.webp)
Salesforce has unveiled a series of new artificial intelligence (AI) models aimed at expediting the training and deployment of autonomous AI agents to streamline workflows and achieve business goals more swiftly.
The newly announced AI models include xGen-Sales, a proprietary model specifically designed to facilitate autonomous sales tasks through Agentforce, and xLAM, a new family of Large Action Models (LAMs) geared towards handling complex tasks and generating actionable outputs. Developed by Salesforce AI Research, these models promise to assist customers in swiftly setting up and deploying autonomous AI agents to enhance operational efficiency.
Silvio Savarese, Chief Scientist at Salesforce AI Research, elaborated on the benefits of the new models: "Building and training your own AI models can be time-consuming, costly, and incredibly frustrating. With Agentforce, we're able to deliver appropriately sized models, built specifically for your business with your data to drive outcomes."
The xGen-Sales model has been designed to enhance the capabilities of Agentforce by automating various sales tasks, including generating customer insights, enriching contact lists, summarising calls, and tracking the sales pipeline. According to Salesforce, this model has already surpassed the performance of other larger models in their evaluations. By fine-tuning xGen-Sales for industry-specific tasks, it aims to provide more precise and rapid responses, thereby allowing sales agents to autonomously nurture pipelines and coach representatives with greater accuracy and speed.
xGen-Sales is considered a precursor to the next generation of language models known as Large Action Models (LAMs). Unlike Large Language Models (LLMs) that predominantly generate content and require significant human intervention, LAMs focus on function-calling capabilities. This distinction enables AI agents to perform tasks independently by triggering necessary actions within other systems and applications.
In addition to xGen-Sales, Salesforce AI Research has introduced a new family of LAMs called xLAM. These models are designed to offer lower costs, faster performance, and greater accuracy compared to many of the larger, more complex models available today. For instance, xLAM-1B, despite having just 1 billion parameters, has outperformed larger and more expensive models. While xLAM-1B serves as a non-commercial, open-source model to aid research, Salesforce utilises a more advanced version for Agentforce.
The xLAM family includes four models: Tiny (xLAM-1B), Small (xLAM-7B), Medium (xLAM-8x7B), and Large (xLAM-8x22B). The smallest model, xLAM-1B, is designed for on-device applications where larger models are impractical, making it suitable for creating responsive AI assistants for smartphones and other devices with limited computing resources. The other models scale up in parameter size and capability, with the xLAM-8x22B model allowing organisations with significant computational resources to achieve optimum performance.
Rena Bhattacharyya, Chief Analyst and Practice Lead, Enterprise Technology & Services at GlobalData, commented on the significance of the open-sourcing of LAM models: "Salesforce's 'Tiny Giant' xLAM-1B exemplifies how advanced, small, action-oriented AI can revolutionise business efficiency and innovation, making high-performance AI accessible to a broader range of companies. Salesforce continues to be a leader in accelerating AI adoption across sectors."
MaryAnn Patel, SVP of Product Management at Salesforce, also shared her vision for the future: "We envision a future in which sellers are augmented by AI to help them drive selling efficiency, freeing up precious time to focus on their customers. The xGen-Sales model is purpose-built to help companies build generative AI solutions that will augment the work of their sales teams with Agentforce."
