The opposite is true.

There is barely any magic in the harness, the magic is in the model.

Try it: write your own harness with (bash, read, write, edit) ... it's trivial to get a 99% version of (pick your favorite harness) -- minus the bells and whistles.

The "magic of the harness" comes from the fun auxiliary orchestration stuff - hard engineering for sure! - but seriously, the model is the key item.

Yeah I agree with this. The only tool that really matters is file patching -- which you can check something like the opencode patch implementation, its fairly straightforward.

Not all percentages are weighted equally. That 1% is worth a lot more than the low-hanging 99% from your example

Is it? Look at pi, for instance.

It turns out that "most of the bell and whistles" could amount to instructing models how to use tools like tmux

This is the complete opposite of my experience.

Does your experience include writing your own agent? Send a link

I agree, but that's not something you can maintain an advantage on for long.

Perhaps there's enough overlap among the low hanging fruit that you can initially sell a harness that makes both genomics researchers and urban planners happy... but pretty quickly you're going to need to be the right kind of specialist to build an effective harness for it.