Apps may fade into the background. Software won’t.
Wall Street has a new consensus: AI is about to kill the software industry. Software stocks are down nearly 30% since the start of the year; pundits call it the SaaSpocalypse. But the story is wrong. AI is rewriting who builds software and how we pay for it—not eliminating it. This piece looks at why the real moats (data, workflows, habits) are getting deeper, how “Software as a Service” is turning into “Service as Software,” and what that means for builders and buyers.
Two 19-year-old high schoolers built an AI calorie-tracking app called Cal AI that brought in over $30 million a year; it was recently acquired by MyFitnessPal. The deal size was not disclosed, but the two clearly came out on top.
On another front, Cursor, the fastest-growing AI coding company in history and less than five years old, was reported in February to have passed $2 billion in annualized revenue.
Whether we talk about AI companies that build apps or the AI-powered apps already in the world, the outlook seems bright.
At the same time, Wall Street is in a panic. The consensus there: AI is about to kill the software industry.
When a model can generate a polished backend with one click, and every employee can summon internal tools in plain language, who will keep paying for heavy enterprise SaaS?
Since the start of the year, many household-name software companies have seen their shares drop by almost 30%. Some call it the “SaaSpocalypse.” The story is simple: AI has made writing code cheap; anyone can spin up an app in a sentence, so why would software companies be worth anything?
What AI Actually Changed: Everyone Can Be an Architect
Cursor’s CEO, Michael Truell, posted a report on X titled “The Third Era of AI Software Development.” One detail stuck out: more than a third of the code at Cursor is now written entirely by AI.
We are not talking about AI-assisted coding. The AI runs on its own in the cloud: it tests, iterates, and hands engineers a review-ready result. The engineer’s job is to break work into tasks, assign them to multiple AIs, and review the output.
In Cursor’s own metaphor, the developer is shifting from “worker operating the machine” to “factory manager.”
Cursor’s main job is no longer writing code but helping developers build a software factory. That factory is made of many agents; developers work with them as teammates, give initial direction, give them the tools they need to work independently, and review what they produce.
This pattern is catching on inside and outside the industry. When Claude Code shipped Claude Cowork earlier this year, Boris Cherny, who leads Claude Code, replied on X that all of the code was written with AI assistance.
Boris Cherny said that building Claude Cowork was not zero human involvement: they still had to plan, design, and go back and forth with Claude, but Claude wrote all the code.
That’s how two high schoolers could build a product earning $30M a year, and why Cursor’s annualized revenue keeps climbing.
Five years ago, building a profitable app usually meant a team of at least ten, several funding rounds, and two or three years of development. Now, two people with AI tools can do it in months.
AI is clearly lowering the bar and the cost of “making software.” What it does not lower is everything else that makes software actually useful.
A Software Company’s Value Is Not in Its Code
A lower bar does not mean incumbents fall off the mountain.
Take Salesforce or Jira (the agile project management tool that countless developers and PMs love to hate). AI has been around for a while. Aren’t there sleeker, lighter, smoother alternatives? Plenty.
So why do teams still pay up while they complain? Because those customer lists, years of sales follow-up, and tangled bug history live inside the old guard’s servers and are hard to leave.
The deepest moats for these enterprise tools are strong historical data and tight coupling to how the business runs. They have become part of the company’s operations.
In 2024, the founder and CEO of fintech startup Klarna decided to drop Salesforce’s flagship CRM after ChatGPT appeared and moved to an in-house AI system. A year later he posted on X that he didn’t think others would or should copy him. “I don’t think this is the end of Salesforce; maybe the opposite.”
So no, we did not replace SaaS with an LLM, and storing CRM data in an LLM would have its limitations. But we developed an internal tech stack, using Neo4j and other things, to start bringing data and knowledge together.
Aneel Bhusri, co-founder of Workday (finance and HR software), recently returned as CEO. On the so-called SaaS apocalypse he gave a sharp, sober take: “However powerful AI gets, it’s still probabilistic at the core.”
And the office platforms have moats too. Switching means moving all historical files, approval flows, and attendance records, and reconnecting to finance, HR, and supplier systems.
That cost has nothing to do with how good the code is.
At the same time, AI is giving another slice of software a stronger engine. Products that are embedded in user workflows, have accumulated data, and benefit from network effects are using AI to do things they couldn’t do before, serve people they couldn’t serve, and enter markets they couldn’t reach.
The office apps we use every day are baking AI into every feature: AI meeting notes, AI search over company knowledge, AI-generated spreadsheets. They are not being replaced by AI; they are using AI to become harder to replace.
In this wave, startups may write lighter code with AI, but the incumbents with the deepest moats still hold what’s scarcest for training super AI agents: huge amounts of proprietary data and real user workflows. AI is making the software industry’s winner-take-all dynamics stronger than ever.
Oracle’s CEO Mike Sicilia has said he does think that if they don’t adopt AI tools and their coding power, that would be a threat—but they are adopting, and fast.
The real value of software lives there. Code is just the ticket in. Process, data, ecosystem, habit, and the deep dependence between the product and how users work do not vanish because AI made code cheaper. Better AI will deepen those moats.
Software as a Service, Service as Software
In 2011, Marc Andreessen said “software is eating the world.” In 2017, Jensen Huang said software is eating the world … but AI is eating software.
This year, at the Cisco AI Summit, Huang said AI will not replace software companies and called the idea “the most illogical thing in the world.” He said the point of AI is not to reinvent tools but to use them.
He later wrote on NVIDIA’s official account about his five-layer AI stack: energy → chips → infrastructure → models → applications. The core idea: AI is less like traditional shrink-wrapped software and more like an industrial product that generates intelligence in real time.
The industry has a term for the last twenty years’ dominant model: SaaS, Software as a Service. A company sells a tool; you (or people you hire) use it; you pay per seat.
That model is being displaced by something some call Service as Software.
Before: a vendor sold you a shovel and charged per shovel; you dug for gold. Now: a vendor digs the gold for you and charges per result. Decagon, which does AI customer support, doesn’t charge per seat but per conversation and per resolved issue.
OpenClaw fits this shift. AI is not attacking software itself but the “per-seat” business model. The death of that model is pushing a better one into place.
For users, we pay for an outcome, not for a tool. Companies no longer need to hire, train, and manage as many people; they tell the new service provider how many support issues they want solved and pay for that.
For software companies, they are moving from tech vendor to outsourced service provider: from giving us a tool to doing the work for us.
This shift is spreading: law, accounting, healthcare, sales, HR. Every domain that used to need a lot of manual work is seeing new software companies that charge by result.
Going from selling tools to selling services and charging by result sounds like the ideal end state. In heavily regulated areas (healthcare, finance, government), long procurement cycles, strict data-privacy review, and the question of who is liable when AI errs will slow the new model well behind the technology.
So even though the code layer has already been disrupted, the business layer will take time. Traditional SaaS incumbents will not collapse overnight. As the model shifts, total software spending will grow, not shrink.
Gartner (February 2026) projects global software spend to grow 14.7% this year, above $1.4 trillion. Forrester projects the global SaaS market to grow from $318 billion in 2025 to $576 billion in 2029. Those are not the numbers of a dying industry.
Maybe one day we really won’t need to open app after app: we’ll say one thing to an AI and it will call a dozen tools in the background. As Karpathy has put it, we might stop browsing and clicking and instead have a single agent do it all for us.
On the surface, apps disappear and UIs vanish. But software does not die; it moves deeper. The primary user of software shifts from humans to AI.
For the first time in twenty years, the software business is being forced to think clearly about the endgame under the pressure of AI agents.
Software is being rewritten. This time, not for humans, but for AI.
TL;DR: AI is making code cheaper and shifting the business from “per seat” to “per outcome,” but the value of software still lives in data, workflows, and lock-in. Incumbents with deep moats are doubling down on AI; the industry is growing, not dying. The shift to watch is API-first design and software built for AI agents, not just for people.