Why AI agents fail without a robust data foundation

Organizations rushing to deploy AI agents often overlook the most critical prerequisite: a clean, well-documented data foundation. According to Niels Zeilemaker, global CTO at Xebia, even the most sophisticated agent will falter if it lacks accurate, accessible data, turning potential innovation into a source of costly operational errors.

The primary challenge lies in data cataloging. When human employees encounter ambiguous data, they rely on colleagues to clarify the context. Agents, however, lack this social backchannel. They are entirely dependent on the precision of the documentation. If a description is imprecise, the agent will inevitably misinterpret the information or attempt to link incompatible fields, leading to system failure.

To bridge this gap, Xebia developed the Agentic Data Foundation (ADF). This framework extends traditional data platforms to host agents, enabling them to operate reliably in both customer-facing and internal environments. By integrating LLM-driven context into the migration process, the company claims it can compress project timelines that previously spanned up to two years into milestone-bound engagements.

Beyond data, Xebia is addressing the risks of AI-generated code through its ACE framework. While tools like 'vibe coding' allow for rapid application development, they often lack the governance required for production environments. ACE embeds AI across the software development lifecycle, aiming for a 40% boost in delivery speed and a 70% reduction in transformation costs, all while maintaining rigorous quality standards. As security concerns mount regarding AI-generated vulnerabilities, Zeilemaker points to emerging solutions like automated pull-request reviewers as the necessary evolution for maintaining control in an increasingly automated development landscape.