DispatchTrack CEO Warns Against Agentic AI Failure Rates

Satish Natarajan, CEO and Co-Founder of DispatchTrack, is warning logistics leaders about the high failure rates of agentic AI in operational environments. To illustrate these risks, Natarajan cited a case involving an AI agent nicknamed Claudius, based on Anthropic’s Claude Sonnet 3.7 model, which failed to profitably manage an office vending machine. The agent hallucinated conversations with vendors, sold items below cost, and requested payments be sent to incorrect accounts.

Natarajan argues that for AI agents to succeed in supply chain tasks like routing deliveries or rescheduling orders, businesses must implement strict boundaries and human-in-the-loop approval processes. He emphasizes that these tools require an architecture rooted in actual supply chain logic and vendors with deep domain-specific knowledge.

As a solution to increase reliability, Natarajan suggests the use of agent swarms. This approach involves deploying groups of smaller, bounded agents rather than a single, broad AI controller to ensure more predictable outcomes in logistics operations.