I think pure style (e.g. that a linter would catch) and DRY could be treated separately. The former should always be followed. To the latter: I find weak or incorrect abstractions much harder to follow than repetitive code. And I find those much more commonly when implemented early rather than later.
There's definitely a balance, and if you know the type of problem you are solving ahead of time you can probably abstract early. But if you're just kind of feeling around in the dark, which seems to be relatively common in complex business applications (for instance), I much appreciate an implement-test-reflect-refator cycle.
For actual redesign, you end up with a long true refactoring iteration when no new features can be added. There is no guarantee you will not repeat the same mistakes or make it worse.
There's definitely a balance, and if you know the type of problem you are solving ahead of time you can probably abstract early. But if you're just kind of feeling around in the dark, which seems to be relatively common in complex business applications (for instance), I much appreciate an implement-test-reflect-refator cycle.