That ML-sidecar for recommending the actual change is a very correct place to put that technology: right in the line where someone is already reviewing the code

That ML-sidecar for recommending the actual change is a very correct place to put that technology: right in the line where someone is already reviewing the code, so they can confirm the ML suggestion looks correct before adding it. I can't overstate how valuable it is for reviewers to give the actual change they're recommending alongside the description, but it happens too infrequently if we make the reviewer write it every time. Actual code is the cleanest language to communicate ideas, but it's a time-sink and a mental drain to turn the English description into code snippets every time. Automating that is precisely the right place to save time for reviewers.