In a more holistic view, a software system includes the program and its developer community. In that sense, certain software such as Emacs and Linux has clearly proved itself antifragile, whereas most other software is clearly fragile.
The test for antifragility is this: Does the software improve with the presence of time and when faced with change?
If it does, it is antifragile. No need to add more conditions like "no developer changes are needed". Developers are part of the software system.
For that meaning, don't use the word "software". The word software has a well-understood meaning that a developer community surrounding a code base doesn't meet.
> The test for antifragility is this: Does the software improve with the presence of time and when faced with change?
Software cannot do that on its own. Remember, we're using the word with its accepted meaning.
Software systems are embedded in broader systems which include the developers, users, etc. But it is not helpful to communication to call those broader systems the "software system", which has a well understood meaning.
> Just wondering if anyone ever developed some more in-depth thoughts of the possibilities of "anti-fragility by design" ...
But that's classic computer science. There's no debate about manual and automatic testing and intelligent error handling -- all software needs to have it.
> but building stuff that somehow "gets better".
To "get better" without human intervention would require a computer system so intelligent that it creates local violations of entropy, sort of like people, and that imitates evolution and natural selection.
We haven't remotely gotten to that point yet -- computers and computer languages would need to be much more powerful than they are. They will eventually be, but more time and energy will be required.