AI coding tools have spread across terminals, editors, desktop apps, and cloud agents faster than most companies have worked out how to govern them. That is a problem because a rule that only applies in VS Code is not really a rule. It is just one guarded doorway next to three open windows.
GitHub’s new enterprise managed settings for the Copilot app and cloud agent are a sensible response. A single managed-settings.json can control approved plugins and marketplaces, whether users may bypass prompts, and the default model selection across the app, CLI, VS Code, and cloud agent. The cloud agent does not use every interactive setting, but it does inherit the plugin and marketplace controls.
That consistency matters more than another clever agent feature. A developer should not be able to avoid a security decision simply by moving the same task from an editor to a desktop client. It also makes the policy easier to explain and audit: define the boundary once, keep it in version control, and know which surfaces enforce it.
There will always be exceptions, and a central JSON file does not replace sensible access controls around repositories and production systems. It does remove an avoidable class of gaps, though. If coding agents are becoming normal developer infrastructure, their guardrails need to behave like infrastructure too.