Salesforce has opened its customer records, business rules and approval workflows directly to Claude, allowing sellers to interrogate live pipeline, update deals and execute approved actions without opening the CRM at all. The arrangement was announced on 27 August as Claudeforce, an expanded strategic partnership between Salesforce and Anthropic that puts each company’s core product inside the other’s.
It launches with Salesforce in Claude, a plugin carrying 37 prebuilt sales skills covering meeting preparation, deal health review and pipeline review, which Salesforce positions as an AI chief revenue officer working alongside the seller. The skills were built jointly by the two companies and draw on Claude’s reasoning, agentic tool use and generative UI, including an onboarding process that reads a seller’s context across Salesforce, Slack and any other connector in Claude, then assembles a dashboard of their accounts, pipeline and live data in views the model generates on the fly.
Underneath sits AIforce, the Salesforce harness that exposes business data and workflows to any agent through MCP servers, APIs and command line tools. The commercial argument is that this removes the integration and middleware work that has historically sat between a model and a company’s system of record.
Access to the rules matters more than access to the data
Plenty of assistants can already read a CRM record. What Salesforce is offering here is the layer above the data, the routing and entitlements and business logic that determine whether an action is permitted in the first place. Salesforce in Claude sends every action back through Salesforce so those rules are enforced at the point of execution. An administrator connects the plugin once, with authentication and permissions handled centrally, and every seller on the team has access from day one, with no per-user setup and no second permissions model to maintain.
That addresses the question most CIOs raise first about agentic deployments, which is what happens when the model gets it wrong. Keeping execution inside the platform means the decision about whether an action is allowed stays in the system that already holds that decision.
“Probabilistic intelligence alone doesn’t run a company, and deterministic systems don’t reason,” said Marc Benioff, chair and chief executive of Salesforce, describing the combination as a dynamic interface that thinks, reasons and acts.
The traffic runs in both directions. Claude is available in Agentforce as a reasoning model for the Atlas Reasoning Engine, powers Agentforce Vibes and Agentforce Coworker by default, and is available in Agent Builder. Through Amazon Bedrock, it also runs inside the Salesforce Trust Boundary, which the companies point to as the route for regulated industries to deploy domain-specific AI while keeping data and workloads contained. Anthropic is the first LLM provider to be fully integrated within that boundary.
In Slack, Claude becomes the default model across the experience, powering Slackbot, feeding team decisions through Claude Tag, and acting as a founding partner for Slack Code. The intent is that teams reason through a decision in the channel where the discussion is already happening and then trigger the resulting action in Salesforce without switching tools.
Each company is now the other’s customer
Salesforce is Anthropic’s preferred CRM and uses Slack as its internal work platform, with Shield, Backup and Sandboxes supporting security, resilience and development. Salesforce says Slackbot has driven 8.1M hours of annualised productivity gains internally, more than doubling quarter on quarter, and it will make Claude Code and Claude Enterprise available to all of its developers and knowledge workers as its preferred assistant and productivity tools.
“Companies can point Claude at the customer information and business context that they’ve been building in Salesforce for decades, and use it to actually run and grow their businesses,” said Dario Amodei, chief executive and co-founder of Anthropic.
Salesforce in Claude is with select pilot customers now and is expected to reach open beta in September 2026, with further prebuilt skills arriving from Q3 2026. Neither company has set out regional availability, which leaves Gulf buyers in banking and government with an open question about how a Trust Boundary deployment maps onto local data residency obligations, and how quickly the skills library grows beyond sales.
