Xebia: AI Agents Fail Without Proper Data Foundation

Nils Zeilemacher, global technical director of consulting firm Xebia, stated in a comment to AI News portal that most failed implementations of AI agents in companies are explained not by the weakness of the models themselves, but by the lack of a prepared data foundation. According to him, agentic AI scales only as strongly as the data on which the agents rely: if an organization has not thought through this issue in advance, the project risks getting stuck long before agents begin to accelerate the processes for which they were implemented.

Why Data Is the Foundation, Not a Detail

Unlike ordinary software that executes a pre-written scenario, an autonomous AI agent decides for itself which sources to access, what data to read, and what action to take next. If corporate information is scattered across disparate systems, unmarked, and inaccessible through clear interfaces, the agent either won't find the necessary information or will receive it in distorted form and make an erroneous action. Zeilemacher emphasizes: the quality of an agent's work is directly limited by the quality and availability of data it has access to — no amount of model power can compensate for this. This fundamentally distinguishes agent systems from classical pilots with generative AI, where model errors rarely lead to real consequences outside the chat.

Similar logic has already manifested in the wave of generative AI projects in recent years: companies massively launched chatbots on internal data, then discovered that answers were either inaccurate or incomplete because the model didn't have access to current versions of documents or couldn't distinguish an outdated instruction from an active one. For agents, the stakes are higher because an agent doesn't just output text — it takes actions: sends emails, updates CRM records, initiates payments. An error caused by poor data in an agent scenario converts not to an inaccurate answer but to a real, potentially costly action.

What Stands Behind Xebia's Position

Xebia is an international IT consulting company that helps corporations with digital transformation, data building, and implementing AI in production processes. Zeilemacher's formulation is directed at those responsible for accelerating business processes using AI agents: according to him, you should start with the foundation — make data suitable for consumption by agents — and only then move to scaling the implementation itself. Such a sequence sounds obvious, but in practice it is precisely this that is most often violated: companies purchase licenses for agent platforms before organizing the sources of data that these agents should use.

Key facts:

• Nils Zeilemacher is the global CTO of Xebia
• Comment published by AI News portal (artificialintelligence-news.com)
• Central thesis: agentic AI scales on data strength, not only on model power
• Xebia specializes in data consulting and digital transformation for corporate clients

What This Changes for Teams Implementing AI Agents

The practical conclusion from Xebia's position is simple: before scaling an agent fleet, it's worth auditing data sources — understanding where they are stored, how current they are, who manages them, and whether it's safe to give an agent access to them via API. Companies that skip this stage and go straight to pilots with impressive demo scenarios most often discover that the agent works in a sandbox but breaks on real, incomplete, or contradictory corporate data. Investments in data preparation — cataloging sources, quality control, access rights management — pay off precisely at the scaling stage, not at the prototype stage.

For technical leaders, this means reviewing budget priorities: some of the funds that previously went exclusively to subscriptions for models and agent frameworks should be redistributed to data engineering — building pipelines that ensure agents have access to current and verified data in real time. Without this step, even the most advanced model will make decisions blindly, and agents themselves risk becoming an expensive demonstration of capabilities rather than a working tool that actually accelerates company processes.