Report Finds Knowledge Gaps Stall Enterprise AI Agents

Executive Summary

• Roughly a third (34%) of organizations' agentic AI projects make it into production, with legacy data systems, security and privacy concerns, and a lack of knowledge and context cited as key failure points, according to a survey of 300 data and AI executives published by MIT Technology Review Insights in partnership with Neo4j.
• Production leaders — organizations where an average of 61% of agentic projects advance beyond pilot — show stronger knowledge capabilities than the rest, especially in semantics, an advantage that tracks with their higher production rate per the same MIT Technology Review Insights report.
• Data fragmentation — the inadequate sharing of data across systems — was the most commonly cited top challenge to expanding agents' access to knowledge, named by 55% of respondents, while 72% of production leaders flagged security and privacy as a major concern according to the survey findings.
• Executives expect the biggest impact from strengthening the structural foundation between organizational data and AI agents, prioritizing retrieval technologies such as ingestion pipelines, AI-ready APIs and retrieval-augmented generation (RAG), AI evaluation agents, and knowledge graphs the report states.

Key Takeaways

• Agentic AI's bottleneck is framed less as model capability than as knowledge: the contextual understanding of what data means inside a specific organization.
• Production success is uneven. The gap between the broader 34% production rate and the 61% average among production leaders correlates with stronger semantic knowledge capabilities.
• Data fragmentation is the most widely cited structural barrier to expanding agent knowledge access, and it is distinct from the security and privacy concerns that production leaders weight more heavily.
• Planned investment is concentrating on the plumbing that connects data and agents, not on new model layers.

Why Agent Knowledge Gaps Stall Enterprise AI Agents

The report's central argument is a distinction between data and knowledge. AI systems continually amass and analyze data, yet enterprise AI agents often lack knowledge — defined in the MIT Technology Review Insights report as the understanding of what data means in the context of an individual organization. Agents need that understanding to reason about situations, make decisions, and take actions. Without sufficient knowledge, the report says, agents are prone to making flawed and unreliable decisions.

The consequences show up at the production line. A lack of knowledge is described as a major reason agentic AI use cases never make it to production, and the survey quantifies how common that outcome is: on average, only around a third (34%) of organizations' agentic AI projects make it into production. The report notes that even high-tech firms struggle with this. The named points of failure are legacy data systems, security and privacy concerns, and a lack of knowledge and context.

Competitive pressure is what makes the gap urgent rather than theoretical. Organizations need to deploy and scale more of their agentic projects to capture the efficiency gains AI promises, and falling short risks wasting investment already sunk into those projects while ceding ground to rivals putting their agents to work more effectively.

Semantic Knowledge Separates Production Leaders From Pilots

The survey's most concrete finding is a correlation between knowledge capability and production outcomes. The report defines agentic knowledge capabilities across three dimensions: semantic knowledge, episodic memory, and procedural knowledge.

A small group of production leaders — organizations where an average of 61% of agentic projects advance beyond pilot — have stronger knowledge capabilities than the rest, especially when it comes to semantics. That advantage tracks closely with their higher production rate, according to the report. The survey measured these capabilities across its sample of 300 data, AI, and other technology executives, and the report's stated purpose is threefold: to gauge organizations' agentic knowledge capabilities, to probe the challenges they face in improving access to knowledge and getting more use cases into production, and to explore the measures they are taking to overcome those challenges.

The report cautions against reading the correlation as a proven mechanism. It presents the knowledge advantage as an observed association with production rates, not as a controlled demonstration that semantic capability causes projects to ship.

Related: Retailers Retreat, Insurers Double Down and AI Moves Upstream in Health Tech

Fragmented Data Emerges as the Top Constraint on Agent Knowledge Access

Asked what limits the expansion of agents' access to knowledge, respondents pointed most often to data fragmentation — the inadequate sharing of data across systems — cited by 55% as a top challenge. That figure makes fragmentation the single most commonly named barrier in the report's findings on knowledge access.

Production leaders diverge from the broader sample on what worries them. This group is more likely to see security and privacy concerns as a major concern, cited by 72% of production leaders, according to the report. The divergence suggests the constraint set shifts as organizations move agents out of pilots and into live environments, where governance questions compete with plumbing questions for executive attention.

The report does not disaggregate the 55% figure by industry or firm size, and it does not publish a comparable fragmentation figure for production leaders alone.

For deeper context, see our Agentic AI analysis: "Oracle Unveils AI-Native Builder for Agentic Apps in Fusion Cloud".

Where Agent Knowledge Investment Is Going

Among steps that can yield higher quality agent decisions, executives expect the biggest impact to come from strengthening the structural foundation between the organization's data and its AI agents, the report finds. The experts interviewed for the report see a knowledge layer as a prime way to achieve this.

Investment priorities to boost knowledge range from pipelines to knowledge graphs. To expand agent access to knowledge, organizations will prioritize investments in retrieval technologies — including ingestion pipelines, AI-ready APIs, and retrieval-augmented generation (RAG) — in AI evaluation agents, and in knowledge graphs, according to the report. The emphasis lands on retrieval and structure rather than on model training or new foundation models, which the report does not list among the stated priorities.

MIT Tech Review AI Implementation Risks

The report identifies legacy data systems, security and privacy concerns, and a lack of knowledge and context as the key points of failure for agentic AI moving into production. Data fragmentation is named as the most commonly cited obstacle to expanding agent knowledge access, while production leaders weight security and privacy more heavily than the broader sample. The report does not quantify the cost of these failures, name specific vendors beyond the report's partner, or specify which industries or regions its 300 respondents represent. It also does not measure whether the investment priorities it lists have produced production gains; those are stated as expected priorities, not demonstrated results.

Additional coverage: Top 10 FoodTech Startups to Watch in the UAE and Saudi Arabia in 2026

Editorial independence disclosure: this article is based solely on the MIT Technology Review Insights report content described above, which was produced by MIT Technology Review's custom content arm rather than its editorial staff. Business 2.0 News has not independently verified the survey data. Source: MIT Technology Review.

What This Means for Practitioners

For CIOs and enterprise buyers, the report shifts the diagnostic question from which model to deploy toward whether the organization can supply context. The 34% production rate suggests pilot-heavy portfolios are the norm, so the practical first step is auditing where agent projects stall — data fragmentation, legacy systems, or governance review — rather than adding another agent. The knowledge-capability gap between production leaders and everyone else points to semantic layer work, retrieval pipelines, AI-ready APIs, and evaluation agents as the nearer-term spend. Treat the stated investment priorities as executive intent, not validated ROI.