The first version of anything is an instrument, not a product. Its job is to convert an argument into evidence — to turn a room full of opinions about what people want into a much smaller room of facts about what people did. Judged as a product, almost every first version is a disappointment. Judged as an instrument, most of them work beautifully, and then get thrown away for the wrong reason.
What I find striking is how rarely the second version is treated as the main event. The first gets the launch, the excitement, the internal demo. The second gets scheduled after something more urgent, and is usually described with some variation of "cleanup" — which is a strange word for the moment when you finally know what you are building.
What the first version teaches
A first version answers questions that cannot be answered any other way. Not the question you thought you were asking, usually. You ship a feature to find out whether people want it, and instead you find out that a quarter of them want something adjacent and unnameable that you would never have proposed in a planning document.
You also learn where the seams are. Every build reveals which of your abstractions were real and which were wishful — the ones you keep having to reach around, the ones that made three unrelated features awkward in the same way. That information does not exist before you build, and it degrades quickly after: six weeks later the awkwardness has become normal, and nobody remembers which parts were surprises.1
The window
So there is a window, and it is short. Right after a first version ships you hold two things at once that almost never coexist: fresh evidence about what people actually do, and a still-vivid sense of where the structure fought you. A second version built in that window is cheap and decisive. The same work attempted a year later is an archaeology project.
A rewrite is what a second version becomes when you wait long enough to forget why you wanted one.
This is the practical argument for treating the second pass as planned work rather than a reaction. Not because rewrites are wasteful — sometimes they are the only honest option — but because the cheap moment is early, and it closes quietly.
How to make it happen
- Schedule it before you ship. Put the second version on the plan alongside the first, with the explicit note that its scope will be decided by what the first one reveals.
- Write down the surprises as they happen. A running list, kept during the build, of every place the design resisted. It takes minutes and it is the entire input to the second version.
- Let it remove things. The best second versions are smaller than the first. If yours is only additive, the evidence probably was not consulted.
There is a version of this essay that is about software, and a broader one that is about any work where the first attempt is how you learn what the work is. Writing, obviously. Organizational structure, painfully. The pattern is the same: the first attempt buys information, and the information has a shelf life.
The people I have learned the most from all shared this habit. They were not unusually fast or unusually right the first time. They were unusually willing to go back while the lesson was still warm.