Build vs. Buy Was the Question. AI Moved the Line

“Don’t reinvent the wheel.”

It’s one of the most repeated lines in technology, forged from years of hard lessons: buy for parity, build for advantage. In 99% of cases, it makes no sense to build a custom purchase order system when you can buy one from a hundred different vendors. Development is expensive, so save your engineers for the things that actually make you money. Most technology leaders can recite the rule in their sleep.

But unchallenged rules have a way of causing problems. When you think about it, this was never really a rule. It was a price quote, and nobody has checked whether the price has changed.

Newton’s fourth law of software might as well be: “an engineering org in motion tends to stay in motion until acted upon by an unbalanced force called the CFO.” The old rule of thumb was “buy the commodity, build the differentiator.” It was just shorthand for where those two forces balanced. Somewhere along the way, though, we printed it, laminated it, and filed it under settled science.

AI-assisted development is the new unbalanced force. It is quietly tipping the cost on the “build” side of the equation while most companies are still navigating by an old rule of thumb instead of checking the latest quote.

The mistake is treating build-versus-buy as a principle when it was always a calculation.

AI is changing the math, and rapidly. Not to zero (anyone selling you zero is selling something else), but the cost is moving materially for a large class of software. When the cost of building drops, the crossover point slides.

That does not mean companies should build payroll systems, email platforms, or ledgers from scratch. The rule still holds for true commodities, especially where compliance, reliability, and scale dominate. A vendor can amortize that regulatory and operational burden across thousands of customers in a way you will never match alone.

But the pool of things worth building yourself is growing. In many cases, buying means contorting your business to match a vendor’s assumptions, and that contortion can leave you executing suboptimally.

AI is allowing companies with real but nebulous ideas to turn those ideas into requirements and working prototypes in a fraction of the time and cost. The design and development compression is real. I see these cases every day.

Sometimes the opportunity is a known pain point that has been around for a decade. It was too expensive to fix before, so everyone just lived with it. Now, with new economics and clearer requirements, it finally makes sense to address it.

The bigger shift is around ideas that never used to survive the budgeting process. Three years ago, a developer would have floated a product suggestion in a retrospective. The development manager would likely have laughed, and even if they didn’t, the idea would never have made it to the backlog. Even if it did, it would never have been greenlit for budgetary reasons.

Now, every couple of weeks, a developer doesn’t sleep for a weekend and shows up Monday morning demoing a beta app to their team with the potential to materially improve how the business runs.

That changes what is possible. But it does not make software free.

AI is compressing the writing of code, but it has not yet compressed everything downstream of the build. Many companies already underestimate the ongoing cost of maintaining homegrown software. Over the next couple of years, AI may make that problem worse before it makes it better, because companies are about to generate far more code, far faster, than they ever have before.

The same organizations that were already below par at testing and support will now own a new wave of internal software. Many are not seriously discussing how they plan to test it, support it, secure it, document it, and maintain it once the prototype becomes part of the business.

Everyone can see that the sticker price of building has dropped. But has the total cost of ownership really even moved?

The wheel got cheap to reinvent. The road you maintain forever did not.

The internet talking heads will happily grant you permission to build everything and anything. “Build everything now, AI makes it free” is a fantasy that ignores the maintenance tail.

On the other side, luddite executives are still treating AI as a fad. Their archaic reflex mantra, “nobody ever got fired for buying IBM,” is quietly putting companies behind their competitors and erecting imaginary ceilings.

Both reactions avoid the arithmetic.

The point isn’t to build more or buy more — it’s that the arithmetic has changed, and it’s still changing rapidly. The variables you locked in two years ago aren’t even close to the current reality.

So rerun them.

Take the three or four decisions you consider settled: the workflow you bought because building was unthinkable, the platform you standardized on because integration was cheaper than authorship. Recompute them at today’s build cost.

Some will hold. A few will not.

Those are where the advantage is hiding, because your competitors will be the ones working off the old quote.