Maybe I'm the only one that uses it, but to me it's a great success.
Anybody made their own harness and using it?
It's been a lot of fun hacking on agent internals and really understanding the nuances of how they work. It's especially useful when trying out new open source models.
After you approve it, the bots come up with the low-level design, iterate on it, implement it, run a review loop consisting of multiple steps and then merge it.
It works spectacularly well, I've never found myself having to revert anything and I do not (as a human) find bugs in literally 95+% of features shipped
We are so early into Ai, building a harness is a useful and educational experience.