Most AI Agents Fail at Work Due to Missing Knowledge

Companies are struggling to get AI agents (software that can act on its own) to work in real business settings. A survey of 300 tech executives found that only about one-third of AI projects make it into production. The main reason? AI agents don't have enough knowledge about the company's data and processes. They can't reason well without that context.

The report says that companies with better 'knowledge capabilities' — meaning they help AI understand the meaning of data — are more successful. For example, production leaders (where 61% of projects succeed) focus on semantics, which is about understanding relationships between data. But most companies struggle with fragmented data, where information is scattered across different systems. That makes it hard for AI to access what it needs.

To fix this, companies plan to invest in tools like knowledge graphs (which map how data connects) and retrieval-augmented generation (RAG, which lets AI look things up). They also want better data pipelines and APIs. The goal is to create a 'knowledge layer' that bridges the gap between raw data and AI agents. This could help AI make smarter decisions and take actions more reliably.

If companies succeed, AI agents could handle more tasks, saving time and money. But if they fail, they risk wasting investments and falling behind competitors. The report suggests that without solving the knowledge problem, most AI projects will remain stuck in pilot mode.