Best Practice for AI Tooling

Since AI and tools like Devon are taking up most headlines in the industry right now, I want to know how people are leveraging these tools in secure manners and develop some sort of community standards for how to integrate these tools. The way I see it, by the end of the decade, maybe half of developer tasking can be automated via AI tooling, but everything else will be domain knowledge, and that knowledge will never be made available in public training data sets, so how do we leverage these tools without leaking proprietary information?