Salesforce Launches AIforce and Koa CRM Model at Dreamforce 2026

Salesforce unveiled AIforce at Dreamforce 2026, creating an architectural layer that exposes live customer relationship management data, permissions, and workflows to external artificial intelligence interfaces. [2][3] This integration allows employees to access Salesforce information from third-party environments, including Anthropic's Claude, Google Gemini Enterprise, and Amazon Q, without opening the Salesforce application. [2] To support this, Salesforce released a Headless Toolkit that provides Model Context Protocol servers, APIs, and plugins for developers. [3] The system routes requests through existing access controls to ensure users only see information they are authorized to view. [3]

The company also introduced Koa, a purpose-built CRM reasoning model developed by post-training NVIDIA's Nemotron 3 Super base model. [2] Salesforce stated that Koa matches or exceeds leading models on complex CRM tasks with three times fewer errors, though this claim relies on a company-designed benchmark without independent validation. [2] A separate Bloomberg investigation found that enterprise customers reported Salesforce's AI agent platform, Agentforce, is missing the outcomes promised by the company's marketing. [2]

By launching AIforce, Salesforce is acknowledging that its competitive future may not rely entirely on the interface it spent 25 years building. As Salesforce's president of applications stated, the company's value lies in its data and metadata, which is now being exposed to new user interfaces like Claude and Slack. The parallel rollout of the Koa model highlights a strategic conviction that enterprise AI requires domain-specific training rather than generalized large language models. However, the lack of independent validation for Koa's performance metrics and the reported ROI gap for Agentforce suggest early friction in deploying these autonomous agents. What the sources don't address: How Salesforce plans to adjust its licensing or billing models when users interact with its platform entirely through third-party AI interfaces.