Yeah, but if your aim is to write FOSS clones of proprietary software (be it SaaS or otherwise, or even if your aim is to write something bespoke rather than a clone of something that exists), doing it with LLMs is either going to take you longer than writing it yourself or result in laughably glitchy, inferior software.

Remember the news that went around not too long ago about how a company "used AI to make an Open Source Photoshop"? The part everyone left out is that even the company behind it calls their "Open Source Photoshop clone" "early alpha". (At the top-right.) It's barely a tech demo. Not a product. And I doubt it ever will be, despite the hype about it in the news. If it ever does end up rivaling Photoshop, it'll only be because they gave up on using LLMs (which will be more work in the long run than if they just hadn't used LLMs in the first place -- or just contributed to Gimp) or because Photoshop will enshittify enough that the terrible quality of whatever PhotoCraft eventually becomes will be preferable.

Same with any other effort to make FOSS clones of existing proprietary software.

I'm 100% all for making compatible FOSS clones of existing proprietary and SaaS software specifically to break the monopolies held by companies like Adobe and Microsoft and Amazon. I would love to be able to recommend ReactOS to relatives. And I've played a "lot" (compared to most people, at least) of FOSS re-implementations and/or clones of originally-proprietary games or game engines (OpenMW, OpenJK, DXX-Rebirth, Lilium Voyager, Luanti). But LLMs aren't magic and won't help.