Claudeforce, Explained: What Happens When the World’s #1 CRM Stops Being a Place You Log Into

For 25 years, “using Salesforce” meant one thing, Opening a browser tab, clicking through Lightning Experience, and living inside the interface Salesforce built for you. On August 26, 2026, Salesforce’s own leadership said out loud that this might be about to change and put real product behind the claim.

The announcement is called Claudeforce, and it’s the clearest signal yet that the next fight in enterprise software isn’t over who owns the best screens.

What Is Claudeforce?

Claudeforce is the name for the expanded partnership between Salesforce and Anthropic, announced during Salesforce’s Q2 FY27 earnings call. Strip away the branding and it comes down to two sentences:
Claude is now woven into Salesforce as a reasoning engine, and Salesforce is now available as a plugin inside Claude governed the entire time by the permissions and business rules your org already runs on.

Salesforce’s own framing on the official Claudeforce page is that this unites “the #1 AI” with “the #1 CRM,” so that Claude can reason over the actual state of your business instead of guessing at it.
Strip the marketing gloss off that line and the underlying idea holds up: A language model, however capable, is only as useful as the data you hand it. Claudeforce is Salesforce’s bet that the data it has spent 26 years accumulating is the thing that makes an AI assistant genuinely useful for revenue work.

The headline product shipping under the Claudeforce name is called “Salesforce in Claude” a plugin that puts 37 prebuilt sales workflows directly inside your Claude conversations.

The Three Layers of Claudeforce

It helps to think of Claudeforce as three separate workstreams bundled under one name, because they have different maturity levels and different audiences.

1. Salesforce in Claude — the new part. A plugin, installed once by an admin, that gives every seller on the team 37 prebuilt skills like pipeline review, deal health checks, meeting prep, account planning, and more, all callable from inside a normal Claude conversation.

2. Claude in Salesforce — the existing part, made official. Claude has quietly been a selectable foundation model inside Agentforce for months, powering the Atlas Reasoning Engine and available in Agent Builder. Claudeforce formalizes that relationship and makes Claude the default model for Agentforce Vibes and Agentforce Coworker.

3. Claude in Slack — the connective tissue. Claude becomes the default intelligence layer across Slack, powering Slack, Claude Tag (an AI teammate that works inside channel threads), and Slack Code, a new space for teams and coding agents to build together in the open.

The thread running through all three: the same underlying data, the same governance rules, whether you’re sitting in Lightning, typing in a Slack channel, or chatting with Claude directly.

How a Request Actually Flows Through the System

This is the part most coverage skips, and it’s the part worth understanding if you’re the one who’ll have to explain it to a security review.

Salesforce in Claude isn’t a new database or a side channel into your org , it’s built on top of Headless 360, Salesforce’s architecture for exposing platform capabilities as tools an AI agent can call directly, without a human clicking through a UI. The specific mechanism is an MCP (Model Context Protocol) server, and rather than exposing hundreds of individual API endpoints as hundreds of separate tools, Salesforce collapsed the whole platform behind a small, deliberately narrow set of operations –> the agent discovers the right operation for the task, describes its technical spec, then dispatches the call.

Here’s what that looks like end to end:

The detail worth sitting with is the box at the bottom: the request runs under your own permissions, not some elevated service account baked into the plugin. That single design choice is what separates “AI assistant with a chat window” from “AI assistant that can actually touch production data safely.”

The 37 Sales Skills — What They Actually Are

Salesforce is positioning the bundle as giving every seller “an AI CRO” — a pipeline co-pilot that never sleeps. That’s marketing language, but the underlying mechanism is more interesting than a chatbot with API access.

A raw MCP connection gives an AI model a pile of operations and hopes it picks the right one. A skill, by contrast, encodes the judgment calls a good rep or manager would already know how to make — which of two overlapping fields to trust when they disagree, what “at risk” actually means for a deal in your specific pipeline, which stage transitions matter and which don’t. That’s the difference between a pipeline review that’s consistent no matter who asks for it, and one that quietly varies depending on how someone phrased their prompt that morning.

Named examples from Salesforce’s own materials include deal health review, meeting prep, and pipeline review, with a first-run onboarding flow that reads a seller’s context across Salesforce, Slack, and any other connected tool to stand up a personalized dashboard — accounts, live pipeline data, and next actions, assembled on the spot rather than configured in advance ahead of time.

Permissions and Governance: The Part That Actually Matters

If you’ve spent any time around enterprise AI rollouts, you already know the real blocker was never “can the model reason well.” It was always “how do we let an AI touch our data without opening a hole we can’t close.”

Salesforce’s answer here is refreshingly boring, in the best sense: there’s no new permissions model to design. Claude sees only what the requesting user is authorized to see, and can only do what that user is authorized to do — the exact same profiles, permission sets, and sharing rules that already govern every login. An admin connects the org once, and every seller inherits access scoped to their own existing permissions from day one, with nothing to re-audit user by user.

On the CRO’s own words, from Salesforce’s official page announcing the partnership:

“My sellers start the day with the pipeline review already done and the account history already there.” — Alexa Vignone, President & Chief Revenue Officer, Salesforce

For write actions, the model is deliberately conservative to start: Salesforce recommends beginning with read-only access, scoped by permission set to specific objects and fields, and adding write access one skill at a time, only once you’ve watched what the agent actually does with read access first. And it’s worth remembering that a write coming from an AI agent is still just a normal Salesforce write — your validation rules, Flows, and Apex triggers all fire exactly as they would if a human made the same change through the UI. Nothing about routing a DML operation through an agent exempts it from your existing automation or your governor limits.

The permission scoping isn’t an abstract promise — it plays out concretely every time two different people ask the same question:

What’s Live Today vs What’s Still Coming

Claudeforce bundles things at very different stages of readiness, and it’s worth being precise about which is which so you don’t plan a Q4 rollout on top of something still in pilot.

Sales is the first persona to ship, deliberately. Salesforce has said it’s using itself as customer zero , its own Account Executives, and Anthropic’s, piloted the plugin against real pipeline before anyone outside either company got access.

What This Means If You Work in Salesforce

For admins and architects, the practical takeaway isn’t “install this immediately.” It’s “audit your permission sets before you have to, not after.”

Every profile that was over-provisioned but harmless behind a slow, click-heavy UI becomes a genuinely different risk profile once an agent can act on it in seconds, at your request, faster than you can double-check. If a sales rep’s profile has edit access to fields they’ve never actually touched in the Lightning UI, that gap was invisible for years. It stops being invisible the moment an AI agent is willing to use it.

For developers, the interesting shift is architectural: Headless 360 treats your org’s capabilities as something to be discovered and called, not just clicked. If you’re already comfortable with MCP concepts — tools, scopes, OAuth flows — you’re most of the way to understanding how this works under the hood. If you’re not, this is a very good moment to get comfortable with them, because “expose it as a tool an agent can call” is quickly becoming as normal a requirement as “expose it as a REST endpoint” was a decade ago.

For anyone building a Salesforce career right now, it’s also worth naming plainly: interviewers are going to start asking about this. Understanding what MCP is, how permission-scoped AI access actually works, and why “the agent runs as the user, not as a service account” matters, is turning into exactly the kind of practical, current knowledge that separates a strong Salesforce candidate from a merely certified one.

How to Actually Try This

Salesforce in Claude itself is still pilot-only, with a public beta planned for September 2026 , you can join the waitlist on Salesforce’s official page to get notified when that opens.

But you don’t have to wait for the plugin to start learning the underlying mechanics. The Headless 360 MCP architecture that Salesforce in Claude is built on is available today through Salesforce’s standard Hosted MCP Servers , just without the managed, single-admin packaging Claudeforce adds on top. If you want a hands-on head start on exactly how Claude and Salesforce talk to each other at the protocol level, walk through our own setup guide:

How to Connect Claude to Salesforce Using MCP Servers: The Complete Setup Guide →

FAQ

Is Claudeforce a Salesforce product I can buy today?
Not directly yet. Salesforce in Claude, the core new product under the Claudeforce name, is available to select pilot customers now, with an open beta planned for September 2026. Claude’s presence inside Agentforce (Atlas, Vibes, Coworker, Agent Builder) is already live and doesn’t require anything new to access.

Does Claudeforce replace Agentforce?
No. Agentforce is Salesforce’s own agent-building platform, and Claude is now one of the foundation models available inside it . Claudeforce sits alongside Agentforce as a separate, complementary product: Salesforce inside Claude, rather than an agent built on Salesforce.

Is this the same thing as the Hosted MCP Servers I might already be using?
It’s built on the same underlying architecture (Headless 360), but Salesforce in Claude adds a managed layer on top: single-admin setup, centrally managed OAuth and permissions, and 37 purpose-built skills, instead of you wiring up and maintaining the connection yourself.

Does Claude get elevated access to my org through this partnership?
No. Every request runs under the permissions of the user who made it — the same profiles, permission sets, and sharing rules already enforced today. There’s no separate elevated service account baked into the plugin for standard use.

What’s the pricing model?
Salesforce has described it as consumption-based, tied to the volume of API/agent activity rather than a flat per-seat fee, with Claude inference itself contracted separately through Anthropic. Exact numbers weren’t published in the initial announcement — expect this to firm up as the open beta approaches.

Should my org pilot this now?
If you’re already comfortable with MCP-based integrations and have healthy permission hygiene, joining the waitlist is low-risk. If you haven’t audited your profiles and permission sets recently, that’s the higher-priority task — do that first, regardless of when you get access to the plugin itself.

Final Thoughts

Salesforce spent over two decades convincing the world that the way to use Salesforce was to open Salesforce. Claudeforce is Salesforce’s own leadership saying, on an earnings call, that this might no longer be true and shipping product that assumes they’re right.

Whether “the UI is the AI,” as Salesforce’s own framing puts it, actually holds up once real sellers use this at scale is genuinely an open question. If you’re building anything Salesforce-adjacent right now, this is worth understanding at the protocol level, not just the press-release level. Start with the MCP fundamentals, and the rest of Claudeforce will make a lot more sense when the beta opens.

Official Salesforce sources for this article:

• Claudeforce — official partnership page
• Agentforce
• Salesforce + Slack
• Headless 360
• Data 360
• Dreamforce 2026 registration