Skip to content
NewsProduct launch

Salesforce Opens Claudeforce Beta at Dreamforce 2026 with 37 CRM Skills Built Into Claude

· by Pondero Newsdesk

The short version

Salesforce and Anthropic revealed the full Claudeforce product stack at the Dreamforce 2026 keynote on September 15. A Salesforce in Claude plugin with 37 prebuilt sales skills opened in public beta this month, and Claude became the default model across Agentforce and Slack.

Salesforce Opens Claudeforce Beta at Dreamforce 2026 with 37 CRM Skills Built Into Claude

Salesforce and Anthropic moved from announcement to open beta in three weeks. At the Dreamforce 2026 keynote in San Francisco on September 15, the companies revealed the full product scope of Claudeforce: a "Salesforce in Claude" plugin now accepting public signups, Claude installed as the default model for Agentforce's reasoning engine and Slack, and a new runtime that lets agents run for days rather than minutes.

What shipped

The "Salesforce in Claude" plugin delivers CRM data inside Claude Cowork. Per Salesforce's press release, the plugin launched to select pilot customers before Dreamforce and opened to the public in September 2026. It ships with 37 prebuilt sales skills: meeting prep, deal health review, and pipeline review are the flagship three, with additional prebuilt skills set to follow in late 2026. First-time setup reads the seller's enterprise context and generates a tailored dashboard. A single admin configuration handles authentication and permissions across the organization.

Claude is now set as the default model for Agentforce's Atlas Reasoning Engine and Agentforce Coworker. It also became the default model underlying Slack. Per Salesforce, Salesforce employees have accumulated 8.1 million hours of annualized productivity gains through the Claude-powered Slackbot, a figure the company said more than doubled quarter over quarter. Claude Tag joined as a founding partner for Slack Code, enabling developers to tag Claude directly in a Slack channel. The agent opens a dedicated code workspace, previews changes, and waits for team approval before applying them.

The enterprise plumbing is AIforce, a Salesforce-built harness that grants agents access through MCP servers, APIs, and CLI tools. Enterprises can run Claude-powered agents through Amazon Bedrock within the Salesforce Trust Boundary, keeping CRM data inside the customer's permissioned environment.

New agents and a longer clock

Salesforce debuted several new Agentforce agents at Dreamforce: Casey for sales, Piper, and Hunter. The most consequential addition is the long-horizon runtime. Per the Dreamforce 2026 recap on LinkedIn, this runtime lets agents pursue goals across days or weeks rather than executing a single bounded task. Hunter became the first Salesforce agent deployed on this runtime.

Two customer deployments provided concrete production data. TSA's Ace agent handles 100,000 monthly traveler conversations and resolves 96% of them without human escalation, per the LinkedIn recap. A Smarsh customer support agent resolves seven out of ten sessions without any human involvement. Salesforce packaged these changes under the Winter '27 product release, framing them as a structural shift from single-task assistance to full workflow execution.

Why it matters

Before Dreamforce, analysts and trade coverage noted that Salesforce's pre-event materials carried almost no customer-reported ROI figures or production cost data. The TSA and Smarsh numbers from the keynote are the first public production data points. They are what enterprise buyers need before committing to a CRM-native AI workflow: not benchmarks, but live session volumes and resolution rates from named accounts.

The beta opens the evaluation cycle to any Salesforce enterprise customer. The 37 prebuilt skills work against actual CRM data today, not a sandbox. The centralized auth model removes one common IT blocker: sales reps access CRM context inside Claude Cowork without a separate credential workflow.

The long-horizon runtime changes the scope of the automation question entirely. An Agentforce agent that runs a pipeline qualification cycle across several days is a background process making decisions about deal prioritization, follow-up timing, and outreach sequences. Governing that kind of agent requires audit trails, approval gates, and override controls that differ from those needed for a bounded chatbot. How Salesforce documents those controls before Hunter reaches general availability is the governance question every enterprise buyer will need answered. The pre-Dreamforce coverage's observation about absent ROI data extends here: production numbers from pilot customers, not just keynote showcase accounts, will be the credible signal.

For teams using Claude through the Anthropic API or Amazon Bedrock directly, Slack Code and Agentforce Coworker create a parallel procurement path through Salesforce licensing. Current materials do not disclose whether that licensing structure carries different unit economics compared to direct Anthropic access.

What to watch next

The "Salesforce in Claude" open beta is accepting signups now. The skills expansion planned for late 2026 will indicate whether Salesforce treats the plugin as a CRM-focused tool or a broader workplace productivity platform. Hunter's long-horizon runtime is the first production test of multi-day agent governance at enterprise scale. Whether Salesforce publishes detailed override and audit documentation before general availability is the concrete governance milestone to track. Slack Code's Claude Tag founding-partner structure also leaves open whether early participants receive differentiated access or rate terms ahead of any general rollout.

Sources