Self replicating machines are one of the things I think both New Earths and Terraforming should have covered, but largely ignored. Though speculative, there are quite a few things we could say about them, both in capabilities and limits. Some of the key points:
- The key advantage of self-replicating machines is that (assuming sufficient inputs) they could provide rapid scaling of physical capability. So, things like (for example) planet diameter sunshades/power collectors wouldn't be out of the question.
- Self-replicating machine <> (doesn't equal) nanomachine. A self replicating machine might be the size of a factory with (relatively) large components, or the size of bacteria. On the other hand, nanomachines wouldn't necessarily replicate. Personally, I suspect we will have macro self replicating machines before the nanoscale kind.
- Self replicating machines can't break the laws of thermodynamics, transmute elements, and so on. They would still require energy and mass inputs, and there would still be limits to how fast certain things could happen (for instance, how fast the Venusian atmosphere could be cooled).
- They would need to be designed, programmed and overseen. A computer is great, but useless without software. To terraform a planet, there would need to be massive planning, software, and control. (It's conceivable that advanced AIs might eventually be able to do this on their own, but if they reach that level, you have to wonder why they would bother. Perhaps as a favor to their pets?)