Connecting AI agents to enterprise knowledge

AI systems gather vast amounts of data, yet many enterprise AI agents lack essential knowledge to effectively utilize this data. Knowledge goes beyond raw information; it is the understanding of meaning within an organization's specific context. Agents require this understanding to reason, make decisions, and take actions. Without sufficient knowledge, agents are more likely to produce flawed and unreliable decisions.

Our research shows that a major reason agentic AI projects fail to reach production is due to a lack of knowledge. To address this urgent issue, organizations must deploy and scale more agentic projects to capture the efficiency gains AI can provide. Failing to do so risks wasting investments and ceding market share to competitors already effectively using agents.

This report, based on a survey of 300 data, AI, and technology executives, aims to assess organizations' abilities to provide AI agents with comprehensive knowledge—semantic, episodic, and procedural. It also explores the challenges in enhancing knowledge access and the measures organizations are taking to overcome these hurdles.

Key findings include that only about a third of agentic AI projects reach production, with high-tech firms also struggling in this area. Legacy data systems, security and privacy concerns, and a lack of knowledge are primary obstacles. Organizations with strong knowledge capabilities—particularly in semantics—have a higher success rate in advancing projects from pilot to production.

Fragmented data hampers knowledge access, cited by 55% of respondents as a top challenge, whereas security and privacy concerns are more of a concern for production leaders (72%).

To improve knowledge access and increase agent decision quality, organizations plan to invest in structural foundations between data and AI agents, such as knowledge layers, retrieval technologies (ingestion pipelines, AI-ready APIs, RAG), knowledge graphs, and evaluation agents for AI.