Microsoft Agent Framework kept moving toward production: Python 1.11.0 (Jul 15) adds message injection, progressive MCP disclosure, and stable skills support, while .NET 1.13.0 (Jul 3) adds skills-source context, per-user session isolation for Foundry hosting, skill approval options, and file-editing tools. Sources: Python release post and agent-framework releases.
Foundry Toolkit for VS Code 1.6.4 (Jul 15) adds Agent Optimization preview and flattens the Foundry navigation into cleaner My Resources and tabbed Models pages, with reliability fixes for hosted-agent creation and the playground. Source: WHATS_NEW.md.
GitHub Copilot CLI 1.0.72-1 (Jul 17) adds --plugin, --mcp, and --skill mutation flags plus skill-removal support, which is a pretty clear sign the CLI is becoming a first-class control surface for agent tooling. Source: copilot-cli releases.
GitHub Copilot in VS Code’s June 2026 release bundle, published Jul 8, sharpened the agent workflow with browser tools, parallel sessions, clearer cost visibility, Marketplace model discovery, and better Autopilot behavior. Source: GitHub Changelog.
GitHub Copilot in Visual Studio’s June update, published Jul 14, adds usage tracking and alerts, trust validation for updated MCP servers, and general availability for the C++ modernization agent. Source: GitHub Changelog.
GitHub pushed a strong enterprise-observability batch on Jul 17: repository-level Copilot usage metrics are now GA, the usage-metrics API now includes the Copilot app, agent session streaming is in public preview across CLI/VS Code/Visual Studio, and Copilot code review got repo-level setup customization. Sources: repo-level metrics, usage metrics API, session streaming, and code review config.
Kimi K2.7 Code became generally available in GitHub Copilot and got a big HN reaction, with the thread already above 400 points; the key signal is that Copilot now offers an open-weight selectable model hosted on Azure. Sources: GitHub announcement and HN discussion.