Ideal Error Handling (NA() vs NA, etc.)

What is everyone's process for error handling?

In the past, I always converted errors to text "NA". This is also consistent with blank results provided via a third party addin/data vendor I use.

Recently, I have been using AI tools to speed up some development steps and see it constantly suggesting to use NA().

I understand the benefit of the cell being an actual error, such as it triggering on ISERROR. However, errors don't play nice with FILTER function and I am finding myself having to constantly build in error handling into every single formula. It feels like a lot more work.

So, what is everyone's best practice for errors?