Everybody’s different, but here’s what I do . . . My workflow is built entirely around the video editor, as though this was “a real movie.”
I start by creating the set to scale. (How high is that wall? How long?) The set is at this point un-decorated. Then characters: once again, simple geometric shapes, but “to scale.” Basic armature: just enough to get by.
Next, the props. Each one is at this point a “bounding box.” The finished prop will just fit inside of it. But it hasn’t been built yet. None of it has. But, “that stand-in thing” is of exactly the right size / dimensions.
And, everything(!) – sets, actors, props, files, scenes, cameras – has "a meaningful name."
All of these are in a library of individual files, designed from the start to be “linked.” In due time, you’ll work on each linked file to replace its contents with “the real thing.” But, you can defer(!) that for a very long time.
I then try to "start shooting." A shot-file consists of a base scene – linked to the background set and all needed assets – which contains some [single …] bit of action. Then, “shot scenes” are linked to that within the same file. Each one has a (named …) camera (and camera rig), a particular output filename, and a meaningful scene-name. I then “shoot” each one. The shot begins slightly before the action ends, and ends slightly after, to give me a “tail” on both ends for editing purposes.
I simply use the Workbench renderer (for speed), and “stamp” each frame with the following metadata: filename, scene name, camera name, and frame number. This allows me to instantly see, in the editor, exactly where each frame came from.
Now, I take all this “footage” to a video editor of my choice. Could be Blender’s, could be any one of several good open-source products, or (as in my case) a familiar commercial product.
I have done everything up to this point "to get to here." In the editor, looking at “footage,” and using my imagination.
As you work with the editor – which is where you will spend most of your time – you begin to get a feel for the timing, pace, and rhythm of “the movie.” Very importantly: every shot – “rough” though it now is – actually corresponds exactly to the “real, final” version of the same shot. The point is that you can shoot it all “fast and easy.” Doesn’t matter if you actually use the shot, in the end. But you can make very detailed decisions, “using just this.”
Each shot corresponds to one action. If the character’s planned to go through several moves, they’re still shot “one at a time.” (Recall: base scene contains the action using all-linked asset, then linked camera scenes in a single file contain the shots. If I want to choreograph the entire move, I can freely do so.) This gives me leeway during edit.
“If I was a real director, directing real actors on the set, what would I direct them to do?” Then, if I was a real cameraman/cinematographer, using real equipment (booms, dollies, trucks …), “how might I shoot it?”
I actually try to get as close as possible to “final cut” at this point. (And, believe it or not, you can actually do it.) Then, once I know what “the movie” is actually going to need, I can begin working on the various linked files, one at a time. Simply re-render the appropriate stuff after each change, and “magically” the movie begins to appear on your editing screen. When you get to “final render,” you now know precisely what material you actually need.
I also keep a daily “Captain’s Log.” What did I do today? What unanswered-question do I have? (I’ll back-fill the answer when I find it.) What’s next? What’s the URL/hyperlink of the cool video that I just found? And, what was I thinking as I was looking for it? Just write it down and never change what you wrote. Because, your memory isn’t as good as you think it is. (I was going to add a nice closing comment here, but I forgot …)