At its Dreamforce 2026 conference, Salesforce introduced AIforce, a new interface layer that lets external AI tools such as Claude and Slack call on the company's CRM data and business logic directly. The goal is to let people work through the AI agents they already use, without having to open Salesforce itself, while keeping the CRM information and permissions available to those tools.

What AIforce Actually Does

Connecting an AI agent to a business system has traditionally meant wiring it to each data source individually, often through the Model Context Protocol (MCP). AIforce flips that approach: Salesforce packages its own business logic, permission settings, and security rules, then delivers them to external AI environments such as Claude, Slack, and Amazon Quick. Each request inherits the sharing rules and permission sets a user already has in Salesforce, so companies do not need to rebuild a separate permission model for AI access.

Three Pillars of the Launch

The announcement centers on three products: Claudeforce, Slackforce, and Agentforce Coworker.

Claudeforce deepens the company's partnership with Anthropic, connecting Salesforce sales data to Claude through an MCP server. It ships with 37 prebuilt sales skills covering everything from prospecting to pipeline management, plus more than 40 skills built into a Claude Code plugin for developers. Deloitte, GitLab, and Legora are among the early pilot customers, and the product is currently in public beta.

Slackforce is built around "Slackforce Surfaces," which display account status, deal progress, and case history directly inside Slack, along with "Slack CRM," which lets users create or update Salesforce records from a chat prompt. It also lets people build dashboards or presentation materials from a text description, with no design skills required.

Agentforce Coworker is an AI teammate built into Salesforce's native Lightning interface. It reviews account records and activity history to offer suggestions, and can hand off tasks to specialized Agentforce agents already in place. Activating it takes a single step and requires no new permission setup. Salesforce says the feature passed 100,000 users within 35 days of its initial rollout, and Fulton Bank has already put it to use across more than 20 workflows with roughly 3,000 users.

Existing Permissions Carry Over

Salesforce says all three products enforce Zero Data Retention, meaning customer data is not used to train external AI models. Claudeforce adds a layer called Enterprise Frontier Safeguards, which combines Zero Data Retention with misuse monitoring. Agentforce Coworker is designed so that no data leaves Salesforce's trust boundary, operating strictly within a company's existing business rules and permission settings.

Deeper Ties with AWS and Other AI Model Makers

Alongside AIforce, Salesforce also expanded its relationship with Amazon Web Services. Amazon's AI assistant, Amazon Quick, can now pull Salesforce deal and customer data through MCP without custom integration work. AWS is also adding DevOps, security, and FinOps agents inside Slack. In addition, Agentforce customers will be able to reach models from Anthropic and NVIDIA through Amazon Bedrock while Zero Data Retention remains in effect, with OpenAI models set to follow.

Salesforce also showed off Koa, a proprietary reasoning model built on NVIDIA's Nemotron 3 Super architecture. It was trained on synthetic CRM scenarios spanning more than 14 industries using Group Relative Policy Optimization, and Salesforce says it produces three times fewer errors than rival models on its internal benchmarks. Pilot customers include 1-800Accountant, the credit union Baxter Credit Union, and Formula 1, with general availability in the US planned for winter 2026.

Salesforce Chair and CEO Marc Benioff described the moment as "an interface revolution" driven by AI, emphasizing that the new products are designed to work with a company's core systems under Zero Data Retention. NVIDIA CEO Jensen Huang also weighed in, saying every company needs AI tailored to its own knowledge and work.

Summary

Salesforce has introduced AIforce, a new layer that extends CRM data and business logic to AI tools outside of Salesforce, built around Claudeforce for Claude, Slackforce for Slack, and the in-house AI teammate Agentforce Coworker. By carrying over existing permissions and security settings while deepening ties with Amazon, NVIDIA, and Anthropic, Salesforce is blurring the line further between business software and the AI agents people use every day.