Enterprise Software Is Dead

Enterprise Software Is Dead

How AI turned a $700 billion industry into a rounding error

Last week, sixteen instances of Claude Opus 4.6 sat down and built a C compiler from scratch.

Not a toy. Not a demo. A Rust-based C compiler that clears 99% of the GCC torture test suite, compiles the Linux kernel across three architectures, and can even run Doom. Nearly 100,000 lines of code, generated autonomously over two weeks, across 2,000 independent sessions. Cost: about $20,000 in API credits.

Let that sink in for a second.

A C compiler is one of the hardest things in all of computer science. It's the kind of project that used to take teams of brilliant engineers years to ship. It's a flex of deep systems knowledge, architecture design, and painstaking debugging. And an AI just did it, autonomously, for the price of a used Honda Civic.

Now ask yourself: if AI can build a C compiler, what exactly is stopping it from building your company's expense report tool? Or your CRM workflow? Or your "enterprise-grade" project management suite that costs $45 per seat per month and still can't do what you actually need it to do?

Nothing. The answer is nothing.

And the market knows it. Enterprise software stocks are cratering. The iShares Expanded Tech-Software ETF entered bear-market territory in early 2026, marking one of the steepest drawdowns since the 2008 financial crisis. ServiceNow, the darling of enterprise SaaS, is down over 25% year-to-date despite posting 21% subscription revenue growth. SAP plunged 16% when analysts realized their cloud backlog couldn't keep pace with expectations. Salesforce, Oracle, Workday — all feeling the gravitational pull of a tectonic shift.

The fascinating part? Their current earnings are fine. Revenue is growing. Margins are healthy. But the stock market doesn't price the present. It prices the future. And the future of enterprise software, as we've known it, is over.

I — The $700 Billion Middleman

The global enterprise software market is worth roughly $700 billion in 2026, projected to hit $1.28 trillion by 2031. SaaS alone accounts for about $375 billion. These are staggering numbers. They represent one of the greatest wealth-creation engines in the history of technology.

But here's the thing nobody wants to say out loud: most enterprise software is a middleman.

Think about what Salesforce actually does. It sits between your salespeople and your customer data. It takes information your team already has, stores it in a database your team could build, and presents it back through an interface your team didn't design. For the privilege, you pay $75 to $300 per user per month. Multiply that across a 5,000-person org and you're staring at millions in annual licensing fees — before implementation costs, consultants, and the army of admins you need to keep the thing running.

And Salesforce is one of the good ones.

Enterprise software has been protected by a moat so deep most people never questioned it: complexity. The argument was always that building custom software is too hard, too expensive, and too risky. Better to buy the off-the-shelf solution that "just works" (it never just works, but that's another essay). Better to pay the vendor who's already solved the hard problems.

That argument made perfect sense when software was hard to build. When you needed teams of engineers, months of planning, and millions in development costs to ship anything meaningful. The enterprise software industry was built on the assumption that writing code is expensive.

II — The Inversion

Here's what changed, and it's not incremental. It's an inversion of the fundamental economics of software.

In February 2025, Andrej Karpathy coined the term "vibe coding" — the practice of telling an AI what you want in plain language and having it generate the code. At the time, it felt like a novelty. Cute demos on Twitter. Hackers building toy apps over a weekend.

Twelve months later, Claude Sonnet 5 hits 82.1% on SWE-Bench, the gold-standard benchmark for real-world software engineering. That means it can autonomously resolve over four out of five genuine GitHub issues — the same kind of bugs and feature requests that fill the backlogs of every software company on the planet. Claude Code's revenue hit a $1 billion run rate just six months after launch. Uber, Netflix, Spotify, and Salesforce itself are all using it.

But the real shift isn't that professional engineers now have better tools. It's that everyone now has the tools.

People with zero programming experience are deploying real applications. Not prototypes. Not demos. Working software that books theater tickets, files taxes, and monitors agricultural systems. Anthropic launched Cowork in January 2026 — essentially Claude Code for people who don't write code. You give it a folder, describe what you want, and walk away. It plans. It executes. It ships.

The implications of this are hard to overstate. Software development used to be a bottleneck controlled by a priesthood of engineers. Now it's becoming something closer to literacy — a basic capability that anyone can access. And when everyone can build software, the value of pre-built software collapses.

III — Why the Stocks Are Crashing (It's Not About Earnings)

This is the part that confuses a lot of people. If enterprise software companies are still posting strong revenue growth, why are their stocks getting destroyed?

Because the market is pricing in a structural collapse of the business model, not a cyclical downturn.

Enterprise software has historically been one of the stickiest businesses in existence. Enterprise churn rates hover around 3-5% annually. Average contract lengths are 24 months. Switching costs are astronomical — you don't rip out your ERP system on a whim. Customer acquisition costs are brutal (industry average: $702 per customer) but lifetime values are enormous because nobody ever leaves.

This stickiness created a beautiful flywheel: high margins (70-90% gross), low churn, predictable revenue, rich valuations. SaaS companies became the blue chips of the tech world. Investors loved the recurring revenue. Founders loved the multiples.

But stickiness is not the same thing as value. Customers weren't staying because the software was irreplaceable. They were staying because the cost of replacing it was too high. The moat wasn't quality. It was friction.

AI just eliminated the friction.

When you can describe what you need and have an AI build it in hours instead of months, the calculus changes completely. The enterprise software industry built a $700 billion castle on the foundation of "building software is hard." And now it isn't. Not for the basic stuff. Not for the internal tools. Not for the dashboards and workflows and integrations that make up 80% of what enterprises actually pay for.

Gartner projects that enterprise software spending will grow 15.2% in 2026. Sounds great, right? Except they buried the lead: over 50% of that growth comes from price increases, not from new customers or expanded usage. The vendors are squeezing existing customers harder because new logos are getting harder to win. That's not growth. That's extraction on a ticking clock.

IV — The Klarna Prophecy (And Its Plot Twist)

If you want to see this story in miniature, look at Klarna.

In 2024, Klarna's CEO Sebastian Siemiatkowski made headlines by announcing they were abandoning Salesforce and Workday to build homegrown AI replacements. Their AI customer service bot replaced 700 full-time contract workers and saved $40 million annually. It was the loudest signal yet that big companies were ready to ditch enterprise software for AI-built alternatives.

Then came the plot twist.

Klarna didn't actually replace Salesforce with pure AI. They replaced it with a patchwork of smaller SaaS tools and custom-built solutions. Siemiatkowski later admitted that the future might not be the end of Salesforce at all. And by early 2026, Klarna was publicly acknowledging their aggressive automation approach had "gone too far" and started bringing humans back into the business.

But here's why this story still proves the thesis, even with its complications.

Klarna didn't go back to the old model. They didn't re-sign with Salesforce at the old price. They built a hybrid — smaller, cheaper tools augmented by AI. The total cost of their software stack went down dramatically. The capabilities went up. They proved that even when pure AI replacement doesn't work, AI-augmented alternatives are cheaper and more flexible than the enterprise incumbents.

And Klarna is just one company. Andreessen Horowitz calls this the "tip of the iceberg." Their data shows 44% of enterprises now using Anthropic in production, up 25% since May 2025. This isn't pilot programs anymore. AI has graduated to a recurring line item in core IT budgets. Companies are reconsidering their entire tech stacks.

VI — The Infrastructure Shift Tells the Whole Story

Follow the money.

In 2025, spending on AI infrastructure was about $60 billion. In 2026, it's projected to hit $230 billion. That's nearly a 4x increase in a single year. This is the fastest reallocation of technology spending in modern history.

And here's the telling part: that capital is going to AI hardware and infrastructure, not to traditional software companies. Semiconductor companies are surging while software companies are sinking. The market is making a clear bet: the future of enterprise technology is AI infrastructure, not SaaS applications.

This makes intuitive sense when you think about it. If AI can build any application on demand, the scarce resource isn't the application — it's the compute that powers the AI. The value shifts from the software layer to the intelligence layer. From the app to the model. From Salesforce to Anthropic, OpenAI, and the chip companies that power them.

It's the same pattern we've seen in every technology transition. When electricity replaced steam power, the value didn't go to the companies that made better steam engines. It went to the companies that built the electrical grid. When cloud computing replaced on-premise servers, the value didn't go to Dell and HP. It went to AWS and Azure. Now, as AI replaces custom-built and off-the-shelf software, the value is moving to the AI platform layer.

VII — What Dies and What Survives

I should be precise about what I mean when I say enterprise software is dead. I don't mean every enterprise software company goes to zero tomorrow. Some will adapt. Some will thrive. But the model - the business of selling pre-built, one-size-fits-all software to enterprises at $50+ per seat per month — is in structural decline.

Here's what I think dies:

Commodity workflow tools. Project management, basic CRM, expense reporting, simple analytics dashboards. Anything that's essentially a database with a UI on top. These are the first to get replaced because they're the easiest to rebuild with AI.

The implementation-industrial complex. The army of consultants, integrators, and admins that exist to configure, customize, and maintain enterprise software. When AI can configure itself to your needs, the multi-billion dollar consulting ecosystem built around enterprise software loses its reason for existing.

Per-seat pricing. When one person with AI can do the work that previously required ten seats of enterprise software, the per-seat model becomes absurd. Expect a brutal transition to outcome-based and usage-based pricing models.

Here's what survives:

Systems of record with network effects. If your value is the data, not the interface — and if that data becomes more valuable with more participants — you have a defensible position. Bloomberg terminals, certain financial data platforms, industry-specific compliance databases.

Deeply regulated verticals. Healthcare, financial services, government. Where the value isn't the software itself but the certification, compliance infrastructure, and audit trail that comes with it. AI can build the tool, but it can't certify it for HIPAA compliance overnight.

AI-native platforms. The enterprise software companies that successfully pivot to become AI platforms rather than application vendors. Instead of selling you a CRM, they sell you an AI that builds and continuously optimizes your CRM. It's a fundamentally different business model, and most incumbents will struggle to make the transition.

VIII — The Builder's Era

We are entering what I'd call the Builder's Era of software.

For the last two decades, the enterprise software industry operated on an implicit assumption: most companies are consumers of software, not builders of it. You buy Salesforce because you're a retailer, not a software company. You buy Workday because you're a bank, not a technology firm. The division was clean. Software companies build. Everyone else buys.

AI obliterated that division.

Now every company can be a builder. Every department can be a builder. Every individual contributor with a laptop and a Claude subscription can be a builder. The barrier to entry for creating functional, useful software collapsed from years and millions of dollars to hours and a couple hundred bucks a month.

I'd recommend watching this podcast episode, Gokul explains it well. He talks about what his portfolio companies are doing and what he's predicting with AI. He also discussed how he benchmarked building a transcription app with AI tools six months ago versus now.

This doesn't mean enterprise software disappears overnight. Gartner is right to note that traditional enterprise applications may be more resilient than the most aggressive predictions suggest. Legacy systems have deep roots. Institutional inertia is real. Procurement departments don't move fast.

But the direction is unmistakable. The same way smartphones didn't kill desktop computers overnight but fundamentally changed what a computer was for, AI isn't going to kill enterprise software overnight. It's going to fundamentally change what enterprise software is.

It stops being a product you buy. It becomes a capability you have.

The companies that understand this will build the future. The companies that don't will become the cautionary tales. And the individuals who learn to wield these tools — who understand that the ability to describe what you need is becoming the ability to build what you need — will have an almost unfair advantage in whatever industry they touch.

Sixteen AI agents just built a C compiler for twenty grand. Your company's bloated, overpriced, under-delivering enterprise software stack doesn't stand a chance.

Anthropics Article on the C Compiler

— Dakshay