The future of data governance in the age of AI agents

Data mesh, agentic AI, the AI Act: we pile up programmes without knowing where to start. This talk starts from a simple observation. An AI agent consumes your data, but also the meaning and the rules that come with it. And most stacks do not have that layer.

Many teams think they are ready because they already govern machine learning models. Except that an ML model is trained on a dataset, validated before going into production, monitored for drift, and produces a score that a human uses to decide. You do not train an agent. It reads definitions, rules and permissions at the moment it acts, then it acts. Its behaviour changes as soon as the context changes, without any retraining. Control therefore shifts from the model to the context, and from validation before production to control at run time.

There are now four layers to govern: data, context, agents and trust. As for roles, none disappears, each moves up a level: the data steward moves from documentation to executable definition, and the governance team becomes data and AI governance.

Who decides, who is accountable when an agent gets it wrong, which skills to amplify, which functions appear, and why you should absolutely not hire at this stage: you leave with a clear reading of how your data governance teams should evolve over the next eighteen months.