I told companies not to build software. Then building got cheap.

In May 2020 I published a piece with a clear conclusion: unless software is your technological base, do not build it, buy SaaS. I was a SaaS founder, and I believed it, and most of it was right. Reading it back six years later, the framework holds. The reason behind the advice does not.

The framework was this: every company has a technological base, the part that is genuinely its core, and most companies confuse a tool that supports the operation with the core itself. A freight company's core is moving things, not the parcel-tracking app. A bank's core is pricing risk, not the mobile channel. Get that wrong and you pour money into building software that was never going to differentiate you, while a specialist out-executes you on the thing you should have bought. So: know your base, build only there, buy everything else. The original is still up, unedited.

That logic rested on one assumption I never stated, because in 2020 it did not need stating: building software is expensive and slow. Every argument against building leaned on it. Do not reinvent Salesforce, because they have years of investment, economies of scale, and a trained team, and you cannot catch up. Buy, because development time is the opportunity cost that kills you. The whole case for "buy" was really a case about the cost of "build."

That assumption is the one that broke.

The reason inverted, not the rule

I now build platforms with AI as a co-architect, and I measure the result straight from the git history. A security and fraud platform that reaches the same complexity as a system I once built over years, close to twenty thousand worked hours, now takes little more than a hundred. I show the numbers and the method here. The point is not the brag. The point is that every reason I gave in 2020 for not building was a function of building being expensive, and building stopped being expensive.

Years of investment as a moat? Compressed. Economies of scale on development cost? They shrink when one person and an agent produce what a team used to. A trained team you cannot match? The leverage moved from the size of the team to the judgment of the person directing it. The Salesforce argument, the centerpiece of my 2020 piece, is the clearest casualty. In 2020, building your own CRM was value destruction. In 2026, building the exact system your business actually runs on, owned and governed, in weeks instead of years, is not obviously wrong anymore. Sometimes it is the better move.

So "buy by default" is gone. Not because buy is bad, but because the thing that made it the safe answer, the prohibitive cost of build, is gone.

What buy still wins, so this is not hype

I want to be precise, because the lazy version of this argument is "AI means build everything," and that is wrong and expensive in a new way.

SaaS still wins wherever the value is shared rather than yours. Payment rails, identity verification, email deliverability, the commodity layers where a network of customers makes the product better and no single buyer should own it. It wins on compliance you do not want to carry, and on the long tail of operational tooling that will never be your differentiator no matter how cheap it is to build. Cheap to build does not mean worth owning. A thing you build is a thing you now maintain, secure, and are accountable for, forever. That cost did not fall as fast as the cost of the first version did.

What actually changed is not the answer. It is the size of the set worth owning. When building was expensive, that set was tiny, basically just your technological base. Now that building is cheap, the set is larger, and its edges are blurry, and deciding where they fall is the entire game.

The new question

In 2020 the question was a cost question: can we afford to build this, or should we buy it? The honest answer was almost always buy.

In 2026 the question is a judgment question: what deserves to be owned, now that owning is cheap? Cost no longer decides it for you. You have to decide it, on the merits, every time. Which capabilities are so close to your core that owning them compounds, and which are commodities you would be insane to babysit. That decision used to be made for you by your budget. Now it is made by you, and it is exposed, because the excuse is gone.

This is the same shift I keep landing on from every direction. When building is cheap, the constraint moves off the keyboard and onto the thinking: why to build a thing, what to build, and what to be accountable for owning. That judgment is the moat now, and the discipline that protects it is recording it, deciding on purpose and leaving a trail, so that cheap building does not just become fast mess.

My 2020 self asked companies a good question: is software your technological base? I would keep asking it. I would just add the 2026 version next to it: now that you can build almost anything, what are you willing to own?

Same instinct. Inverted economics. A better question than the one I had the answer to six years ago.