A survey of 300 data AI and technology executives reveals that most agentic AI projects never reach production. Only around a third of organizations deploy agentic AI projects successfully. The median production rate across high-tech firms stands at 34 percent. Legacy data systems security concerns and insufficient knowledge context rank as the primary failure points.
Knowledge gaps stall agent deployment
Knowledge defined as the understanding of what data means within individual organizations remains the missing layer. AI agents need this context to reason about situations make decisions and take actions. Without it agents produce flawed and unreliable outcomes. The report identifies data fragmentation — inadequate sharing of data across systems — as the most commonly cited challenge at 55 percent. Enterprises recognize that connecting AI agents to enterprise knowledge requires more than raw data access. It demands structural understanding.
Production leaders show different capability patterns
A small group of production leaders — organizations where on average 61 percent of agentic projects advance beyond pilot — demonstrate stronger knowledge capabilities than the broader respondent pool. Their advantage is especially strong in semantic knowledge. This capability difference tracks directly with their higher production rates. Production leaders also differ in their concern profiles. While 55 percent of all respondents cite data fragmentation as a top challenge 72 percent of production leaders identify security and privacy concerns as major. The shift suggests that as organizations mature they focus less on connectivity and more on governance.
Investment priorities target knowledge infrastructure
Executives expect the biggest quality gains from strengthening the structural foundation between organizational data and AI agents. Among the steps identified a knowledge layer emerges as a prime approach. Investment priorities to expand agent access to knowledge include retrieval technologies such as ingestion pipelines AI-ready APIs and retrieval-augmented generation. Also prioritized are AI evaluation agents and knowledge graphs. These investments aim to close the gap between data collection and contextual understanding.
Report methodology and scope
The findings come from a survey fielded as part of an MIT Technology Review Insights report produced in partnership with Neo4j. The study assesses agentic knowledge capabilities across three domains: semantic knowledge episodic memory and procedural knowledge. It probes the challenges organizations face in improving knowledge access and getting more use cases into production. It also explores the measures organizations are taking to overcome identified obstacles. The report was researched and written by humans with AI tools limited to production processes under human oversight.