Anthropic's Claude Cowork announcement wiped $285 billion off software stocks in a week. Thomson Reuters dropped 15%. LegalZoom fell 20%. FactSet, S&P Global, Moody's, and Nasdaq all took sharp hits.
The market has decided that AI agents are coming for enterprise software. There's just one problem: the same investors fleeing SaaS are also betting that AI infrastructure spending is about to collapse.
Both things can't be true at once.
Bank of America's analysts called this out directly: the market is pricing in "internally inconsistent" outcomes. The bear case for AI infrastructure assumes enterprises will slash AI spending because the ROI isn't there. The bear case for SaaS assumes AI is so capable that it will replace their products entirely.
Pick one.
If AI is a bust, SaaS companies keep their customers. If AI is genuinely transformative, the infrastructure companies building it are worth more, not less. The current market posture requires believing that AI is simultaneously too weak to justify investment and too strong for existing software to survive. BofA compares the reaction to the "overblown" DeepSeek selloff in January 2025. They project AI capex will quadruple to $1.2 trillion by 2030, with real bottlenecks being power, land, and data center capacity; not a collapse in demand.
What triggered this
The selloff traces back to Claude Cowork, Anthropic's new product that gives Claude the same agentic capabilities powering Claude Code, but pointed at knowledge work instead of coding. It ships with 11 domain-specific plugins: legal, finance, sales, marketing, customer support, product management, and more. The legal plugin can review documents and flag risks. The finance plugin can build financial models and track metrics. The sales plugin connects to CRMs and knowledge bases.
This is built on the same infrastructure as Claude Code, meaning it works with "much more agency than a regular conversation." It creates plans, executes tasks systematically, and can work through dozens of steps autonomously. The timing compounds the pressure. Opus 4.6 arrived with a 1-million-token context window and a new "agent teams" feature that lets you deploy multiple AI agents working in parallel.
The combination of smarter models, longer context, and multi-agent coordination is exactly what you'd need to replace chunks of enterprise software workflows.
The damage wasn't evenly distributed. Thomson Reuters (-15%), RELX (-14%), and LegalZoom (-20%) took the biggest hits. These aren't random casualties.
Professional services companies are exposed because their products do exactly what Cowork's plugins can do out of the box. Legal research? Contract review? Document risk flagging? That's the legal plugin. Financial modeling and market intelligence? That's the finance plugin. The market is essentially pricing in a future where fewer humans need these tools because AI handles the underlying tasks directly.
Salesforce and CRM vendors face a murkier situation. The data itself is still valuable, but the interface layer is increasingly what AI agents are designed to handle. The part where humans click around in dashboards? That's exactly what agents replace. Your CRM becomes infrastructure that Claude talks to, not software that sales reps spend their days in.
$285B on a research preview
Here's what nobody in this selloff has shown: mass enterprise cancellations.
The capability is demonstrated. The extrapolation to catastrophe is speculation. Cowork launched to Max subscribers first, with Pro subscribers getting access on January 16 and Team/Enterprise on January 23. We're weeks into this product existing. The market has priced in years of disruption based on a research preview.
Our read: the professional services exposure is real. If an AI can review a contract competently, that's direct competition with Thomson Reuters' core business. But "can review a contract" and "has replaced contract review at enterprise scale" are very different claims.
Two distinct pressures are operating on different timelines. The first is immediate: budget starvation. Enterprises are redirecting software spending toward AI infrastructure. This is real and happening now. When your IT budget gets reallocated to GPU clusters and AI platform licenses, existing SaaS renewals get scrutinized harder. The second is longer-term: capability overlap. AI agents increasingly can do what enterprise software does. This is demonstrated but not yet happening at scale. The demos are impressive. The production deployments at Fortune 500 companies are still early.
The market appears to be pricing in both timelines simultaneously.
That's how you get a $285 billion selloff on a research preview. The question isn't whether AI agents can do knowledge work. They can. The question is how fast enterprises adopt them for production workflows, and how much of their existing software stack becomes redundant when they do.
The stocks that got hit hardest are the ones where the capability overlap is clearest. If you're selling legal research tools, an AI that can read a million tokens of case law and synthesize findings is direct competition. If you're selling CRM software, the threat is more indirect; Claude needs your data, even if it replaces your interface. The $285 billion bet is that this transition happens fast. The BofA analysis suggests the market may be getting ahead of itself, just like it did with DeepSeek.
Both can't be right.
The real test: enterprise renewal data over the next two quarters. That's when we'll know if this is a panic or a preview.
Sources cited: Claims as analysis: