Xebia urges data overhaul to speed AI agent rollouts

At the AI & Big Data Expo, Xebia CTO Niels Zeilemaker described steps firms should take to prepare data and deploy Xebia’s Agentic Data Foundation (ADF) and Xebia ACE to accelerate AI agent rollouts and legacy migrations.

Zeilemaker said the first priority for any organisation adding AI agents is to make data available and trustworthy for automated use. He warned that agents will fail if they cannot locate the right records, if datasets are mislabelled, or if fields are incorrectly joined. ‘If you don’t think about that, you can build the best agent, but it will never be able to find the correct data,’ he said.

A large part of Zeilemaker’s presentation focused on data cataloguing. He explained that human teams can call a colleague when documentation is missing, but agents cannot. Agents depend entirely on catalogue entries and accurate metadata. He urged organisations to improve catalogue descriptions and data labels before deploying agents.

Xebia’s Agentic Data Foundation aims to extend existing data platforms so agents can run against controlled data sources and be used in customer-facing products and internal processes. Zeilemaker described work that compresses typical 12- to 24-month migration and integration timelines into fixed-price, milestone-bound engagements, with consultants and clients co-developing migration paths.

Xebia ACE is an AI-native software engineering framework intended to embed AI across the software development lifecycle. Zeilemaker said ACE can shorten delivery times by up to 40% and reduce legacy transformation costs by up to 70% in larger enterprises that need to keep established governance and workflows.

He described combining LLM-assisted coding with added contextual layers inside migration workstreams. Automated coding tools were integrated into the data platform to run migrations faster while preserving oversight and control over generated code.

Security and governance remain industry concerns for AI-driven development. Zeilemaker pointed to experiments that add an LLM as an additional reviewer in pull request workflows, describing the model as a ‘senior team member’ that can perform a third-party review alongside human reviewers.

Zeilemaker also highlighted the company’s practice of sharing methods and lessons at industry events to accelerate adoption and refine approaches. He said that clear metadata, integrated platform services for agents, and governance processes for code generation are the foundation steps organisations should address before scaling agent deployments.