AI Pulse

Salesforce Put Its Entire CRM Inside Claude. Agentforce Just Hit $1.5 Billion. The August 27 Pulse.

By Felix Maru · August 27, 2026 · 7 min read

Yesterday afternoon, Salesforce and Anthropic announced Claudeforce: a partnership that puts live CRM data, account history, pipeline context, and governed record updates directly inside Claude. On the same day, Salesforce reported that Agentforce, its AI agent product, is now generating over $1.5 billion in annual recurring revenue, up roughly 240% year-over-year. Those two pieces of news are connected. One shows where enterprise AI investment is flowing. The other shows how the interface for accessing that investment is changing. Both matter for anyone running support or ops on the Salesforce platform.

What Claudeforce Actually Is

The centerpiece of the announcement is Salesforce in Claude: a plugin with 37 pre-built sales skills that give you live Salesforce data and governed actions from inside the Claude interface. Meeting preparation, deal health reviews, pipeline analysis, record updates: all accessible without opening Salesforce directly. Pilot customers have access now. Open beta is planned for September 2026.

The technical layer underneath is Anthropic's Model Context Protocol (MCP), which is the same infrastructure other teams use to wire enterprise systems into Claude. What Salesforce built is a first-party, production-grade implementation of that pattern, with Salesforce's permission model and audit logging baked in. The "governed actions" framing in the announcement is doing real work: the plugin can update records only within defined permission boundaries, not on an open-ended basis.

The 37 skills currently cover sales-facing use cases. Support teams are not the launch target. But the architecture is the point. Querying live case data from Claude, pulling account context before a support call, surfacing CSAT history or open ticket counts for a customer: those are the exact same pattern as meeting prep and deal health reviews. If you run Salesforce Service Cloud, this announcement is a preview of what becomes possible in the next one or two product cycles, not a capability you have today.

The plugin meets people where they already work. Claude gains full CRM context. The human agent makes the call. That is the right design for customer-facing work.

The Agentforce Numbers That Deserve Attention

Alongside Claudeforce, Salesforce reported its Q2 FY27 results. Agentforce ARR exceeded $1.5 billion, up over 240% year-over-year. The platform delivered 7 billion Agentic Work Units to date, with 3.2 billion coming in Q2 alone, roughly 97% quarter-over-quarter growth. Salesforce raised its full-year revenue and profit forecasts, and its stock surged roughly 13% in after-hours trading.

There is a number buried in those results that does not get much headline space: Slackbot users grew over 150% quarter-over-quarter. Slack handles a lot of support coordination for companies on the Salesforce ecosystem: internal escalations, customer-facing channels for higher-tier accounts, agent collaboration threads. When AI assistant adoption in Slack nearly doubles in a quarter, it is a reliable signal that the AI tools people actually use are the ones embedded in workflows they already live in, not the ones requiring a tab switch to a new platform.

That is the thread connecting Agentforce growth and the Claudeforce design. The product wins when it shows up where the work already happens.

What This Means for Support Teams Running Salesforce

I want to be careful about what Claudeforce is and is not. It is not an autonomous support agent that handles tickets without a human. It is a plugin that gives a human working inside Claude access to live CRM context and the ability to take actions within defined boundaries. The human agent still decides. Claude does the lookup, the summary, and the draft.

That is the right model for customer-facing support work. The biggest friction in most support workflows is not that agents lack judgment; it is that they spend too much time gathering context. Who is this customer? What plan are they on? What was the last ticket about? What did the previous agent promise? That lookup and tab-switching time is time not spent with the customer. A plugin that surfaces all of it inside the drafting tool cuts the friction without removing the person from the loop.

The support teams I have seen get the most from AI tools are not the ones asking "how many tickets can the AI close without a human?" They are the ones asking "how do I give my existing agents better information faster so they make better decisions?" The Claudeforce architecture, live context plus governed actions plus a human in the seat, is an answer to the second question.

If you are not on Salesforce: the same pattern is being built into other CRM and support platforms. Zendesk's Copilot features, Help Scout's AI drafts, Intercom Fin's context lookups, all of these are versions of the same idea. The Salesforce announcement matters because it is the largest enterprise deployment of the pattern to date, and because MCP makes it more replicable than proprietary integrations.

The Constraint No Partnership Announcement Solves

Google Cloud published its State of AI Infrastructure Report earlier this week, and the leading finding is worth sitting with: 79% of technology leaders surveyed cited security, governance, or operational challenges as their biggest barrier to scaling AI agent deployments.

Google's recommended responses are sensible: Secure AI Frameworks, platform-level governance, task-level provenance, human-in-the-loop checks. The Salesforce plugin is an example of task-level provenance done right. The plugin operates within defined permission boundaries, logs its actions, and respects Salesforce's existing role-based access controls. That is not an accident; it is the design choice that makes enterprise adoption possible.

But the governance setup is still on the organization deploying the tool. You still need to decide which users get access, which records can be updated, what the audit review process looks like, and who gets notified when the AI takes an action. Those decisions do not come pre-configured. A plugin that ships with the infrastructure for governed actions is only as governed as the policies the buyer puts in place.

As these integrations get more capable, that governance work becomes more important, not less. An AI agent with access to account history, deal context, and update permissions is meaningfully more powerful than a chatbot that answers FAQs. The blast radius of a misconfigured permission is larger. The time to build the governance framework is before you have 200 support agents using the tool, not after.

The Quick Verdict

Claudeforce: QUEUE IT if you are on Salesforce or evaluating it. Not because the 37 skills are immediately useful to support teams today, but because the architecture it demonstrates (live CRM context in an AI interface, governed actions, MCP as the plumbing) is where the rest of the market is heading. Understanding how it works and what the governance requirements are now gives you a head start on the deployment decisions you will face in the next 12 months.

Agentforce $1.5B ARR: MUST-READ** for anyone who still treats enterprise AI agent adoption as "early market." It is not early. Enterprises are spending real money at scale. The question is no longer whether to adopt; it is how to adopt well, with the right human oversight and governance in place.

Google Cloud governance report: QUEUE IT if you are in the process of deploying any AI agent tooling. The 79% figure is not a reason to slow down; it is a checklist item. Security and governance challenges are solvable. Skipping them is not.

Sources

Thinking through how to apply the Claudeforce pattern in your support or ops workflow, or trying to get your AI agent governance framework off the ground? Drop me a note and I am happy to compare notes on what I have seen work.

Share 𝕏 in ✉

Comments