Anthropic Ships 37 Sales Skills Straight Into Salesforce's CRM With a Human Approval Gate

Anthropic says roughly 7,000 Salesforce sellers are already using Salesforce in Claude, the beta plugin it shipped September 15 that puts a company’s accounts, opportunities, and pipeline directly inside Claude rather than a separate CRM tab. The plugin carries 37 skills built for account-executive workflows — meeting briefings, call prep pulled from Salesforce, Slack, and email, deal scoring against a team’s own sales methodology, and automatic post-call follow-ups that turn a transcript into an opportunity update. Every write-back to a Salesforce record still requires explicit seller approval, and the plugin only sees data a user’s existing Salesforce permissions already allow — the two design choices that separate a genuine productivity tool from a compliance risk.

The release formalizes a partnership Salesforce and Anthropic first announced as “Claudeforce” on August 26, when Salesforce committed roughly $300 million to Anthropic tokens this year on top of an existing $5 billion stake in the company. What’s notable is how fast the pattern moved from partnership announcement to production use: SaaStr disclosed in late August that it had already been running a homegrown Claude-based agent on top of Salesforce’s API for six months, with CRM data usage up 10x and revenue growing 47% year-over-year — evidence that agent-native CRM access, not just chat-assisted CRM lookup, is where the real gains show up.

Named early adopters — GitLab, Siemens, Legora — plus a companion Slack connector for deal-channel summaries suggest this ships as infrastructure, not a pilot feature. For companies still deciding whether a rep needs to open Salesforce at all during the workday, that’s the real question this release forces: not whether an AI assistant can summarize a CRM, but whether the CRM’s own interface is becoming optional.