Last week, Anthropic released a set of plugins for its Cowork desktop agent. Within 48 hours, $285 billion had been wiped from software and services stocks. Thomson Reuters had its worst day on record. LegalZoom dropped 20%. The Nasdaq posted its worst two-day tumble since April.
Then, while the dust was still settling, Anthropic dropped Opus 4.6 — a model specifically designed to make Cowork even better. And OpenAI launched Frontier, its own enterprise agent platform.
The message from both companies is the same: AI isn't just helping you use software anymore. It's becoming the software. Let's break it down.
On January 30, Anthropic released 11 plugins for Claude Cowork - its agentic desktop tool that can read, edit, and create files on your computer. Within 48 hours, $285 billion had been wiped from software and services stocks.
Wall Street panicked. A Goldman Sachs basket of US software stocks fell 6% on Tuesday - its worst single-day drop since the tariff selloff in April. Thomson Reuters plunged 16%. LegalZoom sank 20%. RELX (parent of LexisNexis) dropped 14%. Wolters Kluwer fell 13%. Even Salesforce and Workday took hits. Indian IT outsourcing stocks collectively lost about Rs 2 lakh crore ($24B) in a single day.
Then came Opus 4.6 - on Thursday, Anthropic released an upgraded model specifically optimized for Cowork: - One million token context window (up from 200K) - Better at knowledge work benchmarks than GPT-5.2 - New PowerPoint integration - Files closer to "production-ready" on first try
What everyone's missing: The real story isn't that Anthropic built a legal plugin. It's that a foundation model company decided to compete directly with the application layer. This is the Amazon playbook - use your infrastructure position to move up the stack and eat the businesses that depend on you.
OpenAI launched Frontier - an enterprise platform for building, deploying, and managing AI agents across your entire business.
The pitch: your AI agents should work like employees. Frontier gives them onboarding, shared context across your company's data, feedback loops to improve, and explicit permissions for what they can and can't do.
HP, Intuit, Oracle, State Farm, Thermo Fisher, and Uber are already on board. One financial services firm got 90% more time back for client-facing teams. A tech company saved 1,500 hours per month in product development.
The interesting move: Frontier is open to agents from competitors - including Anthropic and Google. OpenAI wants to be the operating system for enterprise AI, not just the model provider.
Why this matters: Anthropic and OpenAI are making the same bet from different angles. Anthropic is going direct - build the agent, ship the plugins, own the workflow. OpenAI is going platform - be the infrastructure that manages everyone's agents. Both threaten existing SaaS businesses.
Perplexity Model Council - Instead of picking one AI model, Model Council runs your query across three frontier models simultaneously (Claude Opus 4.6, GPT-5.2, Gemini 3.0) and synthesizes a single answer. A "chair" model reviews outputs, identifies convergence/divergence, surfaces unique insights. Available for Perplexity Max subscribers.
The SaaS selloff feels like a wake-up call, but the question isn't "will AI replace software?" The real question: where does proprietary data still matter?
Thomson Reuters has decades of curated case law. LexisNexis has legal research infrastructure. Bloomberg has financial data no one else has. The companies that will survive aren't the ones with the best interfaces - they're the ones with data moats so deep that even the best AI agent needs their data to be useful.
If your business model can be replicated by an open-source plugin on GitHub, it's time to rethink the business model.