Microsoft now has a cleaner split between the developer/runtime side and the IT/security side of agent ops: the new Foundry Control Plane + Agent 365 post says Foundry is for build/operate/govern, while Agent 365 is the governance layer for enterprise oversight. That is the strongest Microsoft “agent platform” framing of the day. Sources: Foundry Control Plane and Agent 365 ; Azure product page ; LinkedIn context
Microsoft also published a practical telemetry/auditing write-up for the Agent Governance Toolkit today. It turns the toolkit from “interesting security posture” into something teams can actually instrument in .NET Core, with the repo showing real traction at about 4.6k stars and 664 forks. Sources: Auditing and Telemetry for the Agent Governance Toolkit ; microsoft/agent-governance-toolkit
GitHub’s issue system got more agent-friendly today: issue fields are now generally available, and MCP integration means Copilot and other AI tools can read and set structured issue metadata when creating or updating issues. GitHub says more than 40,000 orgs adopted the preview already, so this is not a toy launch. Sources: GitHub Changelog
GitHub paired a practical Copilot CLI onboarding post for C# and C++ with a visible traction signal: github/copilot-cli is sitting around 10.9k stars, 1.7k forks, and 291 commits, which makes it more than a demo repo. The article is basically “how to use the terminal agent well inside Visual Studio workflows,” not a fluffy announcement. Sources: Building C# and C++ Apps with GitHub Copilot CLI and Visual Studio 2026 ; github/copilot-cli ; Reddit live-demo thread
GitHub added another enterprise governance control for agent spend: cost centers can now cap how much of the included AI credit pool each cost center can draw, on top of the per-user budgets shipped a few days ago. This is the billing plumbing that makes bigger Copilot/agent rollouts easier to control. Sources: Cost centers now support AI credit pools ; Per-user AI credit budgets available for cost centers
OpenAI’s new “How agents are transforming work” paper/article is worth a read for the field-level signal alone: by June 2026, the 99th percentile of Codex users were regularly generating more than 60 hours of agent turns per day, and OpenAI says Codex has become the primary AI tool across every department internally. That is a strong data point for long-horizon agent adoption outside just engineering. Sources: OpenAI article ; paper link