Salesforce and Anthropic just launched Claudeforce, which embeds Salesforce's entire CRM directly into Claude as a plugin. The core product—Salesforce in Claude—ships with 37 pre-built sales skills covering meeting prep, deal reviews, and pipeline analysis. Users can query and update live CRM data without touching Salesforce's interface. It's available to pilot customers now, with an open beta in September. The framing here matters: Salesforce's president Patrick Stokes compared this to what Claude Code did for developers, calling it "a whole new way to work" for knowledge workers. This represents a radical shift for enterprise software—Salesforce is essentially saying its UI doesn't matter anymore, only the data underneath. The technical architecture is straightforward by design. When a seller asks Claude to update a deal, Claude reasons through available skills (human-readable instructions), finds a match, and executes against Salesforce's MCP server while respecting the user's existing permissions. An administrator sets this up once with central authentication and permissions management—no per-user configuration needed. The security model inherits from Salesforce's existing access controls, so if you can't see a record in Salesforce, Claude can't access it either. This matters to enterprise buyers because there's nothing new to audit or stand up separately. Stokes addressed the obvious tension head-on: if sellers live in Claude instead of Salesforce, doesn't that make Salesforce less valuable? He argues the opposite. Salesforce's value isn't the UI—it's the data, metadata, and encoded workflows built up over years. He gave a concrete example: a seller's morning routine of evaluating opportunities traditionally takes 10,000 clicks and mental synthesis. Through Claude, the same work happens in 30 seconds. The result is paradoxical but compelling: users actually consume Salesforce more because the friction of clicking around vanishes. Salesforce will charge for this through headless consumption pricing (more API calls based on your license tier), while customers separately pay Anthropic for Claude inference. This two-invoice structure signals a deeper shift in enterprise software economics—from per-seat licensing toward consumption-based models as AI agents, not humans, become the primary users of SaaS platforms.