What Happened
GitHub’s latest Copilot release added Claude Fable 5.2 and Gemini 4 Carbon for eligible paid plans. Claude Fable 5.2 requires data retention, and Business and Enterprise administrators must enable its policy. GitHub also deprecated Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra across most Copilot experiences, with replacement models recommended for affected workflows [1][2].
The Copilot app now generally supports GitHub Enterprise Server repositories, issues and pull requests, although it still requires a GitHub.com or GHE.com account and does not support GHES cloud sessions. In VS Code 1.146, experimental Inbox mode prioritizes agent sessions needing attention across workspaces, and agents can run experimental model comparisons. The Copilot harness is now the default for new VS Code sessions; the Local harness is scheduled for removal in February 2027 [1].
GitHub also opened public-preview bulk starts for AI Scan across up to 20 repositories. The scans examine default branches, including languages CodeQL does not support, and place findings in the code scanning backlog. A Copilot license is not required [3]. Sourcegraph’s survey of 1,028 people responsible for software development reinforces the operational question behind these releases: how teams understand, review and change large codebases as AI writes more code [5]. The research notes do not establish comparable releases for Cursor, Windsurf, Replit, JetBrains or Continue.dev.
Why It Matters to Businesses
Model availability is now a configuration and governance issue, not just a developer preference. A deprecated model can disrupt a pinned workflow; a newly available model may introduce a retention decision. Meanwhile, more agent sessions increase the need for clear ownership of reviews and follow-up work [1][2].
Kimbodo Engineering Perspective
We would treat the coding assistant as part of the delivery system, not as an interchangeable editor feature. Model choice, repository access, session behavior and security findings need documented owners and testable policies. AI Scan can extend coverage, but a default-branch scan is not a substitute for pull-request checks or human review [3].
How We Would Implement It
- Inventory pinned Copilot models and integrations, then test recommended replacements against representative coding and review tasks before switching defaults [2].
- Decide whether Claude Fable 5.2’s retention requirement fits company policy before an administrator enables it [1].
- Test GHES access and the new VS Code harness with a small team, documenting account requirements and session limitations before wider rollout [1].
- Pilot AI Scan on selected repositories, assign ownership for its alerts, and compare findings with existing CodeQL and review workflows [3].
Risks, Costs and Security
The immediate risks are untested model migrations, retention settings that conflict with policy, and security alerts without a triage owner. Bulk scanning may increase review workload; the preview should be measured by actionable findings, not scan count. GHES customers should also validate the account and cloud-session constraints before promising a seamless enterprise workflow [1][2][3].
Where Kimbodo Comes In
Kimbodo builds and operates this in production for businesses — see our AI Application Development practice, or Estimate My AI Application.