A new report from MIT Technology Review Insights, based on a survey of 300 data, AI, and other technology executives, argues that enterprise AI agents often fail not for lack of data but for lack of knowledge. Knowledge, in this context, means understanding what data signifies within a specific organization. Without it, agents make unreliable decisions, and the report finds this is a major reason most agentic AI use cases never reach production. On average, only about a third (34%) of such projects advance beyond the pilot stage.
The report identifies a small group of “production leaders” where roughly 61% of agentic projects make it to production. These organizations show stronger knowledge capabilities than the rest, especially in semantic knowledge—the ability to understand meaning and relationships in data. The correlation suggests that improving knowledge access, not just adding more data, is closely tied to successful deployment.
Fragmented data is the most widely cited obstacle to expanding agents’ access to knowledge, named by 55% of executives. Production leaders, by contrast, are more likely to flag security and privacy concerns, with 72% citing them. Looking ahead, executives expect the biggest gains from strengthening the structural link between data and agents, with planned investments in retrieval pipelines, RAG, evaluation agents, and knowledge graphs.