The way I like to speculate about the future of software is by imagining that you have infinite engineering resources. The other day, someone mentioned that they don't want to try PicoClaw (or any of the other spin-offs) because they'll miss out on cool features of the biggest project with the most contributions.
My uncontroversial prediction is that there will be a lot more hyper-personalised software, even products for single users, because that problem will go away. Agents will watch other projects for updates, or the internet for cool ideas, and instantly implement them. Commercial software in competitive spaces will quickly reach feature-parity and stay there, and it'll be harder to differentiate.
A lot of open source software today struggles because the commercial models around them don't work. People can donate money to the developers sometimes (very little in practice) or their time through contributions. A lot of open source software stalls and dies.
With infinite engineering resource, even if it's not completely free, there may be a commercial model where people share the token costs, instead of paying a subscription for a product. Then as long as people keep contributing, features keep getting developed. The more users a product has, the cheaper it becomes to develop.
The downside is there will be a pressure not to fork that software, because the userbase gets reset to 1. It's open source, but instead of donating to devs, or donating your time, you're donating tokens. But if you fork because the maintainers don't like your feature suggestion, and you want it anyway, suddenly you (or rather your agents) have to maintain that fork.
However, I suspect that this will become extremely cheap, in the same way that storage has become cheap. Cost will not be the bottleneck -- you'll be able to clear your product backlog faster than you can fill it. So the asymptote here is that there just won't be any maintainers anymore except your agents, making you your own personal software suite.
My hope is that there will be better interoperation between all this sprawl of software. It's hard to predict as I think the interfaces will change drastically (especially agent-to-agent communication). Collaborative software or social media may be the last to go, as they still have reasons for being unified (technical, or because of network effects and intentional walled-garden-ness).
Data is the only real moat left for SaaS founders. Speaking of data, I think data brokers are in big trouble if everyone starts building their own consumer apps. I would say this is overall a good thing.