IBM's Clowes: feed AI governed context, not all your data

Clowes's argument starts from how enterprise data architectures evolved around changing storage, computing and application requirements. Agents add a new demand: combining information across operational systems while that information is still relevant to a decision. His answer is to move preparation and governance as close as possible to where data is first produced, rather than centralizing everything first.

The second half of the problem is context. Different teams may use the same business term to mean different things, so preparing context for an agent involves establishing the definitions that apply to its task. Clowes describes governed metrics and governed datasets as the foundation upon which the model reasons, with enterprises starting from high-value tasks and improving individual agents as they gain experience.

As agents move from human-assisted decisions to autonomous operations, Clowes says they need operational data without waiting for a person to gather it, including from sources such as oil rigs or cruise ships. He frames the goal as always being able to answer how the state of the business stands right now — the same world in which humans would make decisions.