Microsoft published a practical eval-focused post today, How to test agent experience changes without shipping them, which argues that agent behavior changes should be treated as hypotheses and measured before rollout because obvious docs/UI tweaks often fail in ways you would not predict. Source: https://developer.microsoft.com/blog/how-to-test-agent-experience-changes-without-shipping-them/
GitHub Copilot CLI shipped 1.0.74-0 today with a /model plan picker for plan mode, while 1.0.73 just before it fixed Anthropic subagents when extra directories are configured and resolved relative links in custom agent instructions. The repo still shows visible traction at about 11k stars and 1.8k forks. Source: https://github.com/github/copilot-cli/releases