Anthropic brings Salesforce into Claude with 37 pipeline skills

Why it matters

Anthropic is using Salesforce's permissions and installed base to move Claude from answering questions to performing revenue work. Salesforce keeps control of the records, making the deal a distribution win for both founders' platforms.

Anthropic, led by CEO and co-founder Dario Amodei (@DarioAmodei), launched Salesforce in Claude on September 15th, giving its AI assistant access to accounts, opportunities and pipeline data while keeping Salesforce as the system of record.

Claude on X

The beta release packages 37 skills for account research, call preparation, deal reviews, forecasting and CRM updates. Claude can collect context from Salesforce, Slack, email and call transcripts, then propose changes such as moving a close date, adding a contact, logging a call or creating a follow-up task. Sellers approve proposed changes by default before Claude writes them back to Salesforce.

The arrangement advances the enterprise strategy Amodei has pursued since leaving OpenAI, where he led research on GPT-2 and GPT-3 and helped develop reinforcement learning from human feedback. His research background also includes Google Brain and a doctorate in biophysics from Princeton. Anthropic's founding argument was that increasingly capable models needed to be reliable, interpretable and steerable. Salesforce in Claude applies that thesis to a less theoretical problem: letting an agent touch revenue data without asking a sales organization to abandon its existing controls.

Anthropic president and co-founder Daniela Amodei, who runs work spanning research, engineering, product, governance and commercial execution, has described trust as both a safety principle and a business requirement. In a 2023 interview with Stripe, she said Anthropic's founders wanted to make AI safer "from day one" and keep humans at the center of how the technology is deployed. The approval step inside Salesforce is a concrete version of that approach: Claude prepares the action, while the seller retains the final decision.

Claude wants the work around the CRM

Salesforce in Claude concentrates on the administrative work that surrounds selling. A scheduled morning brief can assemble upcoming meetings, deals nearing their close dates, at-risk opportunities and unread account threads. Claude can prepare a representative for a call by combining CRM history with recent Slack and email discussions, then turn the resulting meeting transcript into a follow-up email, a channel summary and proposed changes to the opportunity record.

For managers, Claude can generate an interactive pipeline view with coverage by stage, likely deal slippage and account-level detail. It can also draft forecast narratives in a sales organization's preferred format. Those tasks have long supported a busy market for sales operations tools because Salesforce may contain the official record while much of the useful context remains scattered across inboxes, calls and internal conversations.

Anthropic says GitLab, Siemens and Legora have deployed Salesforce in Claude and that 7,000 Salesforce sellers use it in their work. Those figures come from Anthropic's announcement, which provides no product-specific revenue, retention data or independently measured productivity results. Anthropic also has not stated a separate price for the plugin. It is included in beta access on paid Claude plans.

The rollout has an administrative gate. Anthropic says the plugin is available in beta across paid plans, while administrators must request access through Salesforce's AgentExchange and connect Salesforce at the organizational level. Salesforce's August partnership announcement described the integration as available to select pilot customers, with open beta planned for September.

Salesforce keeps the keys

Salesforce in Claude runs under each user's existing Salesforce credentials and permissions. An administrator connects the organization once and decides which groups receive the plugin. Claude reads only the records the user is permitted to access, according to Anthropic, and Salesforce's business rules still govern actions taken through the integration.

That architecture gives both sides what they need. Anthropic gains access to one of the most consequential datasets inside a business without building a CRM. Salesforce keeps ownership of the records, permissions and workflow rules while making Claude another interface to its platform.

The integration places permissioning alongside model quality as a competitive feature. Microsoft offers CRM-connected sales assistants, while Salesforce is building its own agents through Agentforce. Anthropic's pitch is that a seller can work from Claude across CRM records, Slack and email, then send approved actions back to the database where management already expects them to live.

Approval defaults matter when an agent can alter forecasts and customer records. In separate cybersecurity evaluations, Anthropic disclosed that research models reached real internet systems after a test environment was misconfigured. The incidents did not involve Salesforce or customer data, and the evaluated models lacked safeguards used in generally available deployments. They still demonstrated why agent deployments depend on access boundaries, monitoring and human confirmation rather than model behavior alone.

Amodei's enterprise distribution bet

Salesforce in Claude extends Anthropic's effort to turn Claude from a general assistant into an operating layer for specific jobs. Coding provided an early path: RuntimeWire reported on September 14th that Anthropic says Claude writes 80% of its merged code, creating enough testing load to force a rebuild of Anthropic's internal test-selection system. Sales offers another high-frequency workflow with clearer links to revenue and large pools of enterprise software spending.

Anthropic has ample pressure to make that expansion work. In May, Anthropic said it raised $65B at a $965B post-money valuation in a round led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital. Anthropic said its run-rate revenue had crossed $47B earlier that month. Both figures were supplied by Anthropic. At that scale, integrations with established systems such as Salesforce become core distribution infrastructure rather than optional accessories.

The partnership also gives Salesforce a hedge. Salesforce can continue selling Agentforce while allowing customers that prefer Claude to use the same data, controls and workflows. Anthropic gets closer to sellers without asking them to migrate their customer records. Salesforce preserves its position underneath whichever assistant wins the user's attention.

The September release turns the integration into a product with defined actions, approval controls and an initial set of sales workflows. Its commercial test will be whether revenue teams keep using Claude after the novelty of an automatically prepared morning brief wears off, and whether administrators are comfortable granting an external model a governed path back into the pipeline.