a practical lifecycle teams can adopt to reduce ambiguity and rework.

Mehrnoosh Sameki, Sandeep Atluri, Minsoo Thigpen and Abby Palia introduce ASSERT, an open-source framework that turns natural-language behavior requirements into executable evaluation pipelines for AI models and agents, generating taxonomies, stratified test cases, traces, and scored results that teams can inspect and iterate on.

Natalie Guevara explains how to give GitHub Copilot CLI real code intelligence by installing and configuring language servers via the LSP Setup skill, replacing brittle grep/decompile workflows with semantic features like go-to-definition, find references, and type resolution in the terminal.

The Visual Studio Code Team shares what’s new in VS Code 1.125 (Insiders), focusing on Agent Host improvements like the new /chronicle command set for session history, clearer file path display in chat, and updates to Cache Explorer for navigating multi-agent sessions and prompt-signature allocation details.

Allison announces an experimental public preview feature in GitHub Copilot CLI: a /security-review command that reviews local code changes for common vulnerability classes and returns severity- and confidence-scored findings plus actionable fixes directly in the terminal.

The Microsoft Foundry Team announces Claude Fable 5 (Anthropic) is now available in Microsoft Foundry, and explains how it’s used to power autonomous agents in Foundry Agent Service and GitHub Copilot, with an emphasis on enterprise guardrails, governance controls, and token-based pricing.

Shawn Henry shares a short overview of Microsoft Agent Framework (MAF) and points to a deeper design write-up on how the SDK is structured for building production-ready agents, including core concepts like agent loops, workflows, and harnesses.

Natalie Guevara explains how to define and run custom agents in GitHub Copilot CLI so repeated terminal tasks become consistent, reviewable workflows. The article shows how agent profiles live in your repo, and includes practical examples for security audits, IaC compliance checks, release notes drafting, and incident response.

Dan Hellem and Andrew Brenner announce a limited public preview that brings GitHub Copilot code reviews into Azure Repos pull requests, and walk through how to enable it at the organization, repository, and user levels. The post also documents preview guardrails and how token usage is billed via GitHub AI credits to Azure Cost Management.

Allison announces that GitHub’s security validation for third-party coding agents is now generally available, bringing the same automated checks used for the GitHub Copilot cloud agent to agent-generated pull requests.

Jon Galloway recaps Microsoft Build 2026 with the main developer announcements across GitHub Copilot, Microsoft Foundry, Azure, Windows, Visual Studio, and .NET—highlighting agentic workflows, new tooling, governance specs, and a curated set of sessions and hubs to follow up on what shipped.

Daniel Roth rounds up the key .NET sessions from Microsoft Build 2026, highlighting what’s new in .NET 11 and C# (including union types), plus sessions on agentic web apps, AI building blocks for .NET, .NET MAUI on-device AI, and tooling like dotnetup.

Microsoft Threat Intelligence and the Microsoft Defender Security Research Team break down recent phishing and malvertising campaigns that abuse popular AI brands (including ChatGPT, Claude, DeepSeek, and Microsoft Copilot) as lures, and provide concrete mitigation steps using Microsoft Defender, Entra ID, and related security controls.

Natalie Guevara answers common beginner GitHub questions, including how to set up SSH keys, create personal access tokens (fine-grained and classic), resolve merge conflicts, undo commits, sync forks, and review pull requests—plus a quick look at using GitHub Copilot for code review in PRs.

Sharlkaur introduces a preview workflow for AI-authored Power BI reports in Microsoft Fabric, using Skills for Fabric and the Power BI authoring plugin optimized for GitHub Copilot CLI. The post shows how agents can generate PBIR-based reports from prompts, iterate using screenshots, and publish to Fabric as part of an end-to-end agentic analytics flow.

Allison announces the deprecation of GPT-5.2 and GPT-5.2-Codex across most GitHub Copilot experiences, and points teams to the supported replacement models and the admin settings needed to enable them in Copilot Enterprise.

Allison announces a public preview feature that lets enterprises centrally configure and distribute GitHub Copilot CLI plugins through VS Code 1.122, using a shared settings.json so standardized plugin marketplaces, hooks, and MCP configurations are applied automatically for licensed users.

Microsoft Defender Security Research Team, Dor Edry and Amit Eliahu break down a prompt-injection pathway in Anthropic’s Claude Code GitHub Action that could leak CI/CD secrets by reading /proc/self/environ, and provide practical hardening guidance for AI-powered GitHub Actions workflows.

davidwright, Arnaud Lheureux, and Suzanne Daniels explain why architecture and governance frameworks only help when they actively change delivery decisions. Using Git-Ape as the example, they show how to turn Azure Well-Architected, Azure Policy (including NIST mappings), and CAF guidance into repeatable repo-driven assessments with prioritized findings tied to code and policy.

Allison announces that Copilot Pro, Pro+, and Max subscribers can use “Fix with Copilot” to automatically investigate and propose fixes for failing GitHub Actions jobs, pushing changes to a branch and tagging the developer for review.

Allison announces a public preview REST API that lets Copilot Pro, Pro+, and Max users start and track Copilot cloud agent tasks programmatically, enabling teams to integrate agent-driven code changes into scripts and internal tooling.

Microsoft AI Red Team updates its agentic AI failure-mode taxonomy based on a year of red team engagements, adding seven new categories and translating real-world attack patterns into practical mitigations teams can apply to deployed agentic systems.

Allison announces new GitHub Copilot capabilities: support for one-million-token context windows and configurable reasoning levels, plus guidance on how these choices affect AI credit consumption across VS Code, Copilot CLI, and the Copilot app.

tinotereshko rounds up the Build 2026 announcements for Microsoft Fabric Data Warehouse, covering GPU-based query acceleration, configurable data retention and time travel, SQL editor upgrades (including inline Copilot chat), and new T-SQL capabilities. The post also highlights DevOps-focused features like REST APIs and Git/DacFx-based schema deployments for the SQL analytics endpoint.