The latest Copilot update brings three distinct shifts: mobile‑app support for GitHub Enterprise Server (GHES) repositories, two new AI models (Claude Fable 5.2 and Gemini 4 Carbon), and a set of experimental VS Code agent capabilities. For engineers responsible for building or operating AI‑assisted workflows, each shift alters the way resources are provisioned, authenticated, and observed.
Enterprise Server support in the Copilot app
Copilot now lets users browse and act on GHES repositories directly from the mobile app, a feature marked generally available. Access still requires a GitHub.com or GHE.com account, and cloud‑based session features remain unavailable for pure GHES instances. The practical implication is a split authentication surface: teams must maintain a cloud identity for app usage while preserving on‑prem GHES credentials for repository access. Automation scripts that previously relied on a single token may need to incorporate dual‑token handling or gateway services to bridge the two identity domains.
New AI models and data‑retention policy
Both Claude Fable 5.2 and Gemini 4 Carbon are offered to Copilot Pro+, Max, Business, and Enterprise tiers. Claude Fable 5.2 enables data retention by default, but Business and Enterprise administrators must explicitly enable its policy, which is off by default. This introduces a compliance consideration: organizations must audit whether retained data aligns with internal policies or regulatory requirements before activating the model. The models’ availability across tiers also means that CI/CD pipelines can now target higher‑capability LLMs without changing subscription levels, but they must respect the optional retention flag.
VS Code agent enhancements
The VS Code 1.146 release adds several experimental agent tools. Compare Models lets an agent locate available models, run side‑by‑side comparisons, and summarize results. Inbox mode aggregates agent sessions across workspaces, surfacing those that need attention first and allowing dismiss, hold, or snooze actions. Chat windows can be detached from the Agents pane, enabling custom layout with editors and terminals. Extension management now occurs within the Agents window, with warnings for unsupported features. Because these features are experimental, teams should treat them as optional pilots and monitor stability before embedding them in production pipelines.
Operational tweaks and accessibility
Project automation can now be scheduled to the minute or via custom cron expressions (e.g., weekdays at 9:30). Multilingual dictation supports French, Spanish, and additional languages after downloading the voice model. Issue and pull‑request filters now include custom fields such as a Project’s Size, improving query precision. Accessibility settings combine high‑contrast filters with color choices and preview them before application; a new GitHub Dimmed theme offers a dark‑only palette. These UI enhancements can reduce friction for globally distributed teams and support accessibility compliance audits.
Related CloudNinjas coverage: AI engineering.
What This Means For Practitioners
Evaluate whether your organization needs GHES integration in the Copilot app and adjust identity management to accommodate dual accounts. Review the default data‑retention setting for Claude Fable 5.2 and align it with compliance frameworks before enabling the model. Pilot the experimental VS Code agent features in a sandbox environment, and decide if Inbox mode or Compare Models adds measurable efficiency. Finally, begin migrating any remaining workflows to the Copilot harness ahead of the February 2027 deprecation of the local harness to avoid future disruption.

