Salesforce Just Said You Won't Need to Open Salesforce
· CX Pulse
Salesforce and Anthropic announced Claudeforce on Tuesday. 37 sales skills, run from inside Claude, without opening the CRM.
Salesforce Just Said You Won't Need to Open Salesforce
On Tuesday, Salesforce and Anthropic announced Claudeforce. The headline feature is a plugin called "Salesforce in Claude," and it ships with 37 prebuilt sales skills: meeting prep, deal health review, pipeline review.
It lets a seller query live CRM data, update the pipeline and take governed actions from inside Claude. Without opening Salesforce.
Read that again with the company in mind. Salesforce built a thirty year business on people logging into Salesforce.
The quotes are worth having
Marc Benioff: "We're bringing together the world's #1 AI and #1 CRM, the best of both worlds."
Dario Amodei: "Salesforce in Claude brings frontier intelligence into systems where commercial activity happens."
And Salesforce's Patrick Stokes, on the ambition: what Claude Code did for developers, "we think we're about to do the same thing for knowledge workers."
Pilot customers have it now. Open beta is September. More skills land late in the year.
What actually just came apart
A system of record is two things stacked together. Underneath, the data plus the rules about who may change what. On top, the interface where people go to work.
For thirty years those were sold as one product, and the interface was the part you paid for per seat.
Claudeforce separates them. Data, permissions and audit trail stay with Salesforce. The place work happens moves somewhere else. Salesforce is betting the governance layer is the valuable half and the screens were never the moat.
That's a real bet, and they may well be right.
If you run CX or revenue operations at scale
You already have a permission model. Roles, sharing rules, field-level security, validation rules, an audit trail your compliance team has opinions about. None of that was designed for this.
The question is whether your controls survive a change of actor. Your sharing rules assume a named human in a session. When the actor is an agent invoked from a different vendor's surface, on behalf of a user, the questions get sharper. Does field-level security still hold. Does the audit entry name the person or the assistant. If a rep's agent updates 40 opportunities in a minute, does anything flag it, and would you want it to.
Enablement loses the surface. Every process you have encoded in page layouts, required fields, guided flows and "don't skip this stage" training lives in the interface. Move the work into a chat surface and the guardrails you built by arranging screens do not come with it. What comes with it is whatever the API enforces, which is usually a smaller set.
There is a licensing conversation coming and it will arrive before you're ready. If a meaningful share of seller work happens in Claude, someone in procurement will ask what the Salesforce seat is now for. The honest answer is the data, the governance and the integrations, which is defensible. It is a different conversation than the one your renewal was built on.
Reporting gets murkier before it gets better. Activity capture, adoption metrics and the dashboards you use to manage the team all assume the work happened in the app. A period where usage looks like it collapsed while performance holds is worth anticipating rather than explaining after the fact.
If you're running a smaller operation
The same shift, with the opposite problem. Large organisations have a permission model that may not survive the change. Most smaller teams never built one, because everyone is an admin and the only way in was a login screen and a human.
That assumption is what this changes, and it changes for a two-person shop and a two-thousand-person org at the same time.
The honest caveat
This is 37 skills, in pilot, for sellers. It is not the end of software interfaces, and anyone telling you it is this week is selling something.
What makes it worth noticing is who shipped it. The company with the most to lose from people not opening its app is the one that built the way around it.
The five minute version
Whatever holds your customer records, pull up the permission model and ask one question of it: what happens when the thing making the request is not a person.
Can an agent acting for a rep edit a closed deal, change a price, alter a forecast category, export a contact list? If it can, does the record say who actually did it?
Most organisations can answer the first question and not the second. That gap has been theoretical for years. It stops being theoretical in September.
Source: Salesforce, "Salesforce and Anthropic Announce Claudeforce," 26 August 2026, and VentureBeat.