What Is Salesforce AIforce?

Salesforce introduced AIforce at Dreamforce 2026 as a “live interface layer” for bringing the governed knowledge already held in Salesforce to wherever employees and AI agents work. In practical terms, it is Salesforce’s attempt to make CRM data, workflows, business logic, semantics, permissions, security, and governed actions available through interfaces such as Anthropic Claude, Slack, and Salesforce Lightning—not only through traditional Salesforce pages. 11

The central idea: Salesforce becomes the governed back end

AIforce reflects a strategic shift from Salesforce as the destination application to Salesforce as an underlying platform and control layer. Salesforce groups this approach under Headless 360: interfaces can change, but the company wants the CRM’s records, rules, workflows, and approved actions to remain connected to Salesforce. 611

That distinction matters for enterprise AI. An assistant may be the visible interface, but Salesforce says requests and actions remain grounded in the Salesforce environment. Rather than creating a separate permissions model for every new agent interface, the objective is to apply the organization’s existing Salesforce access controls and business logic when someone—or an agent acting for them—asks for information or initiates work. 1117

Salesforce describes AIforce as reaching agents through MCP servers, APIs, and CLI tools. The promise is to reduce the amount of custom integration required to give an AI interface access to business context and governed actions. 17

The three AIforce front ends

AIforce launched with three primary experiences, each aimed at a different place where employees may want to work. 11

Claudeforce: Salesforce inside Anthropic Claude

Claudeforce is the Salesforce-Anthropic integration. Its initial product, Salesforce in Claude, is a plug-in that gives Claude access to Salesforce capabilities through a prebuilt MCP-based connection rather than requiring each customer to build a connector from scratch. 171

Salesforce says the plug-in starts with 37 prebuilt sales skills. The cited examples include reasoning over live revenue context, automating pipeline updates, and taking governed actions from within Claude. That can translate into work such as preparing for a meeting, reviewing a pipeline, researching an account, or updating a record—provided the connected user has the appropriate Salesforce access. 17

The important limitation is that Claude is not meant to become an unrestricted export of a company’s CRM. AIforce’s value proposition depends on Salesforce continuing to supply the relevant context, permissions, workflows, and action controls. 1117

Slackforce: Salesforce context and actions in collaboration

Slackforce brings Salesforce-aware work into Slack, including Slackbot and Slack CRM experiences. Salesforce and industry reporting describe Slackforce Surfaces as a capability for richer, interactive experiences in Slack, allowing teams to work with live business context rather than treating a Slack message as a static notification. 611

The related Slack CRM experience is intended to let users view and manage Salesforce records from Slack. The broader goal is to let a team discuss an opportunity, inspect relevant CRM information, and take an approved action without repeatedly switching back to the core CRM interface. 111

Agentforce Coworker: the Lightning-native option

Agentforce Coworker is the AIforce experience for users who prefer to stay inside Salesforce. It operates in the Lightning environment and provides a conversational route to Salesforce context and actions. 911

Together, the three products illustrate the same architecture in different places: Claude for an external AI assistant, Slack for collaborative work, and Lightning for the native Salesforce experience.

How AIforce approaches governance and data protection

AIforce is designed around the claim that access should remain governed even when the user interface sits outside Salesforce. Salesforce says the layer carries its existing enterprise knowledge—including permissions, security, governance, data, workflows, and business logic—to AI interfaces. 11

For organizations evaluating this model, the practical questions are more specific than whether an AI integration exists:

• Which Salesforce objects, fields, and actions can a given user or agent access?
• Which workflows and approval requirements still apply before an action is completed?
• How is the external AI provider configured, including identity, data handling, region, logging, and retention?
• Which prebuilt skills are enabled, and which actions require additional review?

Salesforce has previously described a zero-retention policy for data sent across supported model interactions, stating that the data is not stored outside Salesforce or used to train models. Customers should still validate the contractual terms and technical configuration of each connected provider and deployment before using sensitive data in production. 27

Koa: Salesforce and NVIDIA’s CRM reasoning model

Alongside AIforce, Salesforce and NVIDIA announced Koa, Salesforce’s first CRM reasoning model for Agentforce. Koa is built on NVIDIA Nemotron and is intended to help agents reason through complex, multistep CRM workflows and select the appropriate tools. 32

Salesforce says Koa was created by post-training NVIDIA Nemotron 3 Super on a proprietary synthetic dataset modeled on enterprise knowledge accumulated from decades of CRM deployments, rather than on customer data. 32

Salesforce has also made performance claims for Koa in its own CRM benchmark. Those results should be read as vendor-reported benchmark results, not as an independently established measure of performance across every company’s Salesforce setup. 32

What AIforce means for Salesforce customers

The most consequential part of AIforce is architectural rather than cosmetic. Salesforce is arguing that the interface employees use—Claude, Slack, Lightning, or a future agentic tool—does not need to be the system that owns customer context, policy, and action execution.

For teams, that could reduce context switching and make CRM work available in more natural workflows. For administrators, it raises a parallel imperative: identity, permissions, skills, workflow controls, and provider settings become even more important because the CRM can now be reached from more places.

AIforce’s initial scope is centered on Claudeforce, Slackforce, and Agentforce Coworker, with Salesforce in Claude launching with 37 prebuilt sales skills. Availability, packaging, and feature scope should be confirmed directly with Salesforce because the announcement combines launch products with an expanding platform direction. 1117