On January 12th, Anthropic released Claude Cowork—"Claude Code for the rest of your work." A week later, the Morgan Stanley SaaS index is down 15% for the year. Adobe has lost 45% of its value since end of 2023. The market has rendered its verdict: AI agents aren't coming for software jobs. They're coming for software itself.
The Launch
Claude Cowork arrived quietly, as a "research preview" for Max subscribers. The hands-on reviews were measured—praise for the capabilities, caution about prompt injection risks. By January 17th, Anthropic had expanded access to Pro subscribers at $20/month, warning that they might hit usage limits faster.
Nothing in the announcement suggested an industry-shaking event. This was an incremental product expansion, one AI assistant learning to do slightly more things.
But investors didn't see an incremental product. They saw the first general-purpose work agent priced for mass adoption. Twenty dollars a month to automate "the rest of your work"—the stuff you currently use Salesforce, Adobe, and ServiceNow to organize.
The Math
Enterprise SaaS is a $200 billion annual market built on a simple premise: human work requires organization, and organization requires software. CRM systems because salespeople need to track leads. Project management tools because teams need to coordinate. Creative suites because designers need to iterate.
What happens when the work itself becomes automated?
The bull case for SaaS in the AI era was that these platforms would become the delivery mechanism for AI capabilities. Salesforce pivoted to Agentforce in September 2024, positioning its suite as the control plane for enterprise AI agents. Adobe has been embedding AI everywhere, from Firefly to multi-model image generation.
The bear case is simpler: if AI can do the work, you don't need software to organize the humans doing it.
The Pattern
This isn't the first time AI has spooked SaaS investors. The sector struggled through much of 2025, as each new model release raised fresh questions about which workflows would survive automation. What's different now is the form factor.
Claude Cowork isn't a feature inside existing software—it's a general agent that can operate across applications. Simon Willison noted it's "well positioned to bring Claude Code's powerful capabilities to a wider audience." Those capabilities include reading files, browsing the web, running code, and executing multi-step workflows autonomously.
In other words: doing the work that enterprise software currently orchestrates.
The 15% decline isn't panic. It's a pricing update. Investors are adjusting their models for a future where:
- Per-seat SaaS subscriptions face pressure from per-task AI pricing
- Workflow automation reduces the volume of human actions requiring software support
- General-purpose agents compete with specialized vertical applications
- The value capture shifts from organizing work to performing it
Adobe's Canary
Adobe's 45% decline since 2023 is the sharpest example of this repricing. The company has invested heavily in AI—Firefly generates images, AI Foundry creates custom models, the entire Creative Cloud is getting AI upgrades.
None of it has convinced the market. The fear isn't that Adobe will fail to build AI features. It's that AI features aren't enough when AI can replace the creative workflows that justify $600/year subscriptions.
Why pay for Photoshop if an agent can generate, edit, and iterate on images through conversation? Why pay for Premiere if video generation models eliminate the editing step entirely?
Adobe's response has been to emphasize enterprise value—brand consistency, asset management, commercial safety. These are real differentiators. But they're defensive differentiators, protecting existing revenue rather than capturing new value.
The Incumbents' Dilemma
Salesforce and ServiceNow are attempting a different strategy: embrace the agents, become the control plane. Agentforce positions Salesforce as the orchestration layer for enterprise AI. Marc Benioff has been positioning AI agents as Salesforce's future since spring 2025.
There's logic here. Enterprises won't deploy autonomous agents without governance, audit trails, and integration with existing systems. The SaaS platforms already have the enterprise relationships, the compliance certifications, the data gravity.
But the strategy requires agents to need orchestration. Claude Cowork suggests they might not. When the agent can read your files, check your calendar, send your emails, and update your systems directly, what's the orchestration layer adding?
What We're Watching
The SaaS repricing may be overdone. Markets overcorrect. Enterprise software has deep moats—switching costs, integration complexity, regulatory requirements—that don't disappear because AI can perform tasks.
But the direction seems clear. The software layer between human intent and digital outcome is compressing. Claude Cowork is a $20/month product that can do what required multiple enterprise subscriptions before. As these agents improve—and Anthropic says Cowork is still a "research preview"—the compression will accelerate.
The SaaS companies betting on becoming AI platforms may be right. The ones betting that their existing workflows will remain valuable may not be. And the market, down 15% in nineteen days, is telling us which bet it's making.