According to the tests in the corefx repository: https://github.com/dotnet/corefx/blob/master/src/System.Reflection/tests/TypeInfo/TypeInfo_IsEquivalentTo.cs#L17
this was compatible with .NET core.

And either GetTypeInfo() or AsType() call seems unnecessary. (Previously the Type was compared with TypeInfo, and now - TypeInfo is being compared with Type)

But that example you are linking is actually doing GetTypeInfo and then passing a type (not a typeinfo) to the IsEquivalentTo method, which is exactly what I'm doing too. BTW, I did this change in my coreclr branch because it did not compile for me, I'm not speculating. This works in full dotNET because TypeInfo inherits from Type, but that's not the case in netcore, and it's why we use the GetTypeInfo extension method and the AsType

Then this equality check isn't symmetric, but I believed that it is..

yup, it's not. That's probably why they aren't just using IEquatable in that test.
With this new insight, do you recommend reverting your original PR or keep it this way? or use a different comparison?

There's no need to revert original PR, as its purpose was in extracting allocation causing code into separate method. So it's OK as it is.

but what about doing this at least:
return typeInfo.GetFields().All(f => f.FieldType != typeInfo.AsType()) && IsOrleansShallowCopyable(f.FieldType));
This way we don't get the TypeInfo unecessarily in the first place

Updated with that approach and added some test coverage