GitHub Copilot had a dense enterprise update today: Claude Sonnet 5 is GA, Copilot Agent is now a first-class option inside JetBrains AI Assistant, and per-user AI credit budgets are now available for cost centers. The useful read is that GitHub is bundling model choice, agent entry points, and spend controls into one admin surface. Sources: GitHub Sonnet 5 GA | GitHub JetBrains agent | GitHub budgets | JetBrains integration post | Reddit thread
github/copilot-sdk shipped v1.0.5-preview.1 with MCP OAuth host token callbacks, which matters for agent sessions that need to authenticate against protected MCP servers without awkward glue code. The repo also has real traction, sitting around 9.5k stars and 1.3k forks. Sources: release notes | repo | actions/release activity
Microsoft 365 Copilot’s June update broadened the planner/notebooks story: the official post calls out Planner Agent, while the notebooks update says Copilot Notebooks are gaining broader references/access and customization. A Microsoft PM also said mind maps in Copilot Notebooks are GA worldwide and rolling out broadly across teams. Sources: June Copilot update | Notebooks update | Reddit PM thread | Planner Agent LinkedIn post
Microsoft’s Agent Governance Toolkit kept getting the right kind of external attention: commentary today emphasized runtime policy enforcement, privilege rings, sandboxing, tamper-evident audit logs, and kill switches as the right layer for production agents. The repo itself is already a meaningful open-source signal, with thousands of stars and active discussions/issues. Sources: Microsoft repo | Microsoft Open Source announcement | LinkedIn analysis | LinkedIn commentary
OpenAI published a fresh deployment-oriented paper/update on simulation: “Predicting model behavior before release by simulating deployment.” The practical signal is that OpenAI is trying to validate complex agent behavior in a tool-rich environment before rollout, which is exactly where real agent failures hide. Source: OpenAI post