GitHub vs Protocol Buffers

The last six changes split cleanly in two. One half hardens governance for large customers: a repository property that applies your own config file to code scanning default setup, branch-name control for Dependabot pull requests, and CodeQL 2.26.2 catching up to current Swift and Kotlin. The other half is subtraction — GitHub Spark stopped accepting new users on August 4 and closes entirely on August 31, and the Copilot Billing Preview app is gone, its function absorbed into ordinary billing settings.

The retirements are the story. GitHub keeps standing up separate AI surfaces, then withdrawing the ones that did not become the default path, with the useful part reabsorbed into Copilot or the main product — Spark and the billing preview follow the route the Models retirement took earlier. What is being built up instead is the governance layer: nearly every non-AI change in this window hands administrators a file or a property to set behaviour across many repositories at once rather than one at a time. Meanwhile the AI work that continues is all attached to the Copilot cloud agent, which in this same window gained reasoning-level control and comment-triggered automations.

Expect the August 31 Spark shutdown to land with a Copilot-based migration path rather than a successor product, and expect more repository-property style controls as the standard mechanism for configuration at scale. On the evidence here, GitHub's AI effort is consolidating into the agent that lives in the repository, not into standalone apps.