It's a good point, my first thought is that the AI can replace surface level software easily, but the structure and context it needs to "recreate" git in a accurated and at a professional level would consume trillions of tokens, and even then the chance that it is usable from the get go it's low.

The vibe coders can replace SPA, python scripts, small and low context things. Everything else it is outside the scope. It’s more of a testing tool: you give it snippets, input/output examples, and instructions on how and where it should be used.

PD: My personal take on SaaS, is that is about to go bullish.

my anecdote : I can have an agent harness run every tool that github provided an interface for on my own machines, and I used GH primarily as a git host rather than a collab tool.

To be clear : I could have done this before -- the agent/harness make it seamless, integrated, easy. The same job GH did.

GH was useful for me when it was less hassle than running git and friends. That is no longer the case , agent harnesses are less hassle -- so the friction of moving away from a Microsoft corp with terrible reliability is virtually non-existent.

tl;dr : GH was popular because it lowered the barrier to entry for source control and was reliable enough to take serious. Now it's not reliable, and there are even lower barriers..

This is sort of the same story that plays out for any org that exists because of a tool, really -- and it's not just a CS phenomenon. How is Xerox doing nowadays? And they had the leisure of having made and owned the tool rather than having OSS gift it to them.