Why Salesforce Lets You Pick the Right AI Model for the Job
Salesforce allows customers to select the most suitable AI model for specific tasks, including its purpose-built Koa model and multiplayer AI in Slack, prioritizing flexibility and security.
Salesforce Chief Data Officer Michael Andrew highlights the company’s approach to offering multiple AI models tailored to different enterprise needs. Customers can choose between frontier, open, or custom models like Koa, which is designed for CRM workflows and trained using Salesforce’s 27 years of experience. The flexibility ensures businesses can adapt models based on cost, speed, and precision for each use case.
Andrew emphasizes that Koa operates alongside frontier models rather than replacing them, reinforcing Salesforce’s commitment to providing choice. The model is built in collaboration with NVIDIA and runs within Salesforce’s trust boundary, ensuring data security. This approach aligns with the company’s philosophy of allowing customers to evolve their AI strategy as needs change.
The discussion also covers multiplayer AI in Slack, where agents work collaboratively within shared team contexts. Unlike private agent setups, this model enables entire teams to benefit from exchanges, with agents acting as digital colleagues integrated into workflows. Security and permissions ensure agents operate within designated channels and tasks.
Andrew compares deterministic software systems to the variability of AI, noting that while software guarantees consistency, AI’s adaptability introduces unpredictability. He underscores the importance of selecting the right model for each task, balancing cost, value, and reliability, and leveraging partnerships with providers like Anthropic, OpenAI, and others.