Salesforce’s Agentforce has been fighting for relevance against Microsoft Copilot, ServiceNow, and a growing stack of AI-native competitors. So the company’s decision to partner with Anthropic, rather than lean entirely on its own models, is a telling strategic move. On August 26, Salesforce and Anthropic announced Claudeforce, an expanded partnership that brings Claude’s reasoning directly into Salesforce’s data layer, workflows, and governance infrastructure, and pushes Salesforce’s CRM capabilities into Claude’s interface.
The partnership launches with a concrete product: Salesforce in Claude, a plugin for Claude that gives sellers access to 37 prebuilt sales skills, including meeting prep, deal health review, and pipeline management, without leaving the Claude interface. It’s built on AIforce, Salesforce’s new enterprise connection layer that exposes business data, workflows, and permissions to external agents via MCP servers, APIs, and CLI tools. The pitch is that sellers get an AI that can reason over live Salesforce data and take governed actions, without IT needing to rebuild permissions or integration logic from scratch.
What Claudeforce actually includes
The partnership has four distinct layers, and it’s worth separating them clearly because they serve different buyers:
- Salesforce in Claude: A Claude plugin with 37 prebuilt sales skills. Claude reads the seller’s Salesforce context, including pipeline, accounts, and Slack signals, and generates interactive dashboards. Admin connects once; the whole team gets access.
- Claude inside Salesforce: Claude is available as the reasoning model for Agentforce’s Atlas Reasoning Engine, powering Agentforce Coworker and Agentforce Vibes by default. It runs within the Salesforce Trust Boundary via Amazon Bedrock, which matters for regulated industries.
- Claude in Slack: Claude is now the default model for Slackbot, Claude Tag for team decision-making, and Slack Code for multiplayer development. Salesforce says Slackbot is already driving 8.1 million hours of annualized productivity gains internally, more than double the previous quarter.
- Mutual adoption: Salesforce uses Claude as its preferred AI assistant internally. Anthropic uses Salesforce as its CRM and Slack as its internal work platform. Both companies are customers of what they’re selling together.
Why this matters beyond the press release
The most interesting part of Claudeforce isn’t the plugin count. It’s the architecture underneath. Salesforce is positioning AIforce, its enterprise connection layer, as the infrastructure that lets any frontier model plug into enterprise workflows without bespoke integration work. If that layer gains adoption, Salesforce becomes harder to displace even as the underlying AI models keep changing.
That’s a direct counter to Microsoft’s strategy, which embeds Copilot deeply into Office and Azure to create similar stickiness. Salesforce’s bet is that the CRM data layer, not the productivity suite, is the more valuable anchor point for enterprise AI. And with decades of customer data already living in Salesforce, it’s not a weak argument.
For enterprise buyers evaluating AI tooling right now, the Claude-in-Salesforce path is one of the cleaner options available. Claude running inside the Salesforce Trust Boundary means data doesn’t leave the customer’s governance environment, which removes a major objection for financial services, healthcare, and government buyers. That’s a real differentiator compared to generic API integrations with other frontier models.
Availability and what to watch next
Salesforce in Claude is in pilot now and targets open beta in September 2026. Additional prebuilt skills are expected later in 2026. Salesforce is also making Claude Code and Claude Enterprise available to all of its internal developers and knowledge workers, naming Claude its preferred AI assistant across the company.
The companies say more integrations across Claude, Salesforce, and Slack are coming. So the 37 skills today are the floor, not the ceiling. Whether the prebuilt skill library actually covers enough real-world sales workflows to reduce prompt engineering for enterprise teams is the practical question that pilot customers will answer over the next few months. That result will matter more than any announcement.
