The SaaS-pocalypse is real — but not for the reasons Wall Street thinks

The week $285 billion disappeared

This week, the software sector shed $285 billion in market value. The trigger was Anthropic releasing a set of plugins for its Claude Cowork AI agent last Friday. The plugins automate professional workflows that entire categories of enterprise software exist to support: legal research, contract review, data analysis, CRM. A $20-per-month AI subscription doing work that enterprise software charges six figures for.

The hot takes split on cue. One camp declared software dead. The other called it hysteria. Neither got it right.

Two wrong answers

The doom crowd sees AI agents handling professional workflows and concludes every SaaS company is getting replaced. That ignores how infrastructure works. You don't discard decades of operational data and institutional knowledge because a new tool can draft a contract.

The denial crowd has a different problem. They're correct that enterprise software won't vanish overnight. They're wrong that the market is just spooked. The software index is down 30% from its September peak, and the decline has been building for months, not days. Dismissing a trend that keeps going as irrational is comfortable, not useful.

The real question isn't whether AI kills software. It's why the market reacted this violently to what amounts to a set of prompts and workflow configurations.

The gym membership model

For a decade, the SaaS business model rewarded a specific pattern: add features, add seats, raise prices. The metric that mattered was revenue per user, not value per user. That produced bloated platforms full of features nobody asked for, dashboards nobody checked, and integrations that existed to justify "platform" pricing.

Per-seat licensing worked like a gym membership. The best customers were the ones who never showed up. A company paying for 500 seats where 200 people actually logged in was the model working as designed. The vendor counted all 500.

AI agents don't need seats. They need API access. One agent handling the work of fifty people requires one connection, not fifty licenses. That arithmetic breaks the per-seat model.

But the arithmetic was already broken. Customers were paying for access, not outcomes. AI just made the gap between price and value impossible for the market to ignore.

Software that created work

The structural critique of SaaS pricing isn't new. Analysts have been arguing for years that per-seat models are misaligned with the value they deliver. The selloff isn't a surprise to anyone who was paying attention. It's a correction that was overdue.

What makes this moment different is that AI gave the critique a price tag. When a $20 monthly subscription can replicate workflows that enterprise software charges thousands for, every customer starts asking the same question: what was I actually paying for?

The answer, for a lot of enterprise software, is uncomfortable. Customers were paying for access to a system that required them to do work inside it. Mandatory dashboards that generated reports nobody read. Required manual steps in processes that should have been automatic. Workflows designed to keep users inside the product rather than to make their jobs easier.

Every unnecessary click was a tax on the organization, and the cumulative cost was enormous. AI didn't invent the problem. It sent the invoice.

What the market got wrong

The selloff treated every software company identically, and that's where the market overshot. There's a meaningful difference between software that holds irreplaceable operational data and software that wraps a workflow around somebody else's model.

Systems of record have something AI agents can't generate: decades of context about how companies actually operate. Which approval paths apply to which deals, which exceptions are routine, which escalation rules matter. You can't reproduce fifteen years of institutional knowledge in a prompt. That data gravity is a real moat. Capabilities improve on a curve measured in months. Operational context accumulates on a curve measured in years.

Workflow wrappers are a different story. If your product is an interface layered on top of a model, and the model maker ships a plugin that does what you do, your moat was always temporary. The market is pricing that realization right now.

The indiscriminate selloff created a distortion. Software companies with deep data gravity are trading at the same discount as companies whose entire product can be replicated with a well-written prompt. That gap will close. But the correction for the latter group has further to go.

The bar just moved

We build software. This isn't an analyst's observation from the outside.

What this week makes clear is something that's been true for a while but easy to overlook: software that charges for access rather than doing something obviously worth paying for is running out of time. AI didn't create that standard. It collapsed the timeline.

If you're building software today, one question matters more than it did a month ago: does your product do work, or does it create work?

Software that does work takes a problem off someone's plate and gives them back their time. It charges for outcomes because the outcomes are visible. It doesn't need switching costs or seat-count inertia to justify its price.

Software that creates work forces humans through mandatory steps and into dashboards that exist for the vendor's reporting, not the user's job. It charges for access because the value can't stand on its own.

The selloff isn't a panic. It's a repricing. And the companies that were coasting on seat growth, feature bloat, and switching costs just lost their margin for error.