VENDORiQ: Salesforce Delivers Next-Generation AI Models to Power Agentforce

Why It’s Important

With the simultaneous launch of these two new models, Salesforce is progressing toward the next generation of language models known as LAMs, offering a shift from the content generation delivered by large language models (LLMs) that require significant human interaction, to real-time task execution that can directly interact with business systems and execute tasks such as updating CRM entries, managing customer interactions, and even modifying complex workflows. The most powerful xLAM 8x22B model is currently ranking on the top of Gorilla Function Calling Leaderboard¹ that is used to compare different models for such tasks. This is ahead of top models from OpenAI and Anthropic that rely on prompt engineering. However, Hammer-7B (a fine-tuned model) performs better than xLAM 8x7B, thus raising doubts on the need to train a foundation model for this purpose.  

Salesforce has fully integrated xLAM into their customer relationship management (CRM) platform, which allows businesses to utilise customer data and context to enhance their support interactions. xGen-Sales is designed to integrate not only with other Salesforce applications, but with third party tools, and APIs in addition to marketing automation tools, and e-commerce platforms. The combination of the two models will enable Salesforce customers to rapidly deploy AI agents that enhance productivity, reduce costs, and operate at an unprecedented scale, potentially transforming their business operations and customer interactions.

The release by Salesforce of xGen-Sales and xLAM models developed by its own Salesforce AI research group, coupled with recent acquisitions such as Airkit.ai in 2023 and Tenyx, a startup that develops artificial intelligence-powered voice agents in September 2024, represents a significant advancement in Salesforces’ AI capabilities. These technologies are poised to enhance Salesforce’s product offerings and strengthen its competitive position in the autonomous agents AI market.

Who’s Impacted

• C-suite
• CIO, CTO 
• Sales and marketing
• Customer service process owners

What’s Next?

• It is important to recognise that Salesforce’s investment in enhancing its AI capabilities is intended to support customer interactions and not replace the need for a human workforce. Instead, these capabilities are aimed at supplementing sales and customer service employees.
• To manage the potential for the novel risks associated with the implementation of autonomous customer facing AI agents, organisations should consider the following mitigation strategies:
  • Establish clear guidelines and protocols for the deployment and operation of these agents, ensuring they operate within defined parameters to prevent unintended actions.
  • Incorporate robust monitoring and auditing systems to track the agents’ interactions and decisions, allowing for real-time oversight and quick identification of any issues. 
  • Prioritise data security and privacy by implementing strong encryption and access controls and complying with regulations governing data collection, user privacy and copyright.

Related IBRS Advisory

• AI in Enterprise Customer Service: State of the Art
• Generative AI Risks and Mitigation
• Virtual Service Desk Agent Critical Success Factors
• VENDORiQ: Salesforce Introduces New Generation of Einstein
• Every Breath You Take: The Impact of AI on Privacy
• VENDORiQ: Salesforce Launches Generative AI CRM Technology

1. Berkeley Function Calling Leaderboard V2, 2024, Berkeley CS ↩︎