Browse All Artificial Intelligence Content (1148)

This week's AI roundup centers on a clear operational shift: centralize access, authorization, and auditability as agents move from demos into real systems. Azure API Management and Foundry's AI Gateway control plane show how teams can enforce token limits, routing, and telemetry, while MCP's enterprise-managed authorization patterns tighten end-to-end tool access using Entra ID and App Service auth. On the engineering side, stable Agent Skills for Python, long-running MCP tools on Azure Functions, and richer MCP outputs push agent tooling toward repeatable, debuggable workflows. Copilot and security updates round it out with repo-level usage metrics, more configurable code review, AI-powered code scanning detections, and agentic autofix that make cost, policy, and least-privilege design part of everyday development.

Pamela Fox announces MCP Live, a free livestream focused on the Model Context Protocol (MCP), including updates from MCP maintainers and teams at Microsoft and GitHub, plus sessions on building MCP servers in VS Code and upcoming MCP authorization work.

Authorised Territory demonstrates how to add server-side logging to a .NET MCP server and surface those log messages in a .NET console client when the client invokes a tool.

Lee Stott invites AI engineers to a Microsoft Foundry Discord round table on using the Browser Automation Tool (BAT) to let agents drive real browser workflows via Playwright Workspaces, with a focus on setup basics, practical use cases, and the guardrails needed for responsible, auditable automation.

Allison announces general availability of repository-level GitHub Copilot usage metrics via the Copilot usage metrics REST API, enabling daily per-repository reporting for Copilot coding agent and Copilot code review pull request activity across enterprises and organizations.

Allison announces an update to the Copilot usage metrics API that adds dedicated reporting for GitHub Copilot app activity in enterprise and organization reports, including active user counts and token usage breakdowns.

Allison announces new controls for GitHub Copilot code review, including reading custom instructions from the PR head branch, expanded instruction file support, repository-level setup steps via a workflow YAML, and a default-on firewall with separate configuration from Copilot cloud agent.

Allison announces a GitHub Mobile improvement that lets you start “Fix with Copilot” directly from Copilot code review pull request comments, making it quicker to address review feedback from a phone while keeping PRs moving.

GitHub demonstrates how to onboard the GitHub Copilot for Jira integration, including installing the app from the Atlassian Marketplace, authorizing access, handling SSO, and enabling organizations so Copilot can turn a Jira work item into a draft pull request.

Jordan Selig explains how Microsoft Foundry’s new AI Gateway control plane lets platform teams create or associate an Azure API Management (APIM) gateway from the Foundry admin console, while keeping the runtime on Azure App Service. The post breaks down what Foundry now governs, what still belongs in APIM, and how to adapt an existing App Service agent sample.

Dalia Abuadas argues that in an AI-assisted workflow, writing the first patch is often no longer the expensive part—reviewing, validating, and owning the change is. She proposes using a constrained “price check” patch to replace long scope debates with evidence from a real diff and tests.

Elliot Volkman shares Microsoft Security’s Black Hat USA 2026 agenda, focused on how attackers abuse “trust paths” across software supply chains, identities, cloud services, and AI systems. The post highlights sessions on npm supply chain campaigns, Azure Automation research, GitHub-focused detection ideas, and hands-on skilling with Defender and Sentinel.

VimalVerma outlines Hypervelocity Engineering (HVE) as an operating model for building and continuously evolving Azure AI Landing Zones, with a focus on platform engineering, Infrastructure as Code, Policy as Code, and security-by-design so enterprise AI platforms can scale without losing governance.

Visual Studio Code shows how to give GitHub Copilot more context by sharing your running website through VS Code’s Integrated Browser, so Copilot can reference what you’re seeing while you work.

Fokko at Work walks through selected new features in Visual Studio Code 1.129 related to GitHub Copilot, using short demos and calling out that availability can vary based on enterprise policies and pricing plans.

Waldek Mastykarz explains how to evaluate agent skills that call APIs without paying for test traffic or mutating production data. He outlines why direct evals against real endpoints create cost, state, and non-determinism problems, then shows a proxy-based approach using Dev Proxy to emulate APIs while keeping production URLs unchanged.

Hidde de Smet audits a local SKILL.md “skills” library spread across Claude Code, a shared .agents catalog, and GitHub Copilot’s VS Code setup, showing how silent failures (missing files, duplicates, stale references) accumulate over time and how to stocktake them with a checklist and a couple of shell one-liners.

olivialiu-micro explains how SQL Server 2025 enables governed outbound calls to external AI services while keeping control inside the database security model, including permissions, enablement switches, authentication options, and built-in auditing for REST invocation and external model objects.

Lee Stott invites AI engineers to a Microsoft Foundry Discord round table on scaling agent apps beyond demos, focusing on how Foundry Toolbox, Skills, and Tool Search reduce tool sprawl, prompt bloat, and auth plumbing by centralizing tools behind a governed MCP endpoint with runtime discovery.

Rafia Aqil explains Microsoft’s IQ Platform (Work IQ, Fabric IQ, and Foundry IQ) and how it adds business and organizational context to AI systems. The post breaks down Fabric’s OneLake-based data layer, ontology-driven meaning, and Foundry IQ’s managed services for RAG, memory, ranking, and citations.

Pablo Lopes announces a free, open-source course that guides developers through modernizing a legacy ASP.NET app to .NET 10 using the GitHub Copilot modernization agent, including assessment and planning artifacts, an execution workflow, and a final deployment to Azure App Service.

Microsoft Developer walks through what’s new in Data API builder (DAB) 2.0, focusing on how it sits between an app and a database to expose APIs without writing custom CRUD endpoints. The overview highlights new secrets management, broader authentication options, improved authorization controls, and integration points.

Yesenia Yser and Toby Kohlenberg explain why AI agents need to be treated as first-class security principals, with tightly scoped roles, controlled tool access, and end-to-end auditability to prevent quiet permission creep and hard-to-investigate incidents.

Lily Ma explains why synchronous MCP tool calls break down for long-running work, and shows a practical Azure Functions + Durable Functions pattern you can use today while the MCP Tasks extension rolls out across clients and SDKs.

jordanselig explains the new stable Enterprise-Managed Authorization (EMA) extension for MCP and how it differs from a centrally governed OAuth setup using Microsoft Entra ID and Azure App Service Authentication. The post includes a working sample, a local EMA lab, and practical security details for deploying an Entra-governed MCP endpoint.

Noa Kuperberg breaks down how billing works for the Azure Copilot Observability Agent in Azure Monitor, including what gets metered, which operations are billable, where users can see per-response usage, and how to track costs in Azure Cost Management alongside standard Azure Monitor charges.

GitHub hosts a live Rubber Duck Thursday session exploring the new GPT-5.6 models (Luna, Terra, Sol) inside the GitHub Copilot App, focusing on what changes when you switch models and how that affects Copilot’s behavior during real-time usage.

Authorised Territory demonstrates how to add console telemetry to a .NET AI agent harness by wiring up a trace provider and emitting traces using OpenTelemetry, with a local Ollama model (gpt-oss) used as the LLM runtime.

Manasa Ramalinga lays out a practical reference map for governing enterprise AI and autonomous agents, focusing on how to turn responsible AI policy into enforceable controls, runtime visibility, and audit-ready proof using Microsoft’s governance, security, and observability services.

jordanselig shares a reference implementation for giving an AI agent both short-term conversation history and durable, user-scoped memory on Azure App Service, using Redis for bounded session history and Cosmos DB vector search for recall, with keyless auth via managed identity and a one-command azd + Bicep deployment.

Giles Odigwe announces the production release of Agent Skills for Python in Microsoft Agent Framework, explaining how to package reusable domain expertise (instructions, resources, and scripts) that agents can load on demand, with governance features like approvals, filtering, caching, and controlled script execution.

Davide Mauri shares a practical first tip for developers who are new to Microsoft Azure and want to start building a full-stack AI application while keeping costs under control, pointing viewers toward Azure’s Budget Bytes guidance and related Azure services.

John Edward explains why AI agents that look great in demos often break down in production, and what teams need to engineer around the model to make agents reliable at scale.

Rachel Kang explains new Visual Studio features for working with GitHub Copilot models: pinning favorites in the model picker, comparing model capabilities (like context window size and vision support), and tracking context window and plan usage so long-running chats don’t catch you off guard.

Waldek Mastykarz explains how to design AI coding-agent evaluations (AX evals) that produce real signal instead of misleading scores, focusing on representative prompts, precise pass/fail/skip criteria, judge calibration, and enough repeated runs to separate variance from actual improvements.

Krishna Roy shows how to load test Copilot Studio agents by simulating real multi-turn conversations over Direct Line (HTTP + WebSockets) with Locust, then running the same workload in Azure Load Testing. The post covers measuring TTFB vs full turn completion, handling turn.complete, file uploads, and secret handling with Key Vault.

Pamela Fox shows how to build an MCP server that returns more than plain text: image thumbnails as binary tool results and an interactive MCP app (a carousel) rendered inside VS Code, so GitHub Copilot can search, inspect, and present curated image results.

Allison summarizes the June 2026 GitHub Copilot update for Visual Studio 2026, focusing on new usage tracking and alerts, a trust validation layer for MCP servers, and Copilot features that tie into pull request workflows and C++ modernization.

Allison announces updates to GitHub Copilot for JetBrains, including BYOK custom endpoint support, expanded plugin-based customizations, Claude provider support for custom agents, and local sandboxing in public preview, plus UX and reliability improvements across chat, model selection, and CLI sessions.

Allison announces a public preview feature where GitHub code scanning surfaces AI-powered security detections directly on pull requests, helping teams catch issues in languages and frameworks that don’t have native CodeQL coverage. The update explains what’s included, how it runs, enablement requirements, and how billing works via AI credits.