Teradata’s Josh Fecteau: Why AI Agents Fail at Scale
Teradata’s Josh Fecteau explains why AI agents fail at scale, blaming weak data foundations and uncaptured “temporal truths”
TL;DR
- The biggest misconception about AI agents is that Large Language Models (LLMs) alone can solve enterprise problems.
- AI agents fail at scale because data foundations are weak, lacking metadata, lineage, entity relationships, and embedded business rules.
- The concept of 'temporal truths' refers to decisions that were correct when made but become invalid over time because the context was not captured.
- Incomplete decision records and the failure to capture evolving context lead to AI agents executing outdated or incorrect logic.
- Addressing AI agent failures requires capturing context at the moment a decision is made, assigning ownership to business rules and reasoning, and implementing review cycles.
- The cultural shift of treating context as part of the data asset, rather than residing in human knowledge, is critical.
- The cost of frontier AI models will drive a rethink of where enterprise AI runs, favoring smaller, specialized models closer to the data for specific workloads.
- AI governance is essential for ensuring AI agents align with business intent and for managing costs.