Salesforce and Anthropic recently announced an expanded partnership called Claudeforce, timed to land during a Salesforce earnings call. The stock coverage and the earnings recap have already been written elsewhere. This is the version for people who administer a Salesforce org, not the version for people who trade its stock.
What actually shipped
The first product under Claudeforce is the Salesforce in Claude plugin, built for Claude Cowork and carrying 37 prebuilt sales skills. A seller can ask Claude to prep for a meeting, review deal health, or analyze a pipeline, and get answers pulled from live Salesforce data without opening Salesforce itself. The plugin can also take governed action, meaning it can update records and move pipeline work forward, not just report on it.
It runs on something Salesforce calls AIforce, described as an enterprise harness that brings business data and workflows to any agent through MCP servers, APIs, and CLI tools. In practical terms, the Claudeforce MCP server layer is what lets Claude reach Salesforce data and act on it, the same standard underneath most third-party Salesforce AI integrations, not a proprietary connection unique to this partnership. Actions taken inside Claude route back through Salesforce, so the org's existing business rules and permissions still apply. An admin connects the plugin once, rather than provisioning it seat by seat.
The broader partnership has a few more strands worth knowing. Claude becomes the default model across Slack and Slackbot. Salesforce says it will roll out Claude Code and Claude Enterprise to its own developers and knowledge workers. And the two companies disclosed a reciprocal commercial relationship: Salesforce uses Claude internally, Anthropic uses Salesforce as its CRM and Slack as its internal collaboration tool.
Salesforce also shared an internal adoption number worth noting, since it is the kind of figure that tends to get repeated without context. The company says 83% of its own workforce uses the Claude-powered Slackbot, attributes 8.1 million annualized productivity hours to it, and told analysts on its earnings call that Slackbot reached 1 million active users, up 150% quarter over quarter. That is Salesforce's own reported figure about its own internal usage, not an independent study, and it describes Slackbot broadly rather than the new sales plugin specifically.
Untangling the names: Claudeforce, Agentforce, and Salesforce in Claude
If you've been confused by how many AI product names Salesforce is now running at once, you're not alone. Salesforce's own employees have reportedly had the same reaction, with some internally describing the overlap between Claude Tag and Slackbot as barely distinguishable. Sorting out what's actually new helps.
Claude did not arrive inside Agentforce with this announcement. It has been a foundation model inside Agentforce since late 2025, serving as the reasoning model behind the Atlas Reasoning Engine, the layer that decides what an Agentforce agent does next. Claude also already powers Agentforce Vibes and Agentforce Coworker by default, and is selectable inside Agent Builder for teams building custom agents. If your org has been running Agent Builder with Claude selected for months, nothing about that setup changed this week.
What Claudeforce actually adds is the reverse direction, plus a formal name for the whole relationship. Salesforce data and workflows now reach into Claude through the Salesforce in Claude plugin, rather than only Claude reasoning inside Salesforce's own agent tools. Put simply: Claude inside Agentforce is not new. Salesforce inside Claude is the new part, and Claudeforce is the umbrella term covering both directions plus the deepened Slack integration.
What it doesn't do, at least not yet
A few limits matter more than the headline.
- It's sales-only, for now. The 37 skills that shipped cover meeting prep, deal health reviews, pipeline analysis, and similar sales workflows. Salesforce lists Service, Marketing, Commerce, and several other clouds as coming soon, and its release says additional prebuilt skills will begin launching in late 2026, without firmer dates attached.
- It's pilot-only, not generally available. As of the announcement, Salesforce in Claude is live for a select group of pilot customers. An open beta is expected in September, but Salesforce has not published an exact date. Whatever your procurement timeline looks like, plan around "sometime soon, unconfirmed," not a firm launch date.
- No pricing has been disclosed. Salesforce's own release includes the standard caveat that availability, packaging, and pricing are subject to change, and that customers should base purchasing decisions on what is actually available today, not on what was announced.
- It is not an admin or configuration tool. Salesforce in Claude is built for sellers working with pipeline and account data. It is not built for the kind of work that fills an admin's actual backlog: building flows, adjusting permission sets, standing up validation rules, or restructuring page layouts. Nothing in the announcement changes that scope.
The Trust Boundary detail worth getting right
A lot of the coverage of Claudeforce blends two different data paths together, and the distinction matters if your org has regulated-industry compliance requirements.
Claude's presence inside Salesforce, as the reasoning model behind the Atlas Reasoning Engine, Agentforce Vibes, and Agentforce Coworker, runs through Amazon Bedrock inside what Salesforce calls its Trust Boundary. In practical terms, that means the inference workload and the underlying data stay inside Salesforce's own controlled environment rather than making a round trip to a separate third-party API. Salesforce has described Anthropic as the first model provider fully integrated this way, and named CrowdStrike and RBC Wealth Management as early adopters of that specific deployment path when it announced the arrangement in October 2025.
The Salesforce in Claude plugin runs the other direction: it brings Salesforce data out into the Claude Cowork interface, not the reverse. That is a different data path than the Bedrock-contained Agentforce integration, even though both fall under the Claudeforce umbrella. If your compliance team asks where data goes when a seller uses the plugin versus when an Agentforce agent reasons with Claude internally, those are two different questions with two different answers, and it's worth getting that distinction right before you commit to a pilot, not after.
On the governance side specifically for the plugin, Salesforce has said actions route back through Salesforce so existing business rules and sharing settings still apply, and that you control how much autonomy Claude has on the write side. The example Salesforce gives is Claude checking with you before it emails anyone outside your company, or sending once you are comfortable letting it, rather than acting freely once connected. Worth confirming exactly what that configuration looks like and who controls it before rolling access out broadly, since "configurable" and "configured correctly for your org" are not the same thing.
What this means if you administer the org, not sell out of it
Most of the coverage of this announcement is written for sellers, investors, or people evaluating Claude as a platform generally. If you're the person who actually configures and maintains a Salesforce org, a few things are worth taking away.
Claudeforce is a genuine signal about where Salesforce is headed: more of the day-to-day interaction with the CRM happening through an AI client rather than the Salesforce interface itself, governed by the org's existing permissions and business rules rather than a separate access model. That direction is worth watching regardless of which specific product wins the sales-workflow use case.
But nothing here currently touches the admin's job. Salesforce in Claude is scoped to sales data and sales actions, gated to pilot customers, and has no announced date for general availability, pricing, or an admin/config skill set. If your actual bottleneck is a backlog of flow builds, permission changes, and object configuration, Claudeforce as announced doesn't address it, at least not yet, and Salesforce hasn't said when or whether it will.
If you're the one fielding the "can we get on the pilot?" question from a sales leader, a few things are worth confirming before you say yes:
- Which data path applies to your use case. Ask whether the specific workflow in question runs through the Bedrock-contained Agentforce path or the Salesforce in Claude plugin path, since they behave differently for compliance purposes.
- Who controls the write-access settings. Confirm which role sets how much autonomy Claude has to update records without a check-in, and whether that's an org-wide setting or configurable per team.
- What happens after the pilot. Pilot access, by definition, doesn't come with committed pricing or a guaranteed general-availability date. Get clarity on what changes, and what it might cost, before the team building around it gets attached to the workflow.
That gap is exactly the territory Cirra AI already operates in. Cirra AI is an MCP server built specifically for Salesforce admin and metadata work, connecting to Claude and other MCP-compatible AI clients so admins and consulting partners can create objects, adjust permission sets, build flows, and manage the org's configuration layer using natural language, all within the connected user's own Salesforce permissions. Whatever Claudeforce eventually becomes for sales teams, the admin backlog is a separate problem, and it has a separate answer today.
For a look at the infrastructure layer underneath any Claude-Salesforce integration, including Claudeforce, see Claude Platform and Salesforce Integration: What Actually Changed.
