The Inversion

For twenty years, the sermon was the same: don't build it, buy it. Every conference keynote, every venture pitch, every enterprise sales deck carried the same gospel. Why would you build your own CRM when Salesforce exists? Why maintain your own email infrastructure when SendGrid is right there? Why build analytics when Amplitude has a team of two hundred engineers working on it full time? The logic was ironclad. The economics were obvious. Build vs. buy was a settled question, and buy had won.

Until it hadn't.

What we are witnessing now is not a correction or a cycle. It is the inversion of a fundamental assumption that undergirded a trillion-dollar industry. The assumption was this: software is hard to build, and the difficulty of building it creates a durable moat. SaaS companies did not just sell features. They sold the accumulated difficulty of engineering — the thousands of person-hours embedded in every dropdown menu, every integration, every edge case handled at 2 AM by an on-call engineer who would rather have been sleeping. You were not paying for the software. You were paying for the pain of having built it.

When the cost of creation approaches zero, every business model predicated on the difficulty of creation is living on borrowed time.

AI has not yet made software free to build. But it has made it cheap enough to build that the calculus has flipped for an enormous number of use cases. A competent engineer with Claude or GPT-4 can now build in a weekend what would have taken a team of five engineers a quarter to produce three years ago. The code is not always elegant. The architecture is not always pristine. But it works, it is yours, and you do not owe anyone $48 per seat per month for the privilege of using it.

This is the part that SaaS executives do not want to discuss in their earnings calls. The moat was never the product. The moat was the difficulty of replication. And that difficulty just dropped by an order of magnitude.

A Taxonomy of the Doomed

Not all SaaS is equally vulnerable. The key variable is what I call complexity density — the ratio of genuine, domain-specific complexity to generic software engineering. A SaaS product with high complexity density solves problems that are inherently hard, regardless of how easy the code is to write. A product with low complexity density is essentially a CRUD app wearing a tuxedo.

The pattern is clear. The products most at risk are those where the primary value proposition is we built a nice interface for something you could have built yourself, if you had the time and the engineers. AI does not give you more time, but it radically reduces the number of engineers required. When a single developer can scaffold a custom analytics dashboard in an afternoon — complete with the exact metrics their business needs, integrated with their specific data sources, without the compromises inherent in any general-purpose tool — why would they pay Amplitude $36,000 a year?

The total cost of ownership calculation has been disrupted. For decades, the formula looked like this:

And almost always, $TCO_{build} > TCO_{buy}$, because $C_{dev}$ was enormous. A team of four engineers at $200K fully loaded, working for six months, easily ran to $400K — and that was before maintenance. Against that, even an expensive SaaS license looked like a bargain.

But AI compresses $C_{dev}$ dramatically. An engineer who previously produced $x$ units of working software per month now produces $5x$ or $10x$. The build side of the equation is collapsing:

where $\alpha$ represents the AI productivity multiplier (currently somewhere between 3 and 10, depending on the domain), and the asterisked terms are also shrinking because AI assists with maintenance and reduces opportunity cost through speed.

The Break-Even Horizon

The critical question for any SaaS company is: when does the break-even point shift? At what moment does it become cheaper for a median customer to build their own solution than to keep paying the subscription?

The break-even condition is straightforward:

For a SaaS product charging $50K/year in enterprise licensing, with an AI-augmented build cost of $30K and annual maintenance of $10K, the break-even hits in year one. The customer builds it, owns it, customizes it freely, and is cash-positive by month eight.

The SaaS industry's great innovation was turning software into a recurring revenue stream. Its great vulnerability is that the stream can be dammed by anyone with a text editor and an API key.

This does not mean all SaaS dies. It means SaaS that sells the technology dies. The survivors will be those who sell something that cannot be replicated by writing code: proprietary data, network effects, regulatory compliance, deep domain expertise that is not captured in training data. Salesforce survives not because its software is hard to build — it isn't, and thousands of companies have proven it — but because its data ecosystem is irreplaceable. Veeva survives because the cost of being wrong in pharmaceutical compliance is measured in prison sentences, not downtime.

The New Landscape

What emerges on the other side is not the death of software businesses. It is the death of software businesses whose only moat is the software. The companies that thrive will be those that recognized, perhaps years ago, that the code was never the point. The code was just the delivery mechanism for something deeper — a dataset, a workflow, a community, a standard.

The rest — the beautifully designed CRUD apps, the project management tools with slightly different philosophies about what a task is, the analytics platforms that all show you roughly the same numbers in roughly the same charts — are marked for death. Not by a competitor. Not by a disruption in the traditional Christensen sense. But by the simple, devastating fact that their customers can now build the same thing over a long weekend.

The SaaS revolution promised to democratize enterprise software. AI is completing that promise in a way the SaaS companies never intended. The software is finally, truly, for everyone. It is just no longer from anyone in particular.

This is the great irony of the moment: the technology industry built tools so powerful that they eliminated the need for the technology industry's most profitable business model. The priests automated themselves out of the temple. And the congregation, blinking in the sudden light, is discovering that they can perform the rituals on their own.