The SaaSpocalypse: What Eats Software (And Why the Market Got It Wrong)

Marc Andreessen famously declared that software would eat the world. On February 5th, 2026, Anthropic served up the main course—and $2 trillion worth of software stocks got devoured in the bloodbath that traders are now calling the "SaaSpocalypse."

Within hours of Anthropic launching their Cowork plugins—open-source workflow tools for legal, finance, sales, marketing, and compliance—the entire software sector imploded. Thomson Reuters plummeted 16%. RELX suffered its worst day since 1988, dropping 14%. Wolters Kluwer fell 13%. LegalZoom cratered 20%. Goldman Sachs' software basket shed 6% in a single session, whilst Salesforce has now fallen 40% from its peak.

But here's what the market got spectacularly wrong: SaaS isn't dying. Seat-based pricing is.

The Day Software Ate Itself

I've been building ecommerce systems for 26 years, and I've never seen anything quite like this. Not the dotcom crash. Not 2008. This was different—surgical, targeted, and absolutely predictable if you'd been paying attention.

Anthropic didn't just release another chatbot. They moved up the stack. Way up. Instead of selling APIs that developers integrate, they shipped complete workflows inside Claude. Legal document review. Financial compliance checks. Sales pipeline automation. Marketing campaign management. All the stuff companies have been paying per-seat for since SaaS became a thing.

The Cowork plugins are open-source and available on GitHub. That's not an accident. Anthropic isn't trying to build a moat around individual workflows—they're building one around the entire category of work itself. $285 billion was wiped out in a single session because investors finally grasped the implications.

When the Netlify CEO mentioned that employees had already built AI replacements for their SaaS survey and quoting tools, that wasn't a product demo. That was a funeral announcement for an entire pricing model. When StackBlitz's CEO said "many SaaS vendors we would have previously used are no longer relevant," he wasn't being hyperbolic. He was being accurate.

The Seat-Based Pricing Death Spiral

Here's the uncomfortable truth that every SaaS executive knows but won't say publicly: the per-seat model was always a bit of a scam. Fortune got it right when they pointed out the dirty secret of SaaS—it works like a gym membership. The best customers are those who pay for seats they don't use.

Think about your own company. How many Slack seats go unused? How many Salesforce licenses sit idle because Karen from accounting logs in twice a month to update a contact? How many Jira seats are occupied by stakeholders who create one ticket per quarter?

The per-seat model made sense when software required human operators. One person, one licence, one recurring payment. But AI agents don't need seats. They don't take holidays. They don't get promoted and move to different departments. They don't forget to log out.

An AI agent can process a thousand legal documents whilst your legal team is asleep. It can analyse compliance requirements across fifty jurisdictions whilst your compliance officer is in a meeting. It doesn't need a Zoom seat, a Slack seat, or an Office 365 seat. It just needs compute time and API calls.

That's why Workday's CEO Carl Eschenbach stepped down during the turmoil. When your entire business model depends on counting human seats, and AI can do the work of fifty humans from a single API endpoint, the maths gets ugly fast.

What Actually Dies (And What Doesn't)

The market panic was indiscriminate, but it shouldn't have been. Not all SaaS companies are equal in this brave new world. The survivors fall into three categories:

The Data Owners: Companies that own proprietary datasets or have deep customer integrations aren't going anywhere. Salesforce might lose seats, but they're not losing customers. Your CRM data doesn't magically migrate to Claude because Anthropic released some plugins. The switching costs are still astronomical, and the network effects are still real.

The Platform Plays: Companies that can evolve into platforms for AI agents will thrive. Microsoft isn't worried about losing Office seats—they're building the infrastructure layer for enterprise AI. AWS, Azure, and Google Cloud are the new landlords. AI agents need somewhere to live.

The Workflow Specialists: Highly specialized, industry-specific tools with deep domain expertise will survive. Generic project management? Dead. Specialized regulatory compliance for pharmaceutical manufacturing? Very much alive.

What dies is the fat middle—generic productivity tools that charge per seat for commoditized workflows. Survey tools, basic CRM, simple project management, generic document management. If Anthropic can ship it as an open-source plugin, your $50-per-seat solution is toast.

The Real Winner: Anthropic's $380 Billion Masterstroke

While software stocks burned, Anthropic closed a $30 billion funding round at a $380 billion valuation. That's not a coincidence. They didn't just disrupt SaaS—they absorbed it.

OpenAI launched their Frontier platform the same week, but Anthropic's move was more audacious. They didn't build a marketplace for AI apps. They made AI apps irrelevant by building the workflows directly into Claude. Why would you buy a separate AI legal review tool when Claude can review contracts natively?

This is the difference between selling picks and shovels versus owning the mine. Every SaaS company pivoting to "AI-powered" features is still thinking in the old paradigm. They're building AI tools. Anthropic built AI that makes tools unnecessary.

The Cowork plugins are a trojan horse. They look like productivity tools, but they're actually an acquisition strategy. Every workflow that gets absorbed into Claude is one less SaaS company that needs to exist. Every process that becomes a native AI capability is one less per-seat revenue stream.

The New Economics of Work

Bloomberg's take was harsh but fair: "Salesforce and Other SaaS Companies Deserve Their AI Reckoning." The seat-based model created artificial scarcity in a digital world. It optimized for licence sales rather than actual productivity.

The new model is usage-based, but not in the way most SaaS companies are pivoting toward. It's not about API calls per month. It's about outcomes per dollar. An AI agent that processes your entire legal backlog for $500 is infinitely better value than fifty legal software seats at $200 each.

This shift kills the SaaS growth story. When software required humans, growth was linear and predictable. Hire more salespeople, sell more seats, increase ARR. When AI can do the work of multiple humans, the unit economics break down. Why would a company buy a hundred seats when ten API tokens can do the same work?

The venture capital math breaks too. SaaS companies could raise on projected seat growth. "We have 1,000 customers paying for 10,000 seats, but there are 100,000 potential users in these companies." That multiplier disappears when AI collapses the need for human operators.

What Comes Next

The SaaSpocalypse isn't a one-day event. It's the beginning of a fundamental restructuring of business software. Companies that survive will need to stop thinking about seats and start thinking about outcomes.

The winners will be platforms that enable AI agents, data companies that feed them, and specialists who build workflows too complex for general-purpose AI to replicate. The losers will be anyone charging per-human for work that AI can automate.

For enterprise buyers, this is liberation from the seat-based prison. No more paying for unused licences. No more complex procurement processes for every new hire. No more vendor negotiations based on headcount projections. Just outcomes, measured and paid for directly.

For SaaS companies, it's an existential moment. The question isn't whether AI will disrupt your business model—it already has. The question is whether you can evolve fast enough to survive the transition. Some will. Most won't.

Marc Andreessen was right about software eating the world. But he missed the sequel: AI eating software. And unlike the first course, this one comes with a $2 trillion bill.