Salesforce takes aim at ‘jagged intelligence’ in push for more reliable AI

Salesforce is addressing a significant challenge in artificial intelligence for business applications: the gap between an AI system’s intelligence and its consistent performance in unpredictable enterprise environments, known as “jagged intelligence.”

In a recent research announcement, Salesforce AI Research introduced new benchmarks, models, and frameworks aimed at enhancing the intelligence, trustworthiness, and versatility of future AI agents for enterprise use. These innovations focus on improving both the capabilities and reliability of AI systems, particularly when operating as autonomous agents in complex business environments.

According to Silvio Savarese, Salesforce’s Chief Scientist and Head of AI Research, traditional AI systems may excel in standardized tasks but struggle with consistent task execution in dynamic enterprise settings. This initiative reflects Salesforce’s commitment to “Enterprise General Intelligence” (EGI), which prioritizes AI designed for business complexities over the theoretical pursuit of Artificial General Intelligence (AGI).

How Salesforce is measuring and fixing AI’s inconsistency problem in enterprise settings

The research emphasizes quantifying and addressing AI’s inconsistency in performance. Salesforce introduced the “SIMPLE dataset,” a public benchmark with 225 reasoning questions to assess the capabilities of AI systems. This focus on consistency is crucial for enterprise applications where a single misstep by an AI agent can have significant consequences.

Inside CRMArena: Salesforce’s virtual testing ground for enterprise AI agents

A notable innovation is CRMArena, a benchmarking framework designed to simulate realistic customer relationship management scenarios. This framework enables thorough testing of AI agents in professional contexts, bridging the gap between academic benchmarks and real-world business needs.

New embedding models that understand enterprise context better than ever before

One of the technical advancements highlighted by Salesforce is SFR-Embedding, a model for deeper contextual understanding that outperforms existing benchmarks across various datasets. This model will soon be integrated into the Data Cloud, offering developers efficient code search capabilities.

Why smaller, action-focused AI models may outperform larger language models for business tasks

Salesforce introduced xLAM V2, a series of models specifically designed to predict actions rather than generate text. These models, starting at 1 billion parameters, excel at predicting and executing task sequences, making them ideal for autonomous agents interacting with enterprise systems.

Enterprise AI safety: How Salesforce’s trust layer establishes guardrails for business use

To address concerns about AI safety and reliability, Salesforce introduced SFR-Guard, a family of models trained on both public and internal CRM data. These models strengthen the company’s Trust Layer, setting boundaries for AI agent behavior based on business requirements.

Co-innovation in action: How customer feedback shapes Salesforce’s enterprise AI roadmap

Customer co-innovation plays a crucial role in developing enterprise-ready AI solutions at Salesforce. By incorporating customer feedback, the company has seen significant improvements in AI performance, ensuring accuracy and relevance in enterprise data processing.

The road to Enterprise General Intelligence: What’s next for Salesforce AI

Salesforce’s research efforts align with the growing demand for AI systems that combine advanced capabilities with consistent performance in enterprise settings. By focusing on reliability and real-world business requirements, Salesforce aims to lead the way in the business AI revolution.