Anthropic lit the match on Feb. 5. It rolled out vertical plugins for Claude Cowork (its enterprise agent workspace) in legal, finance, sales and more. Software and data vendors lost roughly $285 billion in market value as investors processed what "agents with workflows" means for SaaS economics. Big decliners included Thomson Reuters (intraday down up to 18%), RELX (about 14%), Wolters Kluwer (about 11%), FactSet (about 10%), plus hits across S&P Global, Moody's, Oracle, Salesforce, and Adobe. See SiliconANGLE for the original writeup.
Quick definitions:
Claude Cowork: Anthropic’s agentic desktop workspace for knowledge work - a research preview that now supports 11 open-sourced "starter" plugins. This is a workspace where AI "agents" act like coworkers to help with tasks. See TechCrunch.
Vertical plugins: Prebuilt, job-specific bundles (for example, contract review or NDA triage; financial analysis and reporting) that connect agents to the right tools and data.
Orchestration layer: The control plane that coordinates agents, data, identity/permissions, and audit across workflows - basically how AI actually runs inside an enterprise and how everything is tied together. See OpenAI.
The money math:
Anthropic raised $30 billion at an implied $380 billion valuation and says it’s at about a $14 billion annualized run rate - roughly a 27x multiple. That sounds spicy if you’re just buying a model, but more reasonable if you’re buying the orchestration choke point where customers centralize access to tools and data. Note this is a private valuation, not a public market cap. See SiliconANGLE.
The "coding wedge," explained:
Researcher Raphaëlle d’Ornano (Decoding Discontinuity; partner at theCUBE Research) argues coding became the first dependable agent beachhead. Win developer workflows, and you win enterprise orchestration - then every other type of work follows. That early lead in developer tools is the wedge that pries open the rest of the stack. Read more at SiliconANGLE.
OpenAI vs. Anthropic - two approaches:
OpenAI Frontier: A horizontal governance platform. It focuses on agent identities, permissions, and audit, and aims to work with agents from multiple vendors. Think of it like "Workday for AI coworkers." See OpenAI.
Anthropic Cowork: A vertical, Apple-like stack. Start with an agent (Claude, plus code-focused agents), standardize connections via MCP (Anthropic’s connector protocol), and extend into job-specific plugins. The bet is that execution quality and deeply embedded context create the moat. See TechCrunch.
Why founders should care:
Own your integration surface, or someone else will. Prioritize open APIs and MCP-style connectors (MCP = Anthropic's connector protocol) so your product can plug into agent platforms without losing data control.
Design for exit ramps. Ship migration paths and exportable "institutional memory" so customers aren’t locked into a single agent vendor’s context store.
Build for agent governance. Expect procurement to ask about agent IAM (identity and access management), audit trails, and fallback review processes. If you can’t answer those, you’re a plugin, not a platform.
Models are table stakes; orchestration is the moat; code is the wedge. For a deeper playbook, read Raphaëlle d’Ornano’s analysis on Decoding Discontinuity and SiliconANGLE’s breakdown.
Get daily insider tech news delivered to your inbox every weekday morning.
