The technique isn't limited to 8 per se, it's just for every additional fragment you want to add, you require an additional joining event to occur. Eventually, your cloning efficiency will just bottom out, and you'll get no positives. Assembly of >8 fragments is possible though - I know a few people who have managed up to 15 or so, though the efficiency again is hugely fragment-length dependent (small fragments are easier than large ones).
To increase the number of fragments, you can try step-wise assembly, sticking subsets of fragments together to make progressively larger ones, or optimise your fragment ratios and overlap lengths. This is a bit of a 'dark art' though - it would be a lot of trial and error with lots of potential variables.
The technique is called 'seamless' as it, in theory, leaves no scars (deleted or inserted bases). NEB has a great page explaining it, but this figure basically summarises:

There would be nothing stopping you going and annealing fragments and then extending them, but this is more like a "joining PCR", and Gibson assembly is designed to mitigate the need to do this. Joining PCRs are very tricky to get working, especially if your fragment sizes differ significantly. The Gibson mastermix contains enzymes in compatible buffers with all the necessary cofactors to allow them to work together.