It’s never too late, in the absolute worst case you bin some or all existing code and start over with very clear goals, and an understanding of any needed external API’s etc. Even better you can almost always extract plenty of useful bits here and there. On the other hand if you start every project thinking I need test cases, separation of concerns, continuous integration, scaling etc you’re working without a clear understanding of the end goal or how you’re getting there. Find out some tool your depending on is insufficient and you may need to toss most of what you’ve been working on down the drain with nothing to show for it.

Jumping right into whatever is the most difficult problem also means your model of the best architecture is likely to evolve as your working on the problem.

Granted this should all be on a sliding scale depending on how novel the project is.