Salesforce and Nvidia have introduced Koa, a reasoning AI model built on the open-weight Nemotron architecture, designed to enhance enterprise tasks in sales and customer support without relying on proprietary frontline AI labs.
- Koa built from Nvidia’s Nemotron open-weight model, post-trained by Salesforce.
- Focuses on sales, marketing, and customer support with synthetic training data.
- Offers a cost-effective, token-efficient alternative to frontier AI models.
What happened
Salesforce announced Koa, its first reasoning AI model, created in collaboration with Nvidia, at the Dreamforce conference. The model is built upon Nvidia's Nemotron architecture and has been post-trained specifically for enterprise tasks related to sales, marketing, and customer support. Instead of relying on real customer data, Salesforce and Nvidia used synthetic data to simulate realistic customer interactions and sales scenarios to train the model.
Koa integrates into Salesforce’s Agentforce platform as an alternative to widely used AI models like Claude and ChatGPT. It is designed to efficiently handle multi-step reasoning tasks for common enterprise workflows such as scheduling, customer inquiry handling, and deal closing. The partnership leverages Nemotron’s open-weight, sovereign AI capabilities, ensuring transparency and compliance with data governance standards.
Why it matters
Koa's development highlights a shift in enterprise AI needs, evolving away from dependence on frontier models controlled by external AI labs towards sovereign, customized solutions. Many enterprises seek AI that can operate securely under their own data governance policies without exposing sensitive information to third parties. By training Koa with synthetic data and offering it as an integrated option, Salesforce is addressing these concerns while optimizing cost and performance.
The model’s token efficiency and tailored reasoning capabilities reduce operational costs compared to the more general-purpose AI models on the market. This advancement could reshape how businesses deploy AI agents, encouraging greater adoption of proprietary or partnered models that provide specific domain knowledge rather than generic language generation.
What to watch next
Monitor how Salesforce’s enterprise customers adopt Koa within Agentforce and whether it can deliver measurable improvements in task automation and cost reduction. Additionally, Salesforce’s continuing partnership with Anthropic to offer ClaudeForce demonstrates a hybrid approach, balancing proprietary models with frontier AI options to serve diverse customer preferences.
Further developments may include Koa being expanded into other enterprise domains beyond sales and customer support, as well as potential improvements through ongoing post-training iterations. The model may also prompt other cloud and AI providers to accelerate their efforts toward sovereign, task-specific reasoning models tailored to regulated industries and complex workflow automation.