On February 3, 2026, Anthropic shipped a product update. Not a new model — a plugin. Claude Cowork, a legal automation tool that could draft contracts, review compliance documents, and handle routine paralegal work. It was impressive. It wasn’t magic.

Within 48 hours, $285 billion in market capitalization had vanished from software stocks.

By mid-February, the damage was closer to $2 trillion. Salesforce and Workday dropped more than 40%. Atlassian plunged 35% in a single week. India’s IT sector had its worst trading day since March 2020. Financial pundits scrambled for metaphors. Morgan Stanley reached for poetry. In 2011, Marc Andreessen had famously declared that “software is eating the world.” Morgan Stanley’s new line? “Something is eating software.”

SaaS stock decline in February 2026 showing Salesforce, Workday, and Atlassian losses

I remember watching it happen in real time. I was sitting in my home office, half-reading a thread about Anthropic’s announcement, half-working on a client project that runs on — you guessed it — a stack of SaaS subscriptions. The first headline felt like noise. The second felt like a trend. By the third day, when my timeline was nothing but red charts and panicked takes, I started wondering if my own clients would start asking uncomfortable questions about their software budgets.

Some of them did.

What actually triggered the selloff

Let’s be precise about what happened, because the narrative got muddied fast. Anthropic released a plugin. It was good. It was not a general-purpose replacement for enterprise software. But the market didn’t react to the product. It reacted to the implication — that if AI agents could automate legal workflows today, they could automate sales workflows, project management workflows, HR workflows, and finance workflows tomorrow.

And if they could do that, who needs all those software subscriptions?

The math that spooked investors was brutally simple. Morgan Stanley put it in a research note that became the most-quoted sentence of the month: if AI cuts staff by half, it cuts software subscriptions by half. That’s it. That’s the whole thesis. SaaS companies sell seats. Fewer humans means fewer seats. It doesn’t matter how good the software is if there’s nobody left to log in.

“If AI cuts staff by half, it cuts software subscriptions by half.” — Morgan Stanley

This isn’t AI replacing software. It’s AI reducing the humans who use software. That’s a pricing model crisis, not a technology crisis. And it’s a distinction most of the hot takes completely missed.

The seat compression problem

Here’s where things get genuinely uncomfortable for SaaS companies, regardless of whether you think the selloff was rational.

The entire SaaS business model is built on a simple assumption: more employees means more licenses. Company grows, headcount goes up, software spend goes up. This has been the engine of SaaS growth for two decades. It’s why Wall Street loved recurring revenue metrics so much — they tracked neatly with employment growth.

Now reverse the assumption. What happens when a company can do the same work with significantly fewer people? Not because the software got worse, but because AI agents handle tasks that used to require a human sitting in front of a screen?

Why pay for 100 Salesforce seats when 3 AI agents can process the same pipeline? Why pay for 200 Jira licenses when an AI project coordinator manages task tracking for the whole org? The software itself might be perfectly good. The pricing model is the problem.

According to IDC, pure seat-based pricing will be functionally obsolete by 2028. That’s not a prediction from some fringe blog — that’s one of the largest technology research firms in the world. The shift is already measurable: the share of SaaS companies using seat-based pricing dropped from 21% to 15% in the past year alone.

Foundation Capital goes further. They’ve coined the term “Service as Software” to describe what’s coming — not software as a tool for humans, but software that is the worker. They estimate the total addressable market at $4.6 trillion. That’s a big number, but it’s also a different number. It accrues to whoever builds the agents, not necessarily to whoever sells the seats.

The Klarna reality check

No company embodied the “SaaS is dead” narrative more neatly than Klarna. In late 2025, CEO Sebastian Siemiatkowski made headlines by announcing the company was shutting down Salesforce and Workday. The story was perfect: a major fintech company dramatically ripping out its enterprise SaaS stack and replacing it with AI. Commentary sections lit up. “This is the future,” everyone said. “SaaS is finished.”

Except that’s not quite what happened.

When reporters actually dug into the details, Klarna hadn’t replaced Salesforce with an AI agent. They’d replaced it with Deel — another SaaS product — and built an internal graph database for some customer data. They didn’t eliminate SaaS. They consolidated it. They got more efficient, yes. Revenue per employee jumped from $400,000 to $700,000. That’s legitimately impressive.

But here’s the part that didn’t make the headlines: Siemiatkowski himself later admitted that most companies couldn’t do what Klarna did. They had the engineering talent, the technical infrastructure, and the organizational willingness to rebuild core systems from scratch. Your average mid-market company running 47 different SaaS tools does not.

Most companies couldn’t do what Klarna did. — Sebastian Siemiatkowski, CEO Klarna

The Klarna story followed a pattern I’ve seen over and over with AI narratives: “AI replaces X” is the headline. “AI enables consolidation of X” is the reality. Less dramatic, more accurate, worse for engagement metrics.

Which SaaS survives and which doesn’t

Not all software is equally vulnerable, and this is where the analysis gets interesting.

Bain published a framework in late 2025 that I keep coming back to. They divided enterprise software into two categories: deterministic systems and probabilistic systems.

Deterministic systems are ones where accuracy is non-negotiable. Financial accounting. Regulatory compliance. Tax calculation. Payroll. These systems work because they follow rules precisely, every single time. You don’t want your accounting software to be “approximately correct.” You want it to be exactly correct, down to the penny, or someone goes to jail. AI agents are bad at this. They hallucinate. They approximate. They’re probabilistic by nature.

Probabilistic systems are ones where “good enough” is actually good enough. CRM data entry. Task tracking. Email drafting. Meeting scheduling. These are workflows where human judgment was always a bit fuzzy anyway, and an AI agent that’s 90% accurate is replacing a human process that was maybe 85% accurate. (I’ve seen how most people use Jira. Let’s not pretend the data was pristine to begin with.)

The Bain framework predicts that deterministic SaaS — your accounting software, your compliance tools, your ERP systems — will be relatively resilient. The data has to be right, the audit trail has to be complete, and the regulatory stakes are too high for “probably correct.”

Probabilistic SaaS — your project management tools, your CRM platforms, your low-end marketing automation — is in serious trouble. These are the categories where AI agents can plausibly do the work without a human in the loop, which means the seat-based model collapses fastest here.

Both sides of the argument

I want to be honest about the fact that smart people disagree sharply about what’s happening.

Bank of America released a note in the first week of February calling the selloff “bizarre” and suggesting the tech free fall “doesn’t make any sense.” Their argument: SaaS companies are the beneficiaries of AI, not the victims. Who builds the infrastructure that AI agents run on? Software companies. Who has the data that AI agents need? Software companies. The selloff, in their view, was a classic panic — investors pattern-matching from the DeepSeek scare a year earlier without doing the actual analysis.

Bloomberg, on the other hand, published an opinion piece arguing that SaaS companies “deserve their AI reckoning.” The argument: for years, these companies got away with annual price increases, feature bloat, and vendor lock-in because switching costs were too high. Now AI is lowering switching costs by making it easier to migrate data, replicate workflows, and build custom solutions. The SaaS tax is finally being challenged.

I think both sides are partially right, which is the boring answer but probably the correct one. The selloff was overdone in the short term — a $2 trillion haircut based on a plugin launch is not rational price discovery. But the long-term structural challenge to seat-based pricing is real. The question isn’t whether SaaS business models will change. It’s how fast, and who adapts.

The business model shift

Here’s what I’m actually watching, past the daily stock ticker drama.

The companies that survive this transition won’t be the ones that ignore AI or the ones that panic. They’ll be the ones that fundamentally rethink how they charge for software. Seat-based pricing was a convenient proxy for value: more users meant more value delivered. But if AI agents are doing the work, the proxy breaks.

What replaces it? A few models are emerging:

Outcome-based pricing — charge for results, not access. Instead of $50 per user per month for a CRM, charge per deal closed, per support ticket resolved, per invoice processed. This aligns the vendor’s incentive with the customer’s actual goal.

Consumption-based pricing — charge for what’s used. API calls, compute time, data processed. AWS pioneered this model for infrastructure; it’s now spreading to application software.

Agent-inclusive licensing — treat AI agents as a new category of user with different pricing. Some SaaS companies are already experimenting with this: a human seat costs $X, an AI agent seat costs $Y (usually less, but not zero).

None of these are settled yet. The SaaS pricing model of 2030 probably hasn’t been invented. But the direction is clear: the era of “multiply headcount by license fee” is ending.

What this means for developers (and everyone else)

If you’re a developer, you might be wondering why a stock market story matters to you. Here’s why.

The companies that buy software also employ developers. If those companies are cutting SaaS spend because they need fewer humans, they might also be cutting development teams. Or they might be growing development teams because they need custom AI solutions instead of off-the-shelf SaaS. Both things are happening simultaneously right now, and the balance depends on the company.

If you build on top of SaaS platforms — Salesforce developers, Atlassian plugin authors, HubSpot integrators — the ground is shifting under you. Not disappearing, but shifting. The Klarna lesson applies: companies aren’t eliminating software, they’re consolidating it. That creates demand for developers who can build the bridges, migrate the data, and design the custom solutions that replace five generic tools with one purpose-built system.

If you build SaaS products, the message is even more direct: rethink your pricing before your customers rethink their subscriptions.

The SaaSpocalypse might look ridiculous in hindsight — a $2 trillion overreaction to a plugin launch. Or it might look like the first tremor of a genuine restructuring. I don’t know which. But I do know that the seat-based pricing model, the one that built the SaaS industry into a multi-trillion-dollar category, is facing its first serious existential challenge. And the companies and developers who adapt to that reality early will be the ones still standing when the dust settles.

Marc Andreessen was right in 2011. Software did eat the world. The question for 2026 is simpler and scarier: what eats software?

Sources