Collibra CEO Felix Van de Maele takes on the AI 'hallucination tax'

Van de Maele's argument starts from a change in how enterprise AI is deployed. Two or three years ago, he said, there were people in the middle whose judgment could be counted on. That layer has gone away, and agents now take action across enterprise systems rather than only answering questions — which is why he frames governance as a runtime problem rather than a documentation and policy-setting exercise.

The pieces Collibra is pitching map onto that shift. Live Map handles context, pulling reusable relationships out of curated unstructured documents such as SharePoint files so a model does not re-parse the same data for every question, with Collibra's broader Knowledge Graph based on Neo4j supporting traversal among enterprise assets. Maestro applies agents to governance tasks traditionally done through manual stewardship, and Van de Maele said it can automate up to 80% of that work. Agent Contracts set out permitted behavior based on data sensitivity, operating context and risk, and Guardian Agents enforce them by compiling the contracts into a customer's existing AI gateways or orchestration tooling and blocking an action that does not align.

Van de Maele is candid that no single retrieval mechanism fits every task — vector stores, structured files, semantic search and graph grounded in business ontology all have a role, and the real question is routing the right context to the right mechanism at the right time. The stakes therefore seem to hinge on whether contract-based blocking holds up inside live gateways, where a false block and a missed one carry very different costs. Compliance and platform teams evaluating agent deployments may find the automation figure and the pre-action enforcement model the points worth testing first.