Enterprise AI agents do not fail only because they lack data. According to a report based on a survey of 300 data, AI, and other technology executives, the bigger problem is often that agents do not understand what that data means inside a specific organization.
That gap matters because agentic AI is expected to reason about situations, make decisions, and take actions. When agents lack enough knowledge and context, their decisions become less reliable, and many projects never reach production.
Data alone is not enough for agentic AI
The report draws a clear distinction between data and knowledge. Data is what AI systems gather and analyze. Knowledge is the contextual understanding that tells an agent how that data should be interpreted within an enterprise.
That distinction helps explain why many agentic AI efforts stall. Organizations may have large stores of information, but if those stores are difficult for agents to understand in context, the agent cannot reliably use them for decisions or action.
The report frames agentic knowledge capabilities across three areas: semantic knowledge, episodic memory, and procedural knowledge. Together, these capabilities describe an organization's ability to give AI agents a fuller contextual view of the data they ingest.
The business pressure is direct. Companies want to capture the efficiency gains promised by AI, but projects that remain stuck in pilots can waste investment. The report also warns that organizations that do not scale agentic projects risk falling behind rivals that are already putting agents to work more effectively.
Most projects still do not reach production
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 transition.
The main points of failure are not framed as a single technical issue. Instead, several weaknesses appear together:
- Legacy data systems that make useful information harder to connect and apply.
- Security and privacy concerns that limit how agents can access knowledge.
- A lack of knowledge and context that leads to weaker decisions.
This helps explain why the production problem is difficult to solve with model capability alone. An enterprise agent may be able to process data, but it still needs the organizational context that makes the data usable.
The report's central implication is that production readiness depends on the structure around the agent as much as the agent itself. Without a stronger connection between enterprise data and AI agents, use cases can remain trapped before deployment.
Production leaders show a different pattern
The report identifies a small group of production leaders: organizations where on average 61% of agentic projects advance beyond pilot. These organizations have stronger knowledge capabilities than the rest.
The difference is especially clear in semantics. That matters because semantic knowledge helps connect data to meaning, and the report says this advantage tracks closely with the leaders' higher production rate.
This does not mean the report presents one simple fix. It does show a relationship between stronger agentic knowledge capabilities and better movement from pilot to production. In practical terms, enterprises that make data more meaningful to agents appear better positioned to scale agentic AI.
The finding also shifts the question leaders should ask. Instead of focusing only on whether an agent can access data, organizations need to ask whether the agent can access knowledge that is structured enough to support reliable decisions.
Fragmented data is the biggest knowledge barrier
Data fragmentation is the most commonly cited top challenge to expanding agents' access to knowledge. The report defines this as inadequate sharing of data across systems, and says it was cited by 55%.
Fragmentation makes it harder for agents to build a complete view of what is happening across an organization. If relevant data is split across systems and not shared effectively, the agent's understanding is incomplete before it begins to reason or act.
Production leaders report a different pressure point. Among that group, security and privacy concerns are more likely to be seen as a major concern, cited by 72% of this group.
That contrast is important. For many organizations, the first problem is gaining better access to knowledge across fragmented systems. For production leaders, the challenge appears to move toward managing access responsibly as agentic AI becomes more operational.
Where companies plan to invest next
The report says executives expect the biggest impact on agent decision quality to come from strengthening the structural foundation between an organization's data and its AI agents. Experts interviewed for the report see a knowledge layer as a prime way to achieve that.
Investment priorities are broad but connected. Organizations plan to focus on retrieval technologies, including ingestion pipelines, AI-ready APIs, and retrieval-augmented generation (RAG). They also plan to invest in AI evaluation agents and knowledge graphs.
These priorities all point toward the same goal: making enterprise knowledge easier for agents to access, interpret, and use. The work is not simply about adding more information. It is about building the connective structure that lets agents use information with context.
For enterprise AI leaders, the message is practical. Agentic AI will be judged by whether it can move beyond pilot projects and make dependable decisions in production. According to the report, that depends heavily on whether organizations can close the gap between the data they have and the knowledge their agents need.