How much do you want to spend on this? The things that are working best (at my work) are using multi-agent processes, where some agents break down work into plans, some implement code, some write tests, some review the code, some approve the results, and most of those are done as best-of-N runs.
Lots of tokens, but surprisingly good results.