More Software, Fewer Software Businesses

Static software gets commodified. Dynamic intelligence doesn't.

There's a tweet going around: "Software engineering is like Tetris now in that you have to go faster and faster until you die."

It's funny because it's true. But it's also wrong about what game you're playing.

The Tetris framing assumes the game stays the same — you're still building software, just faster. Stack the blocks quicker or lose. But that's not what's happening. The game itself is changing, and the people who see it as Tetris are about to discover they've been practicing for the wrong sport.

The AI Twitter crowd is having the wrong conversation. They're excited about acceleration. "I built a SaaS in a weekend!" "Vibe coding is the future!" But acceleration of what? Of building static software. They're using a fundamentally new capability to do the old thing faster.

This is what happens in every disruption. The first instinct is to speed up the existing paradigm, not to see the new one. The "I shipped a SaaS in a weekend" crowd is actually proving the opposite of what they think. If anyone can ship a SaaS in a weekend, then SaaS has no moat. They're celebrating the very thing that destroys the value of what they're building.

It's like printing press operators getting excited about how fast they can copy manuscripts while missing that the game just became what you publish, not how fast you can copy.

Here's what's actually happening: AI coding tools are commodifying static software.

A church management system is fundamentally "arrange data in a database and show it in a UI." That's the first thing AI coding tools commodify. Not tomorrow — but the $99 lifetime deal that's differentiated today might be competing against "describe what you need and get a custom tool" in two years.

The small vertical SaaS playbook — find underserved niche, build adequate tool, charge monthly, enjoy 80% margins — every step of that is getting undermined. The margins existed because building was hard and switching was expensive. When building is trivial and switching means "tell the AI what you want," both of those evaporate.

You end up in a commodity market competing on price toward zero.

And notice the irony: small price-sensitive organizations are both the classic target market for vertical SaaS and the market most likely to be served by a generic AI builder first. A church treasurer who can say "I need to track 80 members and their donations" to Claude and get something functional? That person was never going to pay $99 for your thing. The ones who would pay are the ones who can't articulate their needs clearly enough to prompt an AI — and that population shrinks every year as the tools get better.

This is the distinction that matters:

An AI coding agent builds you an app. The app is code. It runs deterministically. It's a thing that was made once and then sits there. That entire category — "thing that was made once" — is what's becoming free.

But there's another category: running intelligence. Something that requires real-time reasoning about this specific situation. The value isn't in the code, it's in the inference. No coding agent can "build you" one of these because it isn't a build — it's an ongoing process. It's closer to an employee than a product.

Static software gets commodified. Dynamic intelligence doesn't.

The unit of value in software is shifting from the artifact (code) to the intelligence (ongoing reasoning). Code becomes commodity. Intelligence becomes product. The people who win aren't the fastest builders — they're the best designers of minds.

This maps to a completely different business model. You're not selling software. You're selling labor — AI labor that's 100x cheaper than a human but delivers 80% of the value.

A bookkeeper/scheduler costs $35-50K a year. An AI that delivers the useful subset of that could cost $50-100 a month. That's not a SaaS pitch. It's a staffing pitch. And staffing doesn't get commodified by coding agents because it's not code.

The cost structure actually protects you. Real marginal cost (every interaction is inference) means "just spin up a competitor" isn't free the way it is with static software. And the value you're delivering is priced against human labor, not against other software. That ceiling is much higher.

Someone at a field service management company is going to say "we need AI" and they'll bolt a chatbot onto their dashboard. Because they're thinking in software. The chat becomes another feature of the product, not the product itself.

They literally can't conceive of removing the dashboard because the dashboard is what makes it "software" in their minds. The conversation as the entire interface — no dashboard at all — is counterintuitive enough that most competitors won't even try it.

But that's exactly where the insight lives. You don't manage an employee via a dashboard. You talk to them. The interface and the economic model aren't two separate decisions. They're the same insight.

So what's the new game?

Building software is engineering. Crafting an intelligence is something else — closer to character design, or pedagogy. It's understanding what it means for an AI to know someone's business, to have the right judgment about when to ask versus infer, to feel like someone you trust.

That's a genuinely different skill. And almost nobody has it yet because almost nobody has been practicing it. The R&D required isn't technical — it's relational. It's spending years in genuine dialogue with AI and understanding something about what makes it feel real.

There will be more software than ever. But the era of software businesses as the ultimate high-margin business will end.

The thesis, stated plainly: the business moves up the stack to the intelligence layer. More software, fewer software businesses.

Craft intelligences, not software.