Ideally, admins' clicks-not-code should also result as metadata in a repo somewhere. Flows, processes, page layouts, etc can all be sourced to a repo and then distributed through CI, thus enhancing the experience. It would be even more ideal if your admin could simply create a scratch org, deploy your existing metadata, make any clicks-not-code changes, then pull those changes back to the repo (this is actually the intended pattern to use).
There are, of course, situations that this won't work for (not all metadata can be deployed via API), but for the majority of use cases, your admins should follow the same steps as a developer in terms of making repo changes.
In fact, with Unlocked Packages, it goes one step further; any rogue changes made by an administrator can be overwritten by packages, thus ensuring a consistent state of metadata. This sounds annoying, but it's actually incredibly liberating knowing that you can replicate your org in mere minutes with close to 100% accuracy.