Software Isn’t Dying. It’s Being Demoted.

On January 28, 2026, Anthropic released a set of open-source plugins for Claude Cowork targeting legal, sales, marketing, and data analysis workflows. It wasn’t a flashy keynote. It was a quiet Tuesday product update. Within 72 hours, it became the catalyst for the largest AI-driven stock selloff in market history.

The S&P North American software index dropped 15% in January alone—its worst monthly decline since October 2008. By February 11, roughly $2 trillion in market capitalization had been erased from the software sector. The iShares Tech-Software ETF (IGV) retreated 30% from its late-2025 highs. It wasn’t a correction. It was a repricing.

Then OpenAI piled on. On February 5, it launched Frontier—an enterprise platform that doesn’t just offer AI tools but treats agents as digital employees with onboarding, permissions, identities, and performance reviews. Uber, Intuit, and State Farm were already live customers. OpenAI’s applications CEO Fidji Simo framed the ambition plainly: Frontier is meant to be the intelligence layer that sits on top of every enterprise system.

The mood on trading desks was visceral. James St. Aubin, chief investment officer at Ocean Park Asset Management, captured it: the seemingly wide moats of established software companies suddenly feel much narrower as AI-created products intensify. One Jefferies analyst summed up the client calls more bluntly: when asked for their “hold-your-nose” price level, nobody had an answer. They were just selling everything.

The combined message from the two biggest AI labs was unmistakable: we’re not just building models anymore. We’re building the software that replaces your software.

The Real Thesis: It’s Not Replacement. It’s Demotion.

Most coverage has framed this as “AI vs. software.” That’s the wrong lens, and it’s leading to bad decisions.

Software isn’t being replaced. It’s being pushed down the stack.

Think of it like the plumbing in your house. You need pipes. You’ll always need pipes. But nobody chooses their home based on the brand of copper tubing behind the walls. The value—and the pricing power—lives at the interface layer: the kitchen, the bathroom, the thermostat you actually touch. That’s the layer AI agents are claiming.

Here’s the shift in three stages:

Stage 1: Software as Product (2000–2024)

You buy Salesforce. You log into Salesforce. You interact with Salesforce. The interface is the product. You pay per seat. SaaS companies own the customer relationship.

Stage 2: Software as Infrastructure (2025–2026)

AI agents log into Salesforce for you. You interact with the AI layer—Claude, GPT, Gemini—which uses Salesforce as a backend data source. The interface shifts to the agent. The SaaS product becomes plumbing. The AI lab starts to own the relationship.

Stage 3: Software as Commodity (2027+)

If the AI agent can use Salesforce, it can also use HubSpot, or a custom database, or nothing at all. When the user never touches the underlying tool, switching costs collapse. The software becomes interchangeable. Pricing power evaporates.

This is why Goldman Sachs drew the parallel to newspapers. Newspapers didn’t die because people stopped wanting news. They died because Google and Facebook became the interface through which people consumed news—and captured the relationship, the attention, and the ad revenue. The content producers became commodity suppliers. Goldman analyst Ben Snider warned clients this may represent a long-term structural decline, comparing the trajectory to how newspaper stocks only stabilized once earnings estimates finally bottomed. For software, that bottom hasn’t arrived yet.

The Seat-License Time Bomb

The business model problem is arguably more urgent than the technology problem.

The entire SaaS industry runs on per-seat pricing. If your company has 200 salespeople, you buy 200 Salesforce licenses. But if one AI agent can handle the prospecting, CRM updates, and follow-up emails that 10 salespeople used to do manually, your company doesn’t need 200 seats anymore. Maybe it needs 50—plus five agents.

Fortune’s Jeremy Kahn nailed the analogy: the SaaS industry has long operated like a gym—your best customers are the ones paying for memberships they don’t fully use. AI agents expose this gap by actually using every feature, all the time, at machine speed. When that happens, the “over-provisioned seats” revenue disappears.

The incumbents see it coming. Here’s how the biggest players are responding:

The Bull Case: What the Panic Is Getting Wrong

Before you dump every software stock in your portfolio, here’s the counter-argument—and it has teeth.

JPMorgan’s Dubravko Lakos-Bujas told clients the market is pricing in worst-case disruption scenarios that are unlikely to materialize in the next three to six months. His reasoning: enterprise software is deeply embedded across corporate infrastructure, protected by multi-year contracts and switching costs that provide a significant buffer against near-term displacement. He argues that AI is more likely to be additive to software workflows in the near term—not a substitute.

The data backs him up on timing. Thomson Reuters, despite getting hammered 28% year-to-date, reported Q4 numbers on February 5 and guided for roughly 8% organic revenue growth in 2026 alongside margin improvement. Revenue is still growing. Earnings aren’t collapsing. The stocks are falling on future fear, not present pain.

Goldman Sachs CEO David Solomon himself called the selloff “too broad”—even as his own strategists warned of long-term structural risk. Morningstar’s analysts added that European tech is now one of the cheapest sectors on the continent, trading at a 6% discount to fair value. The Sycomore Sustainable Tech fund, which has beaten 99% of its peers over three years, was buying Microsoft shares during the rout.

The honest take: The bears are right about the direction but probably wrong about the speed. Enterprise software contracts don’t unwind in quarters—they unwind over years. The companies that use this breathing room to build agent layers will survive. The ones that assume multi-year contracts will protect them forever are the newspapers of this analogy.

Who Wins, Who Survives, Who’s Exposed

Not all software companies face the same risk. The key variable isn’t AI capability—it’s data moats and switching costs.

Likely Winners

Infrastructure and hardware: NVIDIA, TSMC, and the hyperscalers (AWS, Azure, GCP) benefit regardless of which layer wins. AI agents need compute. Every scenario requires more of it. Hyperscaler capex guidance for 2026 is up 24% year-over-year—$117 billion more than last year, per Wells Fargo. That money flows directly into the infrastructure layer.

Cybersecurity: AI agents create new attack surfaces nobody has patched yet. The ServiceNow “BodySnatcher” vulnerability—disclosed by AppOmni in January—demonstrated how attackers can hijack AI agents to move laterally through enterprise systems, weaponizing the tools meant to simplify workflows. ServiceNow’s Now Assist and Virtual Agent products are used by nearly half of AppOmni’s Fortune 100 customers. More agents = more attack vectors = more security spend. Allianz’s 2026 Risk Barometer ranked AI as the #2 global business risk, jumping eight spots in a single year.

Likely Survivors (But Not Unscathed)

Data-moat companies: Thomson Reuters and RELX took enormous hits, but their legal and research databases represent decades of accumulated, proprietary, paid-access data. AI agents can’t replicate what doesn’t exist in the open web. Morningstar’s analysts argued that switching costs remain the critical moat—and both companies still have wide ones. Thomson Reuters’ Q4 report confirmed this: 8% organic growth guidance for 2026 is not the number of a company being disrupted out of existence.

But here’s the nuance: not all data moats are equally deep. Legal databases built over 30 years are durable. A CRM’s customer records, on the other hand, sit in your own company’s data—the AI agent doesn’t need your CRM vendor to access them if you give it database credentials. The moat isn’t the software; it’s the data the software contains that you can’t get elsewhere.

Platform builders: Salesforce and ServiceNow are spending aggressively to build their own agent layers. Salesforce’s entire identity is now Agentforce—CEO Marc Benioff mused in December about renaming the company after it. ServiceNow’s COO Amit Zavery declared 2026 “the year of agentic collaboration in the enterprise.” If they execute, they become the orchestration platform—the contractor who manages the plumbing, not just the pipes themselves. If they don’t, OpenAI and Anthropic will build that orchestration layer for them.

Most Exposed

Workflow-thin tools: Products where the primary value is a UI layer on top of commodity logic—expense reports, basic analytics dashboards, simple project management—face the highest risk. If an AI agent can replicate the workflow without the tool, the tool becomes optional. The test is simple: does this product hold data I can’t get anywhere else, or does it just present my

own data back to me in a pretty interface?

Per-seat-dependent companies: Any company whose revenue model depends on billing for human users touching the product is structurally exposed to the agent shift. Intuit (−34%) is watching AI tax-filing agents approach its core value proposition. Atlassian (−35%) is seeing developers use Claude Code to build internal coordination tools that bypass Jira entirely. These are the canaries in the coal mine.

What You Should Do This Quarter

Whether you’re a team lead, a CTO, or an individual contributor, this shift creates both risk and opportunity. Here’s the framework:

The 5-Question Enterprise Exposure Audit

Which of our software tools does a human need to physically interact with? If the answer is “none—an agent could do it,” that tool is in the blast radius.

What’s our per-seat spend, and how much is over-provisioned? Most companies are paying for 30–50% more seats than are actively used. AI agents will expose this gap—and your CFO will notice.

Are our vendors building agent layers—or are third parties building agents on top of them? If your CRM vendor isn’t shipping its own agent platform, someone else’s agent will sit between you and it. Ask your rep directly.

What data do we hold that AI agents can’t replicate? Proprietary data, domain-specific training sets, and institutional knowledge are the only durable moats. If your competitive advantage lives inside a third-party tool, it’s not a moat—it’s a rental.

Who on our team is already using AI agents to bypass our tools? This is already happening. Shadow AI is the new shadow IT. The sooner you know, the sooner you can lead the transition instead of chasing it.

Run this audit before your next vendor renewal. The companies that renegotiate now—while the market is panicking and vendors are desperate to retain accounts—will lock in favorable terms. The ones who wait will pay full price for software that’s losing leverage every quarter.

Start with the SaaS Audit prompt from this week’s newsletter—it maps your personal tool stack in five minutes. Then bring the 5-Question framework to your next leadership meeting.

The Bottom Line

The SaaSpocalypse isn’t the death of software. It’s the death of software’s privileged position at the top of the enterprise stack.

For two decades, SaaS companies were the interface between businesses and their data. That’s what justified the premium valuations and the per-seat pricing. Now AI agents are claiming that interface role—and the market is repricing accordingly.

The winners will fall into two camps: AI-native companies that own the agent layer (OpenAI, Anthropic, Google), and legacy vendors that successfully build their own agent platforms before they become commodity infrastructure. Everyone else faces a slow slide toward lower margins, consumption pricing, and a fundamentally different relationship with their customers.

JPMorgan thinks the selloff is overdone. Goldman Sachs thinks it’s just the beginning. Both might be right—oversold in the short term, structurally challenged in the long term. The range of outcomes is wider than the market has priced in years.

The only wrong move is assuming this doesn’t affect you.

Go Deeper

Bloomberg: “What’s Behind the SaaSpocalypse Plunge in Software Stocks” — The definitive timeline of how Anthropic’s plugin triggered the selloff.

Fortune: “AI Agents Aren’t Eating SaaS—They’re Using It” (Jeremy Kahn) — The best bull-case argument for why SaaS isn’t dead yet. Includes the gym-membership analogy.

Deloitte: “SaaS Meets AI Agents: Transforming Budgets, CX, and Workforce Dynamics” — How pricing models are shifting from per-seat to consumption-based. Essential for vendor negotiations.

OpenAI Blog: “Introducing Frontier” — OpenAI’s own pitch for treating AI agents like employees. Read to understand the playbook.

AppOmni: “BodySnatcher—Agentic AI Security Vulnerability in ServiceNow” — The scariest AI security disclosure of 2026 so far. If you use ServiceNow, read this today.

This deep dive is a companion piece to the iPrompt Newsletter.
Stay curious—and stay paranoid.

— R. Lauritsen