Salesforce: AI Needs Real Data & Rules, LLMs Alone Unreliable for Business

• Salesforce emphasizes that Large Language Models (LLMs) alone are insufficient for running businesses, requiring integration with accurate data, business logic, and governance.
• The company re-evaluated its AI strategy after customers experienced reliability issues and unpredictable behavior with extensive AI agent use.
• Salesforce developed Agentforce, an AI infrastructure designed with "tight guardrails and deterministic frameworks" to ensure enterprise-grade reliability and reduce randomness.
• Internal testing and real deployments revealed challenges like AI agents faltering with complex instructions (over 8) and "AI drift" where chatbots lost focus.
• CEO Marc Benioff now advocates for a grounded AI approach, stressing that AI cannot replace strong data foundations to prevent hallucination and inconsistency.

Why It Matters: Salesforce shifts AI focus to trust, reliability, and real data integration, moving beyond standalone LLMs for enterprise success.