I mean currently yes, but writing a test/patch/benchmark loop, maybe with a seperate AI that generates the requests to the coder agent loop, should be doable to have the AI continually attempt to improve itself, its just no ones built the loop yet to my knowledge