What it is, one sentence per part
Claudeforce is the name of the agreement, not of a piece of software. It covers everything Salesforce and Anthropic build together from August 2026 on, in two directions: Claude moves into Salesforce as the reasoning engine, and Salesforce moves into Claude as a plugin.
The first direction is already visible in the org: Claude is available as a model in the Agentforce Atlas Reasoning Engine, and it is the default in Agentforce Vibes and Coworker. Agent Builder keeps the choice, with Claude next to the Amazon Nova models and no single-model architecture forced on you.
The second direction is the "Salesforce in Claude" plugin: 37 ready-made sales skills, such as meeting prep, deal health review and pipeline review, that read live CRM data and act within the user's permissions. Today it is for selected pilot customers, with open beta expected in September 2026.
In between sits Slack, which for Salesforce is where this happens: Slackbot, Claude Tag and Slack Code default to Claude, and the Salesforce-connected version of Claude Tag is in public beta.
What it is not
- Not a merger or an acquisition: two separate companies with a commercial and technical agreement.
- Not a product with a list price: you do not buy "Claudeforce", you buy Salesforce licences and Claude inference.
- Not an obligation: agents built on another model keep working, and the model choice in Agent Builder stays yours.
- Not finished: the most interesting part, the plugin inside Claude, is still in pilot, and skills beyond sales are announced for late 2026.
What ships and when
Status as of 21 September 2026, from official communications. Beta dates are expected, not guaranteed.
Claude in the Agentforce Atlas Reasoning Engine
Available
Agentforce Vibes and Coworker defaulting to Claude
Available, no configuration on your side
Claude selectable in Agent Builder
Available, next to the Amazon Nova models
"Salesforce in Claude" plugin, 37 sales skills
Pilot customers; open beta expected September 2026
Salesforce-connected Claude Tag in Slack
Public beta
Hosted MCP Server (Headless 360)
Beta since July 2026, requires API v67 or later
Skills beyond sales
Announced for late 2026
Winter '27 release (API v68)
Production on 4 September, 2 and 9 October 2026: overlaps with the beta
What changes for you, depending on who you are
If you are a company that simply uses Salesforce
In the short term, little: nothing to buy, install or migrate. What changes is the quality of reasoning in any agents you switch on, and the fact that sellers can work on the CRM from inside Claude and Slack without opening Salesforce. If you have no agents yet, this is the moment a first use case starts to make sense.
If you administer the org
The plugin uses a single org-level connection, with access scoped to each user's permissions. That makes permissions the thing to check before any pilot: a profile that is too wide today is one more thing the agent sees tomorrow. Validation rules, Flows and triggers apply to agent writes exactly as to manual ones, so the overlapping automations you tolerate now become visible errors.
If you already have Agentforce
In Vibes and Coworker Claude is already the default, with nothing to do. In Agent Builder you can pick Claude for new or existing agents and measure the difference on real conversations before changing everything. Usage costs follow the same rules as before: credits or conversations, plus Claude inference.
If your team lives in Slack
Salesforce-connected Claude Tag lets you query and update the CRM from a channel. The point to watch is security: Claude Tag works with a dedicated integration user, while hosted MCP servers use each user's own access. Those are two different postures, and the marketing does not separate them.
What it costs
The press release gives no prices. What the technical communications say: the model is consumption-based, tied to calls rather than seats, and Claude inference is contracted separately with Anthropic. Salesforce licences and Agentforce usage, credits or conversations, stay as explained in the Agentforce cost guide.
The part that depends on me is unchanged: a first Agentforce use case starts at a fixed €1,400 and includes the model choice, guardrails, tests on real conversations and a usage estimate at your volumes. Before committing the whole org, measure a pilot group: it is the only way to know what you will actually spend.
What is worth doing now
Permission review
Before any pilot with the plugin: who sees what, and who can change what. It is the work that makes everything else safe, and it usually takes half a day.
One use case, measured
A process that repeats every day, a number attached to it, Claude as the model, a few weeks of real conversations. Then decide whether to scale.
Sequence with Winter '27
The October release and the plugin beta land together: upgrade the sandbox first, test your agents on the new version, and only then open the pilot.
Rules for Slack
Decide now who may connect Claude to Salesforce and with which user. Do not let every team set up its own connection: that is shadow IT with better branding.
Frequently asked questions
Is Claudeforce a merger between Salesforce and Anthropic?
No. It is a strategic partnership announced on 26 August 2026: Salesforce adopts Claude as the default reasoning model in Agentforce and brings its data inside Claude through a plugin. The two companies remain separate.
Do I have to change anything in my org?
If you do not use agents, no. If you use Agentforce Vibes or Coworker, Claude is already the default with no configuration. If you want the "Salesforce in Claude" plugin, wait for the open beta and do a permission review first.
Am I forced to use Claude?
No. Agent Builder keeps the model choice, with Claude next to the Amazon Nova models. Existing agents keep working as before.
Does my data leave Salesforce?
According to the official communications Claude is served through Amazon Bedrock inside the Salesforce Trust Boundary, with traffic contained in Salesforce's virtual private cloud. It remains your job to control what a user, and therefore an agent acting for them, can see.
What does Claudeforce cost?
It has no price of its own: you pay Salesforce licences, Agentforce usage by credits or conversation, and Claude inference on consumption, contracted separately with Anthropic. Implementing a first use case with me starts at a fixed €1,400.
When can I use the Salesforce in Claude plugin?
Today it is available to selected pilot customers; open beta is expected in September 2026 and skills beyond sales are announced for late 2026. These are dates communicated by Salesforce, not guarantees.