A Value-First Guide to Surviving the SaaSpocalypseContents
Part I — Diagnosis
Chapter 1
What Happened
It started with a short post on a blog.
On January 30, 2026, Anthropic, the company that makes the AI model Claude, published a post called "Customize Cowork with plugins." Cowork is Anthropic's agentic desktop tool, and a plugin packages up what the AI needs for one kind of job: instructions, connections to other tools, a few commands. The post said: "To get you started, we're open-sourcing 11 plugins built and used by our own team." Sales. Finance. Data. Marketing. Customer support. And one more, described in a single line: "Legal — Review documents, flag risks, and track compliance."
That was a Friday. The plugins were open source and up on GitHub, and paid Claude users could use them the same day.
The following Tuesday, February 3, the market noticed.
Thomson Reuters, one of the big publishers of legal information, was down by nearly 18% during that day's trading, by Reuters' count. RELX, the owner of LexisNexis, was down 14%. Wolters Kluwer was down about 13%. LegalZoom slumped 19.7%. Reuters quoted Mike Archibald of AGF Investments on what set it off: "I think Anthropic came out with some plug-ins to tackle the legal space."
The selling didn't stay in legal. Bloomberg reported that the new Anthropic tool "sparked a $285 billion rout in stocks across the software, financial services and asset management sectors."
The week's name came from a trading desk, by way of Bloomberg. "We call it the 'SaaSpocalypse,' an apocalypse for software-as-a-service stocks," said Jeffrey Favuzza, who works on the equity trading desk at Jefferies. "Trading is very much 'get me out' style selling."
A small catalyst
Here's what still strikes me about it: the catalyst was small, set against what it moved.
Anthropic didn't announce a machine that could replace a workforce. It released plugins. Packages of instructions that told its AI how to do specific jobs in specific fields. Eleven of them. Open source.
A post like that doesn't move hundreds of billions of dollars unless the weakness is already there. The market had been sitting in a burning room for years, insisting everything was fine. The plugins didn't start the fire. They just made it impossible to keep pretending the room wasn't on fire.
The repricing logic
For a generation, software-as-a-service companies built their businesses on two assumptions: humans drive the work, and humans need interfaces.
If humans drive the work, you charge per human. Seat-based pricing. The more people an organization employs, the more seats it buys. The seat-based model was built on that arithmetic.
If humans need interfaces, you build interfaces. Dashboards. Workflows. Click paths. Feature sets. Each new feature justifies a higher price per seat.
Both assumptions held for a generation. That week, the market started pricing in the possibility that neither one would hold much longer.
When an AI agent can review a contract, it doesn't occupy a seat, and it doesn't need the screen the seat was paying for. The seat economics and the interface economics come under pressure at the same moment, for the same reason.
So what got repriced wasn't "software." It was inventory. The companies hit first and hardest that week were legal and data businesses, whose value sat in a warehouse of information that people paid to get into. The thing that keeps its value when an agent can do the looking is context: the understanding of what's needed, when, and why.
Who got hit, and the pattern
Here's the pattern as I see it, from most exposed to least.
The most exposed are narrow data wrappers: companies whose value is organizing data behind a proprietary interface. When AI can get to the information directly and do the job the interface was for, the wrapper has a hard time saying what it adds.
Next are point workflow solutions: tools that handle one step in a process. When AI can carry the whole process end to end, a tool that handles one step starts to look like a feature, not a product.
The least exposed are broad platforms, where many business functions already operate together. For them the question isn't whether AI replaces them. It's whether they become the place where humans and agents work together.
In May, Harvard Business Review published an article by Christopher Stanton arguing that leaders were drawing too broad a lesson from the selloff, and that software that pairs pooled, proprietary data with predictive work is in a different position from workflow and record-lookup software. That's the same line, drawn for the people deciding what to buy.
Markets overreact
Markets overreact. That week was no exception, and some of the companies sold off hard may well turn out fine.
So don't read this chapter as "SaaS is dead." It isn't. The SaaS model is being renegotiated. Seat-based pricing for commodity workflows is on its way out. So is wrapping data in a proprietary interface and charging admission.
Stock prices move back and forth. A structural shift doesn't move back.
Two versions of "This Is Fine"
You know the meme. The cartoon dog sitting in a room engulfed in flames, coffee cup in hand, calmly declaring: "This is fine."
Here's what's easy to miss about that meme: there are two versions of "This Is Fine."
Version One: Denial. The room is burning. You see the flames. You choose to believe they'll go out on their own. You keep running the same playbooks. You keep paying for the same tools. You're not fine. You're choosing not to notice.
Version Two: Preparation. The room is on fire — and you're genuinely calm. Not because you're in denial, but because you built the room differently. Your walls are fireproof. Your exits are clear. You understood the structural vulnerability and chose to address it.
About six weeks after that Tuesday, I wrote this, in a letter to a fellow practitioner about a different upheaval in our own corner of the industry: "Here is what I've come to believe: the disruption is rarely the end of the story. It's usually the beginning of a better one." It holds here too. Version Two isn't luck. It's what a better story looks like from the inside.
What the months since have shown
The first edition of this book was written close to that week. This one is written with the months since in view, and two things those months made plain run through the chapters ahead.
The first is that the weakness was financial as well as technical. In May, CIO.com laid out how a decade of buying software companies with borrowed money had rested on the same assumptions the plugins put under pressure. Chapter 2 tells that part.
The second is that the companies that make the AI models didn't behave as if the model were the whole product. A week apart in May, Anthropic and OpenAI each launched a company built around engineers who work closely with customers to build the AI into how each customer works. Chapter 6 tells that part, because it's the strongest evidence I know that context, not the model, is where the value sits.
Neither one changed the argument. Both made it harder to argue with.
What this book is about
This book is about the difference between those two versions of "This Is Fine." It's about why one group built a room that caught fire and another group built a room that didn't. It's about what the second group understood that the first group missed. And it's about how to move from one room to the other.
The next two chapters explain the diagnosis: why what happened in February was predictable in its shape, if not its date, and what the disease underneath it is called.