Claudeforce Beta Puts 37 Salesforce Skills in Claude

Salesforce moved Claudeforce out of a hand-picked pilot and into beta at Dreamforce '26 this week, so it is now worth looking at what the partnership actually ships. Claudeforce is not a product with a price tag. It is the umbrella Salesforce and Anthropic put over everything they build together, and its first shipping piece is Salesforce in Claude, a plugin carrying 37 prebuilt sales skills, live CRM records, and Salesforce's own permission model into the Claude interface. The architectural claim underneath is the interesting one: Salesforce has decided the system of record does not have to be the place you go to read the record.

• Salesforce in Claude reached beta at Dreamforce '26, held September 15-17, 2026, following the August 26 Claudeforce announcement and a limited pilot
• The plugin ships 37 prebuilt sales skills, including daily-briefing, pipeline-review, close-plan, stakeholder-map and call-prep
• A separate Salesforce Development plugin for Claude Code carries more than 40 skills, with a wider skills library published on GitHub
• Claude runs inside the Salesforce Trust Boundary via Amazon Bedrock, keeping model traffic in Salesforce's virtual private cloud

What Does Salesforce in Claude Actually Do?

The plugin gives Claude a live connection to a Salesforce org through a prebuilt Model Context Protocol server. A seller can ask for a pipeline review, get an answer grounded in current opportunity data rather than a stale export, and then update the record from the same conversation. The write path matters more than the read path here. Plenty of tools can show you CRM data in a second window; far fewer can take a governed action against it.

Salesforce groups the 37 skills by the job they serve: daily operations (daily-briefing, pipeline-review, forecast-narrative), deal strategy (stakeholder-map, close-plan, handle-objection), account management (account-plan, lead-triage, customer-health) and execution prep (call-prep, conversation summarisation). A data hygiene set with an sfdc-hygiene-check rounds it out, a quietly practical inclusion given how much CRM value evaporates through incomplete records. Engineering teams get a separate and larger surface: a Salesforce Development plugin for Claude Code with more than 40 skills, backed by a public skills library on GitHub.

AIforce Is the Layer Underneath Claudeforce

The Dreamforce announcement put a name on the plumbing. AIforce is Salesforce's enterprise harness, the layer that exposes business data, workflows, business logic and governance to agents through MCP servers, APIs and CLI tools, without each integration being rebuilt by hand. Claudeforce is AIforce pointed at Claude. Slackforce points it at Slack. Agentforce Coworker points it at Salesforce's own surfaces.

That structure is a bet that the interface is becoming interchangeable while the governed data layer is the durable asset. If one harness already serves three front ends, adding a fourth becomes a configuration problem rather than a rewrite.

Salesforce named Deloitte, GitLab and Legora among the pilot customers, and says Agentforce Coworker activated 100,000 users in its first 35 days. That figure is the company's own, so it is best read as a vendor metric rather than an audited one.

How Does Salesforce Keep Permissions Intact?

This is the question that decides whether any of it reaches a regulated industry, and the answer is that enforcement stays server-side. Salesforce applies its existing permission model, sharing rules and business logic on the back end, so an agent acting through the plugin inherits the access the person already has. Nothing is granted by virtue of the request arriving through Claude instead of a Salesforce screen.

Running the model inside the Salesforce Trust Boundary via Amazon Bedrock extends the same idea to inference. Anthropic describes Claude as the first large language model provider integrated this way, with traffic contained in Salesforce's virtual private cloud. That design targets financial services, healthcare, cybersecurity and life sciences, where deployment topology is often the blocker rather than model quality. Anthropic cites RBC Wealth Management using Claude in Agentforce for meeting prep. It is a familiar pattern for large enterprise Claude deployments, where the rollout question is governance long before capability.

One practical budgeting note: Salesforce's Patrick Stokes told CIO that token consumption for these sales skills is certainly not zero, but nowhere near what a software development use case consumes. Sales skills are short structured calls rather than long agentic coding runs.

What Comes Next for the Claudeforce Partnership

Salesforce lists Service, Marketing, Commerce, Revenue and Field Service as coming soon, alongside Tableau, MuleSoft, Informatica, Data 360 and Headless 360. Additional prebuilt skills begin arriving in late 2026. Pricing and packaging are explicitly subject to change and availability varies by region, so the beta label is doing real work.

The pattern matters beyond the Salesforce ecosystem. A frontier model vendor and an enterprise data vendor agreed the model should come to the data under the data owner's rules, rather than the data being shipped out to the model. Combined with steady capability gains in the underlying models, covered in our look at Claude Fable 5.1, that arrangement is likely to show up well outside CRM. Our broader AI coverage tracks where it lands next.

Sources: Salesforce - Salesforce and Anthropic Announce Claudeforce - August 26, 2026; Anthropic - Salesforce and Anthropic expanded partnership - August 26, 2026; CIO - August 27, 2026; TechNode Global - September 16, 2026.