Salesforce says customer trust in large language models has fallen over the past year, prompting the company to rely more heavily on deterministic automation inside its Agentforce product. Executives say predefined workflows improve reliability, reduce hallucinations, and lower operating costs compared to fully LLM-driven agents. Salesforce says some customers have struggled with LLM “drift,” where AI agents lose track of objectives or skip required steps when conversations go off-script.
