I'd argue this is more like "I want to build a competitor to spotify so let's decide between using mariaDB or writing an SQL compliant database from scratch"

In your example, a database is the end goal and you can either start with a premade or make your own.

Here, a social media platform is the end goal. Activitypub is a very important part of it but it's not the entire piece.

If we replace the parts of your analogy with the original your example would parse out to "I want to make a competitor to lemmies ActivityPub integration, so let's start with fedify" which is not the same as the article states.

Now, should you re-impliment a protocol yourself or use a generic library is the real question. Both have their benefits. With option A you have full code ownership and can wrap your solution around your end goal without the issue of dealing with the original to get needed changes accepted. You don't have to worry about code not written by or understood by you. With option B, you get a more robust and almost certainly more accurate implementation. Along with, for free, better integration with any service using the same library. Very useful for a federated service when talking about cross platform.

Both have many more positives and negatives of course and each person should decide on their own how to proceed.

My opinion? I think it's usually best to own anything which could feasibly be understood by a single dev. Even if each dev doesn't. Anything larger shouldn't be internal in my strong opinion unless very good, specific reasons apply that makes an external solution impossible or increadibly difficult. Most negatives of an external library also apply at that point with enough time.