I always found it too fragile. There’s a reason why smalltalk VMs have such good recovery features, it’s incredibly easy to brick your system when it’s all a network of objects, and I’m sure anyone who has worked in smalltalk has done so many times

How can it be fragile if there are "such good recovery features"?

Just restore unsaved changes when you launch the same image again. That's robust not fragile.

https://drcuis.github.io/TheCuisBook/The-Change-Log.html

I used to find it amusing that you could say `Smalltalk := nil` in the early days of Squeak, and poof! all the classes in your system are garbage collected.

(I’m not sure that’s exactly what happened, probably the system crashed before garbage collection could happen. But it was definitely a guaranteed insta-crash).