That's not a new flip side, though. That's always been with us. SaaS will be more able to take on more requirements then they used to be but no one SaaS will ever be able to take on everything. And the decision between "roll our own and own it forever versus buy this one service that
almost does everything we need, but not quite, but maybe it's worth living with it because it's still better than rolling it ourselves" isn't going anywhere either.
I would not want to be a SaaS that offers some really simple service that can be replicated in a heartbeat, though. Something like "how do I pay people all over the world" is already very complicated, and over the next decades as governments start writing laws with the understanding that AIs can implement them in code no matter how complicated they are, it's likely these problems will become even more complicated and even more important to just buy a service that can deal with them. (I'm not celebrating that, merely predicting it.) But I sure wouldn't want to be selling some super simple scheduled reminder service or something else really small.
In the worst case, envision a world where home owner associations or townships or whatever other local governmental division of just perhaps a few hundred people start levying sales taxes, with their own complicated exclusions and offsets and conditions, because LLMs make it possible to handle the code for all of the literally hundreds of thousands or millions of such jurisdictions. Even if you can throw tokens at that problem to solve it yourself, you probably don't want to. And again... I'm not celebrating that. More a world-weary bowing to the inevitable despite it being an obviously bad idea.