The Fastest Way to Build the Wrong Thing Is to Skip the Thinking

In 1966 a NASA engineer named Robert Propst published a study on the relationship between planning and construction cost in large infrastructure projects. His finding had been observed in shipbuilding, in civil engineering, in the construction of railways across three continents. A change caught at the specification stage cost a fraction of the same change caught during the build. A change caught after the thing was in use cost an order of magnitude more again. The numbers varied by industry and by project. The shape of the curve was always the same. The later a decision was made, the more expensive it became to make it.

This is a property of complex systems rather than a principle of project management. The further a structure develops from an unresolved decision at its foundation, the more of itself it builds on top of that decision. By the time the decision surfaces as a problem it has load-bearing material stacked above it. Changing the foundation means disturbing everything it is holding up.

The people who understood this built specifications before they built anything else. The specification was the cheapest place in the entire process to make a decision. It was the moment where changing your mind cost nothing, where the gap between what you thought you were building and what you were actually building could be closed with a conversation. Everything that came after carried a higher price.

Software inherited this principle and spent decades in productive tension with it. Agile methodologies, lean development, rapid prototyping, each a legitimate response to processes that over-specified and under-delivered. The pendulum swung toward shipping. Moving fast became the culture. The specification shrank and the build began earlier. The results were often good enough to validate the approach.

Then AI coding tools arrived and the friction in the build phase disappeared entirely. What used to take a developer a week now takes an afternoon. What used to take an afternoon now takes minutes. The build became frictionless at exactly the moment the specification was already at its thinnest.

That friction had been doing work nobody noticed until it was gone. It created natural pauses where decisions got made, where misalignments surfaced, where the thing being built had to be compared against the thing that was wanted. Fast enough that nobody felt it as a delay. Slow enough that the thinking had somewhere to happen.

A frictionless build running ahead of an unresolved specification is the accumulation of unconscious decisions at a pace that makes them progressively harder to unpick. Every sprint compounds the decisions that were never made explicitly. Every feature added to an unclear foundation makes the foundation more expensive to revisit. The debt is not in the code. It is in the absence of the thinking that should have preceded it.

The teams discovering this moved fastest. The codebases that look most impressive at six months and most expensive at eighteen. The products that shipped quickly and scaled in directions the architecture was never designed to support. The builds that got done and then got done again because the definition of done was never clear enough to recognise.

Propst’s curve has not changed. It has become more expensive to ignore.

Constructable was built for the thinking that has to happen before any of this starts. A plain-English description of what you are trying to build becomes a PRD, an architecture document, a security model and a launch checklist before a line of code is written. The specification as the product. The foundation laid while changing your mind still costs nothing.