I use OpenCode with a three agent combo (architect, developer, reviewer), as I've found it's crucial that different models write the code vs review it.

More details here:

https://www.stavros.io/posts/how-i-write-software-with-llms/