Ask HN: Which agent harness do you use for working with multiple Models?

There is a lot of stuff out there at the moment, from Multiplexer tools such as herdr and Rex, classic Tmux/CLI setups, OpenCode and pi, Codex talking to Claude and vice versa.

Some people have minimal setups like Claude and ChatGPT models talking to each other via .md files, while others let herdr spawn 20 subagents for each task.

What is your approach and why did you choose it?