Browse Artificial Intelligence Videos (291)

GitHub explains how hill climbing can be used to iteratively improve AI agent performance by running targeted evals, benchmarking results, and using the feedback loop to steer responses toward higher accuracy over time.

John Savill shares a quick Azure Weekly Update covering recent platform changes and previews, including VM ephemeral OS disk caching, Azure Functions Flex Consumption custom TLS, user delegation SAS updates, AZD extension framework work, Azure Load Testing Playwright expansion, and new Copilot-related items.

Martin Woodward shares his pick for a new GitHub Copilot feature: the GitHub Copilot app, a native Windows and Mac experience aimed at agentic coding workflows across issues and pull requests, with isolation via separate Git worktrees.

Fokko at Work demos selected new GitHub Copilot features in Visual Studio Code 1.137, including automations, attaching GitHub issues and pull requests directly in chat, and voice mode, with notes on how enterprise policies and pricing plans can affect feature availability.

Visual Studio Code shows how to connect your own Azure-hosted AI models to GitHub Copilot in VS Code using Bring Your Own Key (BYOK), so teams can control which models are available in Copilot and pick the right model for different coding tasks.

Visual Studio Code announces a new “VS Code pet” coding companion that you can trigger from GitHub Copilot Chat in VS Code using the /vscode-pet command, and invites the community to submit name ideas during a one-week window.

Microsoft Developer introduces Toolboxes in Microsoft Foundry as a way to make MCP (Model Context Protocol) integrations more enterprise-ready, with centralized governance, consistent authentication, and reusable configuration across multiple agents and tools.

Microsoft Developer discusses the MCP Triggers & Events experimental extension, showing how agents can poll MCP servers for events and be triggered via webhooks, and what new event-driven agent architectures this enables.

Caitie McCaffrey reviews the current state of the Model Context Protocol (MCP), focusing on the major changes in the 2026-07-28 specification release and what they mean for developers building AI agents and agent-to-agent communication patterns.

Microsoft Developer explains why Dynamic Client Registration becomes a scaling problem for MCP clients, and how switching to Client ID Metadata Documents turns the client_id into a durable, URL-based identity that servers can fetch on demand.

Jeremiah Lowin shows how MCP Apps can render interactive UI (buttons, charts, forms) directly inside AI clients, and walks through building these experiences in Python using FastMCP and Prefab—from basic components to dynamically generated interfaces.

Den Delimarsky explains how MCP authorization evolved, covering the original OAuth profile, protected resource metadata, changes to client registration, and how enterprise-managed authorization fits into the model.

Microsoft Developer demonstrates how to build modern Model Context Protocol (MCP) servers in VS Code, moving beyond simple local tool calls to more production-ready agent integrations with notifications, elicitation, stateless HTTP, and plugin-based distribution.

Sam Morrow discusses GitHub’s Model Context Protocol (MCP), covering what his team is shipping across MCP server and client implementations, plus the spec areas he’s most excited about and what’s coming next.

Authorised Territory demonstrates how to build domain-specific AI agent evaluators in .NET using Microsoft Agent Framework connected to a local Ollama model, then run automated checks against the agent’s responses to validate both content and output structure.

Microsoft Developer demonstrates how to build a no-code AI agent in the Azure AI Foundry Portal: create a Foundry project, deploy a model, connect an MCP server for a cupcake store scenario, then validate the agent using tracing, token usage metrics, and evaluations.

Marco Casalaina breaks down Foundry IQ, Fabric IQ, Work IQ, and Web IQ, explaining them as grounding capabilities that give AI agents the right context. He maps each “IQ” to the underlying data source (knowledge bases, OneLake, Microsoft 365 apps, and the public web) and what it enables.

GitHub explains how to make GitHub Copilot work better in a real project by giving it durable context about your repo. It covers adding custom instructions under .github, defining repeatable “skills” for tasks like testing, using custom agents for bigger workflows, and connecting MCP servers such as Playwright.

GitHub highlights research by Louis Abraham on how AI tools are influencing the language developers use in GitHub pull request descriptions, based on analysis of more than 460,000 PR descriptions and the rise of distinctive AI-associated terms and punctuation patterns.

Maddy Montaquila explains how a “Aspire-ified” .NET app can give a coding agent more context and control, including a single command to start containers, processes, and the debugger, plus the ability to manage resources and view logs from one place.

Andrea Griffiths and Meagan Cojocar demonstrate GitHub Copilot Cloud Agent running inside Slack and Microsoft Teams, showing how teams can create issues, review diffs, preview HTML, and analyze repository data directly from chat, plus what admins need to enable it safely in enterprise environments.

Visual Studio Code announces “MCP Live” (September 9, 9 AM PT), a live event with experts from Microsoft, GitHub, Anthropic, Okta, Prefect, and Amazon focused on updates across the Model Context Protocol (MCP) ecosystem.

John Savill recaps the major Microsoft AI updates from August 2026, including new and updated models, routing and content-understanding changes, and a set of Copilot and Azure Copilot improvements that affect how teams use AI features across developer and enterprise environments.

Authorised Territory demonstrates how to build an AI agent with the .NET Microsoft Agent Framework using a local Ollama model, then evaluate the agent locally with built-in checks to confirm response quality and that required tools were actually invoked.

GitHub explains why newer, larger AI models don’t automatically make better code reviewers, highlighting benchmark results where highly formatted, verbose reviews can still miss fundamental bugs.

John Savill runs through the Azure Weekly Update for 4th September 2026, covering a set of platform changes across compute, AKS, networking, monitoring, security, and Azure AI Foundry, with quick pointers to what’s new and where the updates land in the Azure stack.

GitHub introduces Project HydraFusion and shares benchmark results claiming higher verified task quality and lower estimated cost versus Claude Opus 5 on Terminal-Bench 2.1. The short also points developers to try HydraFusion as a selectable model via the /experimental feature in GitHub Copilot CLI.

GitHub introduces Project HydraFusion, a research preview for GitHub Copilot that orchestrates multiple AI models at runtime to balance quality, cost, and latency. The video explains how routing strategies like single, cascade, and critique workflows can automatically choose an execution path per task to improve outcomes without manual model selection.

GitHub shares a quick “The Download” roundup covering a new GitHub CLI capability for attaching images and videos to pull requests, plus several AI-related updates including local coding-agent models, large-scale LLM releases, and benchmarks and analyses focused on AI code review and PR-writing language.

Andrea Griffiths talks with Angela Wen about making a first open-source contribution on GitHub, from finding a good first issue through opening a draft pull request and getting CI checks green. They also cover practical expectations when working with maintainers and what “using Copilot responsibly” looks like in a real PR workflow.

James Montemagno and Matt Soucoup run a beginner-friendly livestream on the GitHub Copilot SDK for .NET, focusing on how to embed Copilot’s agent runtime into your own apps for multi-turn conversations, tool calling, streaming responses, and MCP-based extensibility.

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.

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.

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.

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.

John Maeda shares a practical take on what’s genuinely new in AI: setting a clear goal, defining how you measure progress toward it, and using iterative loops that “hill climb” toward better results as tokens get cheaper and more capable.