Signal

At Dreamforce, Salesforce announces Koa, a specialized CRM reasoning model built on Nvidia's Nemotron architecture, designed to power AI agents in Agentforce

First reported by Siliconangle ·

The signal ●●●● Compiled by AI from Siliconangle, Techmeme, Investor's Business Daily, Unite.AI, Constellation Research and 1 more
Why you might care

AI agents can now perform complex CRM tasks, reducing the need for human intervention in sales and support processes.

What happened

Salesforce has announced Koa, a specialized AI model for customer relationship management (CRM) data, developed in partnership with Nvidia. Built on Nvidia's Nemotron architecture, Koa is designed to enhance AI agents within Salesforce's Agentforce platform, enabling them to process complex, multi-step CRM workflows. Salesforce emphasized that Koa was trained using a large synthetic dataset derived from its 27 years of CRM deployment experience, rather than actual customer data. This approach aimed to simulate realistic enterprise workflows across 14 industries without compromising customer privacy. The training involved supervised fine-tuning, reinforcement learning, and group relative policy optimization to ensure accuracy in executing multi-step tool calls. Salesforce claims Koa outperforms leading general-purpose models on CRM tasks with significantly fewer errors. Koa is currently in an expanded pilot program and is slated for general availability in winter. Separately, Nvidia's Nemotron models will be integrated into Salesforce's Missionforce platform for secure, regulated sectors, with rollout expected in October.

What it means

Salesforce's introduction of Koa signifies a strategic push towards domain-specific AI models tailored for enterprise CRM operations. By leveraging Nvidia's Nemotron architecture and synthetic data for training, Salesforce aims to deliver AI agents capable of intricate reasoning and multi-step task execution within its ecosystem. This move allows for the automation of traditionally human-intensive processes, potentially boosting efficiency and accuracy in sales, lead qualification, and customer support workflows.

The emphasis on training with synthetic data, avoiding direct use of customer information, addresses critical privacy and security concerns prevalent in the CRM industry. Salesforce's approach offers a pathway for other enterprise software providers to develop specialized AI models without compromising sensitive client data, positioning them for a future where AI agents are integral to business operations.

AI-written summary. May contain errors.

AI Chips