TL;DR: on August 26, 2026, Salesforce and Anthropic announced Claudeforce, a partnership that makes the Salesforce CRM fully drivable from Claude, with 37 prebuilt sales skills and an open beta planned for September 2026.
This goes beyond a connector. Salesforce, which has spent 25 years selling software whose value flowed through its screens, now says its users will no longer need to open them. If your team already avoids opening your CRM, this news deserves five minutes of your attention: it says where the real value of a CRM actually lives.
Key takeaways
- Salesforce and Anthropic announced Claudeforce on August 26, 2026; the Salesforce in Claude plugin ships 37 prebuilt sales skills (meeting preparation, deal health review, pipeline review).
- The plugin has been in pilot since August 26, 2026; an open beta is planned for September 2026, and skills for other business functions will arrive in late 2026.
- Salesforce expects to spend roughly $300m on Anthropic tokens in 2026, and its stake in Anthropic is valued at around $5bn according to Bloomberg.
- The announced billing model is double-sided: one Salesforce invoice for API consumption and one Anthropic invoice for model inference.
What Salesforce and Anthropic announced on August 26
Claudeforce is a two-layer partnership, made official in a Salesforce press release dated August 26, 2026. First layer: Claude, Anthropic's model, becomes the default reasoning engine across Salesforce's AI products, including the Atlas Reasoning Engine, Agentforce Vibes and Agentforce Coworker. Second layer, the one drawing the headlines: a plugin called Salesforce in Claude lets users query, update and act on live CRM data directly from Claude, without ever opening Salesforce.
The plugin delivers 37 prebuilt sales skills: meeting preparation, deal health reviews, pipeline analysis and updates, with so-called governed actions that respect permissions and leave an audit trail. According to VentureBeat, the architecture relies on MCP (Model Context Protocol) servers that inherit the user's rights: if a salesperson cannot see a record, the MCP server cannot see it either.
The table below sums up the announced components and their availability.
| Component | What it is | Availability |
|---|---|---|
| Salesforce in Claude | Claude plugin with 37 sales skills over live CRM data | Pilot since August 26, 2026, open beta in September 2026 |
| Atlas Reasoning Engine | Claude becomes the default reasoning model behind Agentforce | Live |
| Agentforce Vibes and Coworker | Claude by default for app building and the internal assistant | Live |
| Regulated industries | Claude access via Amazon Bedrock, inside Salesforce's Trust Boundary (banking, healthcare) | Live |
On the financial side, The Next Web reports that Salesforce expects to spend roughly $300m on Anthropic tokens in 2026, and that its equity stake in Anthropic is valued at around $5bn, a figure first reported by Bloomberg in June 2026. No public pricing has been disclosed for Claudeforce so far.
Why "the UI is the AI" is an admission
The line from Marc Benioff, Salesforce's CEO, quoted by VentureBeat, deserves a literal reading: "we're bringing together the world's #1 AI and #1 CRM; here, the UI is the AI." Patrick Stokes, Salesforce's president of applications and marketing, goes further in the same piece: evaluating a deal used to take "like 10,000 clicks inside of Salesforce"; the same evaluation now happens "in like 30 seconds" in plain language.
A vendor that bills per-seat licences has just publicly conceded that its screens are a cost, not a value. Stokes locates Salesforce's value elsewhere: in "data and the metadata, the years worth of encoded workflows." In other words, what is expensive in a CRM is clean data and accumulated business rules; the interface displaying them becomes interchangeable.
That shift carries a direct financial consequence. The announced business model stacks two consumption-based invoices: Salesforce bills API calls according to the licence edition, Anthropic bills model inference. The fixed per-seat subscription, much criticised but predictable, gives ground to a variable cost driven by actual usage. That is fairer for light users, and harder to budget for heavy ones.
What does this change for a small or mid-sized business?
First consequence: the number one CRM problem in smaller companies, adoption, changes in nature. A CRM the team never opens used to be a training or discipline problem. If even Salesforce is organising the bypass of its own screens, the conclusion writes itself: the right goal is not getting people to open the CRM, it is getting information in and out without friction. A note dictated to an assistant, a record updated from the messaging tool, a pipeline queried in plain language; that is the standard now settling in.
Second consequence: the value of your CRM shifts to your data and your rules. A business owner hesitating between switching tools and cleaning up data now has a clear answer: screens get replaced, clean data and formalised processes get carried over. Investing in record quality, well-defined sales stages and follow-up rules pays off whatever tool ends up displaying them three years from now.
Third consequence, a more cautious one: none of this is available to a typical SME today. The plugin is in pilot, the open beta is announced for September 2026, pricing is not public, and consumption costs are hard to estimate before the first real-world reports. Add the hosting question: Salesforce answers regulated industries with Claude access through Amazon Bedrock inside its Trust Boundary, but a European company committed to local data hosting should ask where its customer data gets processed before wiring its CRM to any AI assistant.
How this connects to my own work
I see this movement from both sides. On the Salesforce side, I worked as a Salesforce Administrator at e-Enfance, the French nonprofit that operates 3018, the national hotline against cyberbullying: Apex development, LWC components and a 3CX telephony integration. The logic behind Claudeforce, permission-governed actions on top of business data, matches exactly what makes a well-administered Salesforce deployment solid.
On the custom side, the bet Claudeforce is making is the same one behind Emma, the CRM I co-built for the 3018 hotline counsellors: the value sits in the data model, the 5 unified channels and the business rules serving 12 counsellors and managers, not in a catalogue of screens. When the data structure is sound, changing how you access it, through a screen, an AI or an automation, becomes a reasonable project rather than a rebuild. That is the test I recommend before any custom CRM project: would your data model survive a change of interface?
Salesforce just said out loud what the market had been whispering: a CRM is a well-governed database, the rest is packaging. The useful question for a business owner is no longer "which CRM should we pick?" but "are our data and processes worth plugging into an AI?". If the answer is no, the tool was never the problem.
Frequently asked questions
What is Claudeforce?
Claudeforce is the expanded partnership announced on August 26, 2026 between Salesforce and Anthropic. It makes the Salesforce CRM drivable from Claude through a plugin with 37 prebuilt sales skills, and makes Claude the default reasoning model behind the Atlas Reasoning Engine, Agentforce Vibes and Agentforce Coworker.
Can you already use Salesforce from Claude?
Partially. The Salesforce in Claude plugin has been limited to select pilot customers since August 26, 2026. An open beta is planned for September 2026, with skills beyond sales arriving in late 2026. No pricing has been published.
What does Claudeforce change about CRM pricing?
The announced model stacks consumption-based billing on top of the classic subscription: one Salesforce invoice for API calls and one Anthropic invoice for model inference. Total cost becomes variable and usage-dependent, which makes upfront budgeting harder than with per-seat licensing.