AI Didn't Make Software Cheaper. It Made Fit Affordable.

There is a lazy version of the AI story that says software is now cheap. It isn’t, though that is not the interesting change. Plenty of software costs exactly what it did. What collapsed is something more specific and more consequential: the cost of building software that fits your business precisely.

That distinction reorders a decision most companies settled years ago.

The old math

For two decades the logic was airtight. Off-the-shelf solved the common 80% (the parts every business shares) at a fraction of the cost of building. Bespoke software was reserved for the few with IT budgets in the tens of millions, because the binding constraint was always the same: skilled engineering hours, in large numbers, for a long time.

So everyone else adapted. You took the near-fit, bent your process to it and hired around the gaps. Not because anyone preferred it, but because building the thing that actually fit was a rich company’s privilege.

What actually changed

AI didn’t remove the need for engineering judgment. It removed the need for so many hours of it.

A small, senior team with AI-augmented delivery now ships in months what a large team once took years to build. The boilerplate, the plumbing, the test scaffolding, the first draft of nearly everything (the work that used to consume most of the calendar) compresses. What is left is the part that always mattered most: deciding what to build and architecting it well.

The cost driver of custom software was never the idea. It was the labour. Cut the labour-hours by a large factor and the entire economics of fit move with it.

Why the equation flips

Hold the five-year math in mind. The reason renting usually won wasn’t that renting was cheap forever. It is that building was prohibitively expensive up front. Drop the build cost by enough and the crossover where owning beats renting moves from “year fifteen, theoretically” to “year two, in practice.”

The build that didn’t pencil out on a mid-market budget in 2022 pencils out in 2026. Not because the software got cheaper to run, but because fit got cheaper to make. The privilege became an option.

The caveat that matters

Cheap-to-build is not the same as trivial. This is where the lazy version of the story does real damage. AI lowers the cost of construction; it does nothing for the cost of knowing what to construct. A payroll system that doesn’t understand provident-fund rules, an order flow that doesn’t model cash-on-delivery, a rental platform blind to post-dated cheques: none of these fail because they were hard to code. They fail because nobody who understood the domain shaped them. Faster building makes domain judgment more decisive, not less, because the bottleneck moves from typing to thinking.

Re-open the question you closed

If you decided build-vs-buy in 2019 or 2021, you decided it under the old math. The inputs have changed. The honest move is to re-run the question for the one or two systems that have been quietly taxing you ever since, not as a rich company’s privilege this time but as a number that now works.

A two-week Fit Assessment re-runs that decision on current economics and tells you whether the build that didn’t make sense three years ago makes sense today.