GitHub Copilot vs Semantic Kernel

Copilot's cloud agent keeps gaining surface — comment-triggered automations on issues and pull requests, a reasoning-level dial per delegated task — while GitHub prunes around it. GitHub Spark stops accepting new users and new apps on August 4, with existing users cut off at the end of the month, and the separate Copilot Billing Preview app is retired into standard billing settings. Enterprise controls are moving in the opposite direction from the pruning: managed settings can now be targeted at individual enterprise teams with itemized configuration files, and model policy targeting is in public preview. Model churn continues underneath all of it, with a September 1 deprecation list already published.

The consolidation is the story. GitHub spent a year running natural-language app building as its own destination and is now retiring it, at the same time as the cloud agent grows the triggers and tuning knobs that make the same work happen inside a repository. That points at one agent surface rather than a family of AI products, with the differentiation moving to where the agent is invoked from and how tightly an enterprise can bound it. Governance is scaling by delegation — central admins hand configuration down to team-level files instead of arbitrating every change — which is what a platform does when adoption has outrun a single approval queue.

Expect more comment- and event-driven triggers for the cloud agent as the replacement path for what Spark offered, and expect team-scoped configuration to spread from managed settings and model policy to the rest of the Copilot admin surface.