Only 34% of AI agent projects reach production, review finds

The gap between pilots and production

The difference is sharpest among the small group of companies leading on production deployment. They move an average of 61% of their agent projects beyond the pilot stage, compared with 34% across companies overall. Their knowledge capabilities are also stronger, particularly in semantic knowledge.

That association does not establish that better knowledge capabilities alone cause projects to succeed. But it makes the knowledge layer a more concrete part of the deployment problem than the familiar ambition to scale AI.

Companies point to several obstacles:

The most common complaint is fragmented data: 55% of respondents say insufficient data sharing between systems is the main problem. Among production leaders, security and privacy are more prominent, cited by 72% of respondents in that group.

A knowledge layer, not just more data

Executives expect the biggest improvement in decision quality to come from strengthening the structural link between organizational data and AI agents. Experts surveyed in the review say a knowledge layer could provide that link.

Companies’ planned investments reflect the breadth of the challenge. They include data ingestion pipelines, AI-ready APIs and retrieval-augmented generation (RAG), as well as AI agents for evaluation and knowledge graphs.

Source: technologyreview.com

The numbers suggest that the production gap is not simply about getting more pilots underway. Leaders move a much larger share of projects forward, but they also face a different bottleneck: 72% cite security and privacy. I think that distinction matters. Better access to knowledge may help agents use company data, but access also creates a harder question about how to govern it. The review points to the technologies companies plan to buy; it does not say whether those investments will resolve the trade-off.