Agentforce
AIforce questions, answered: what it is, what it costs, and what changes for you
Is AIforce a product, is it free, does it replace Lightning, and which surfaces does it cover today. Twenty of the questions people are searching since Dreamforce 2026, each answered in plain words from Salesforce's own announcement and the coverage around it, with the unpublished parts marked as unpublished.
The word on the Dreamforce keynote screen was AIforce. Within a day the same four questions were circulating in different orders. What is it. Do we have to buy it. Does it mean the Salesforce screens go away. Where does the data go.
Salesforce’s announcement answers some of these, the Salesforce Ben coverage answers more, and several have no published answer yet. This page collects the questions people search and gives each a short answer. I wrote a longer piece on what AIforce is under the branding the week of the keynote, and this is the quick reference beside it.
What AIforce is
What is AIforce?
AIforce is the name Salesforce gave on September 15, 2026 to a live interface layer that sits above Data 360, Customer 360 and Agentforce. Its job is to make your Salesforce data, workflows, business logic and permissions available inside whatever AI tool a person is already using, rather than requiring them to open Salesforce. The announcement’s phrasing is that Salesforce comes to the user.
Is AIforce a product I can buy?
Not as a single line item. AIforce is an umbrella over things that already have names: Salesforce in Claude, Slackforce, Agentforce Coworker and the Headless Toolkit underneath them. You buy or enable those. The Dreamforce recap has the full list of what shipped.
What does “AI replaces the UI” mean?
It is the keynote’s slogan, and Patrick Stokes gave the clearer version: the value of Salesforce was never in the user interface, it was in the platform where customers encode how their business works. AIforce is the claim that a model can be the front end to that platform. It does not mean Lightning is retired, and Salesforce said nothing of the kind.
Who announced AIforce, and when?
Marc Benioff announced it in the Dreamforce keynote on September 15, 2026, with Dario Amodei of Anthropic and Jensen Huang of NVIDIA on stage. The Salesforce Ben write-up lists the sequence as AIforce, then Claudeforce, Slackforce, Agentforce Coworker and the Headless Toolkit.
What AIforce includes
What is the Headless Toolkit?
The Headless Toolkit is the architecture underneath AIforce: an open set of MCP servers, APIs, plug-ins and skills that expose Salesforce platform elements to any client. If you have read about the Headless 360 MCP server, that server is one part of the toolkit. AIforce packages those parts into surfaces you can switch on.
What are Claudeforce, Slackforce and Agentforce Coworker?
They are the three branded surfaces. Claudeforce is the Anthropic partnership, and its product is the Salesforce in Claude plugin with 37 sales skills. Slackforce is Slackbot with live Salesforce surfaces inside Slack. Agentforce Coworker is the agent in the Lightning search bar for people who stay in Salesforce.
Each has its own post: the Claudeforce questions page, Slackforce Surfaces, and Coworker.
Which surfaces does AIforce cover today?
Claude, Slack and Lightning, per the announcement, which adds that more will be announced soon. Salesforce Ben’s admin roundup reports that several MCP connections ship out of the box with more in beta, naming Slack, Microsoft Teams, Claude and ChatGPT as targets. Treat Teams and ChatGPT as connections you can make rather than branded surfaces Salesforce has named.
Does AIforce work with ChatGPT or Microsoft Teams?
The AIforce announcement does not mention either. The headless connections Salesforce Ben describes include both as destinations for MCP connections, so the plumbing exists. There is no Teamsforce or ChatGPTforce product, and I would not plan around one until Salesforce names it.
Cost and availability
Is AIforce free?
Nobody has said, because AIforce itself is not priced. The pieces are.
Coworker is available to all customers, with credit consumption that depends on a permission-set choice at setup. Slackforce Surfaces are included on the Slack Business+ and Enterprise+ plans. Salesforce in Claude has no published price. The announcement carries the standard line that pricing and packaging are subject to change.
What does AIforce cost?
For the plugin in Claude, consumption runs on two meters, your Salesforce API entitlement and your Anthropic contract, and no rate has been published. For Coworker and the agents it calls, the Flex Credit model applies as before. If you want a single number for AIforce, nobody can give you one yet.
How do I get AIforce?
By enabling the surfaces. Salesforce in Claude is now available to all customers in beta. Coworker is described as instantly available. Slackforce Surfaces are on for eligible Slack plans. The headless connections are configured by an admin as MCP connections against an External Client App. Nothing arrives as an AIforce switch.
Is AIforce available outside the United States?
The announcement says availability may vary by region and is governed by customer agreements, and gives no regional list. The surfaces have their own footprints: Slack plans are global, Salesforce in Claude is a beta with no regional statement, and Koa, which is separate, is US-only at general availability. Ask your account team for your region.
Security and permissions
Can AIforce see data a user cannot?
No. Salesforce’s statement is that every request runs on existing permissions and business rules, and every agent sees only what the person asking can see. A request from Claude or Slack is an API request as that user, so sharing rules, field-level security, validation rules and Flows all apply.
The consequence is that an over-broad profile is now reachable from a model, which is why I would review user permissions before enabling any surface.
Where does my data go with AIforce?
Salesforce says business data is used to answer the question at hand and is not retained by the model provider, and that nothing is retained outside your trust boundary. For Claude specifically, inference runs through Amazon Bedrock inside the Salesforce boundary, and Anthropic promises zero data retention on the plugin. Anthropic also advises confirming the retention setting is in effect rather than assuming it.
Does AIforce need Data 360?
Not to answer questions about CRM records. The surfaces read Salesforce data through the headless connections. Agents that ground on unstructured content or unified profiles still use Data 360 the way they did before Dreamforce, so if your agents already needed it, they still do.
Does AIforce replace Lightning?
No. Coworker lives inside Lightning, and the announcement positions Lightning as one of the three surfaces. What changes is that some work moves to Claude or Slack for people who prefer to be there. My earlier post covers the screen-or-surface decision for a given workflow.
For admins and developers
What do I have to configure to turn AIforce on?
Per surface. Coworker is a permission set and a credit decision. Salesforce in Claude is an org-level connection through an External Client App with per-user OAuth. Slackforce needs the Slack plan and the Salesforce connection. The headless MCP connections are the admin’s to create, and each one runs as the user who connects it.
Salesforce’s line is that there is no new permissions model to build, which is true and is the reason to check the old one.
How is AIforce different from Agentforce?
Agentforce is where you build and run agents that act on Salesforce data. AIforce is how people and external tools reach that platform without the Salesforce UI. Coworker is where they meet: a person asks in Lightning, Coworker routes the work to an Agentforce agent. I compared AIforce, Agentforce and Headless 360 in a separate post.
Is AIforce the same as Headless 360?
Related but not the same. Headless 360 is the set of MCP servers and APIs that expose Salesforce to any client. AIforce is Salesforce’s packaging of those into surfaces you turn on, plus the branding. If you are building your own client, you are using Headless 360. If you are enabling Claude or Slack, you are using AIforce.
What should an admin do before enabling any AIforce surface?
Three things. Review the profiles and permission sets of the people who will use it, because a model will use whatever access it is given. Decide whether each connection is opt-in per user or on for everyone. And test one surface in a preview sandbox against your real validation rules, since a write from Claude or Slack triggers them the same as a write from a page layout.
Understanding the basics
Is AIforce the same as Agentforce 360?
No. Agentforce 360 was the name Salesforce gave the whole platform in October 2025. AIforce, announced in September 2026, is the interface layer above it. The two coexist, and Salesforce also reversed the Agentforce application names back to Sales Cloud, Service Cloud and the rest at the same Dreamforce.
Do I need Agentforce to use AIforce?
Not for reading and updating records from Claude or Slack, which run through the headless connections as the user. You need Agentforce when you want an autonomous agent to do the work, and Coworker is the surface that hands work to those agents.
What is the difference between AIforce and Claudeforce?
Claudeforce is one branded surface under AIforce, covering the Anthropic partnership: Claude as a default model inside Salesforce products and Salesforce as a plugin inside Claude. AIforce is the umbrella that also covers Slackforce, Coworker and the toolkit underneath.
What should I do about AIforce right now?
Pick one surface, one team and one preview sandbox. Review permissions first, measure what people ask for and what it costs, and decide on a wider rollout from that. Turning on three surfaces for everyone because the keynote was exciting is how the first support tickets arrive.
The question underneath all of these
Every AIforce question comes back to the same fact. The model on the other end sees exactly what the user sees, and does exactly what the user may do. That is the security story and it is also the exposure. Your permission model was designed for people clicking through pages, and it is now being exercised by something that reads every field it is allowed to read.
Get that model right and AIforce is a settings change. Leave it as it is and the settings change is the easy part.
If you want the permission model reviewed and one AIforce surface piloted before your users ask for all of them, talk to us, or start with the Agentforce readiness assessment.
