Salesforce says AI needs real data and rules to run businesses, LLMs alone are not reliable enough at scale

Salesforce

Salesforce, one of the world’s biggest enterprise software companies, has been rethinking its reliance on large language models after facing reliability challenges when customers began using AI agents at scale. A report by The Information had earlier revealed that executives inside Salesforce were less confident in open-ended generative AI than they were a year ago, especially after AI models behaved unpredictably in real business environments.

Now, Salesforce has responded with a clear official statement backing its current AI direction. A company spokesperson said, "While LLMs are amazing, they can’t run your business by themselves. Companies need to connect AI to accurate data, business logic, and governance to turn the raw intelligence that LLMs provide into trusted, predictable outcomes. That’s why we built Agentforce: trusted AI infrastructure that drives real business value. We ground AI in tight guardrails and deterministic frameworks, optimizing LLMs to deliver enterprise-grade reliability. Trusted. Reliable. Secure. This is what AI is meant to be."

The statement confirms what had already been emerging inside the company a strategic shift toward deterministic automation in Agentforce, Salesforce’s flagship AI agent platform. The goal is simple: reduce randomness and make AI agents dependable enough for businesses that can’t afford mistakes.

During internal testing and real deployments, Salesforce found that AI agents sometimes stumble when given complex instructions. Muralidhar Krishnaprasad, CTO of Agentforce, had previously pointed out that when agents receive more than eight instructions at once, they occasionally start skipping steps. Another issue known internally as AI drift showed chatbots losing focus when users asked unrelated questions, pulling them away from the original task.

Real world use cases made the problem even more obvious. Vivint, a home security company that uses Agentforce to handle customer support for millions of users, discovered that its AI assistant occasionally failed to send customer satisfaction surveys even when the instructions were clearly defined. The company then worked with Salesforce to introduce deterministic triggers that ensured surveys were delivered every single time, removing the uncertainty of AI remembering to complete the step on its own.

Salesforce CEO Marc Benioff, once one of the loudest voices championing generative AI, has also been emphasising a more grounded approach lately. In a recent podcast, he admitted that the company has redeployed employees to other roles after implementing AI agents, but also stressed that AI cannot replace strong data foundations. Without accurate structured data and governance, AI can hallucinate, drift, or behave inconsistently something no enterprise can afford in mission-critical software.

Even as Salesforce expects Agentforce to generate over $500 million annually, the company’s stock has dipped about 34 percent from its December 2024 peak. The fall, however, isn’t slowing its AI ambition, it’s reshaping it. Instead of pushing AI that sounds intelligent but fails unpredictably, Salesforce wants AI that works with real business logic, real data, and strict safety rails. Its latest statement makes that stance even sharper: AI is not just about intelligence, it’s about trust.