Connecting AI agents to enterprise knowledge is crucial for effective decision-making. A strong structural foundation that links data and agents is essential, particularly when it comes to semantic knowledge, episodic memory, and procedural knowledge. Without sufficient knowledge, AI agents risk flawed decisions, which can stall projects and give rivals an edge. Organizations must address data fragmentation and security concerns to scale their agentic AI projects successfully.

The findings from a survey of 300 data, AI, and technology executives reveal that only around 34% of agentic AI projects make it into production. High-tech firms struggle just as much as others, indicating that the problem is widespread. Production leaders, who have a higher rate of success (61% of projects advancing beyond pilot), attribute their success to stronger knowledge capabilities, especially in semantics. Fragmented data is the biggest challenge for most organizations, with security and privacy concerns taking precedence for leaders.

To improve access to knowledge, organizations are prioritizing investments in retrieval technologies, AI evaluation agents, and knowledge graphs. A knowledge layer is seen as key to achieving this. However, the real challenge lies in the ability of AI agents to reason and make genuinely innovative decisions, which is still a work in progress.

The report highlights the importance of connecting data and AI agents more effectively, but also acknowledges the current limitations of AI in terms of creativity and innovation. While AI agents are advancing, they are not yet capable of carrying out genuinely innovative, open-ended research. This underscores the need for ongoing development and investment in AI technology.