Our commentary on this week’s Claudeforce announcement. If you read our Built to Last series on the enterprise context layer, consider this the bonus chapter.
Salesforce and Anthropic announced Claudeforce on August 26th, an expanded partnership unveiled as Salesforce opened its earnings call. The name alone says something. Salesforce has never before put its “force” suffix on another company’s product.
10K partners with both companies. We’ve focused on Salesforce for more than ten years, and Anthropic became our other strategic partner this summer, so we work with customers in this space every day.
What was actually announced
The headline product is Salesforce in Claude, a plugin that gives sellers live access to their Salesforce data from inside Claude. It comes with 37 prebuilt sales skills, including meeting prep, deal health review, and pipeline review, and it can take action, from updating records to drafting outreach, without the seller opening Salesforce. Select pilot customers have it now, with an open beta planned for September and skills for more business functions coming after that.
Underneath the plugin sits AIforce, the piece worth studying: a harness that exposes business data, workflows, business logic, actions, and governance to agents through MCP servers, APIs, and CLI tools. The plugin is one product built on it. The bigger story is that AIforce can connect Salesforce to any head: any AI tool or interface a company chooses to work through.
The rest of the announcement is largely a recap of what was already in place. Claude has been a foundation model inside Agentforce since late 2025, already serving as a reasoning model for the Atlas Reasoning Engine, already powering Agentforce Coworker and Agentforce Vibes by default, and already available inside the Salesforce Trust Boundary through Amazon Bedrock, which matters for regulated industries. The Slack piece extends earlier work: Claude Tag already put Claude inside Slack, and the announcement now positions Claude as the default model across the Slack experience. Claudeforce bundles the existing arrangements under one name and adds the plugin on top.
What matters here
The direction of travel matters most. Salesforce putting its data, its actions, and its permission model into another company’s interface is the clearest statement yet that the CRM’s value no longer depends on the CRM’s screens. That shift has a name, headless, and it’s good news for anyone who owns the data underneath. Salesforce just agreed in a very public way.
The harness matters almost as much. Connecting an AI model to your CRM raises immediate questions: who can see what, which actions are allowed, and how usage stays observable. The announcement answers the first two directly. Every answer and action routes through Salesforce under the requesting user’s existing permissions, so Claude sees what that seller can see and does what that seller can do, nothing more. Observability gets less airtime so far, though routing everything through Salesforce means the activity lands where admins already watch, and the companies say jointly built safeguards and further enterprise controls are coming. Governance that travels with the connection is what lets a company say yes to a new tool quickly.
The Trust Boundary piece deserves a note too. When teams decide where to run AI against customer data, security usually casts a heavy vote, and running inside Salesforce’s own boundary used to be an argument for staying with Salesforce’s own tools. Claude operating inside the Trust Boundary shrinks that gap. The tool choice gets to be about the work more often, which is where it belonged all along.
And meeting prep earning a spot among the headline skills says something about where the value is. It’s one of the highest-frequency, highest-friction workflows in sales, and the first place an account team feels what a unified view of the customer is worth.
What matters less
The number, 37 skills, matters less than it sounds. Prebuilt skills are starting points, and they will multiply. What a skill can actually tell your sellers is capped by the data it can reach. Meeting prep against a thin CRM produces thin prep, whatever the packaging says.
The keynote language matters less too. “An AI CRO” and the suggestion that sellers will never open Salesforce again are stage productions. Real teams will work in both places for a long time, and in-flow work inside Salesforce remains exactly what Agentforce and Coworker are built for.
The thing that didn’t change
Every skill in that plugin reasons over your revenue context, and it can only be as good as that context. A unified profile that joins core CRM with product usage and engagement signals gives these tools something worth reasoning about. Disconnected tables give them fragments, whichever logo is on the head.
The heads are changing faster than anyone predicted a year ago. The companies that built the context layer underneath get to point every new head at it the week it arrives. The companies that didn’t will watch the demos and wait.
That’s the argument we spent our Built to Last series making, and this week is the strongest evidence for it yet.
If you’re working through what Claudeforce means for your stack, we’d love to have a conversation.