The issue is that these definitions make sense to individuals, not machines. A human studying a textual content definition of “buyer” can fill in the remainder from expertise: which programs prospects stay in, how they relate to orders, areas, and income. An AI can’t fill in what it was by no means given. It wants entities, properties, and relationships: a buyer is an individual, belongs to a site, and locations orders. Give an agent that graph and it could actually construct its personal mannequin of the enterprise and infer new data from it. Ask what number of prospects you serve in Europe, and it could actually purpose its strategy to a solution. Give it a definition with no construction behind it, and it’ll guess confidently as a substitute.
BI semantics assist inside a slim scope, however they’re hardly ever constructed on the open requirements that might let AI programs purpose over them.
The dream of enterprise semantics in 2026
The facility of frontier giant language fashions (LLMs) tempts individuals into pondering that merely loading enterprise knowledge into these fashions will get you near the dream of enterprise semantics. It doesn’t. LLMs nonetheless must be advised what the structured knowledge means, how enterprise ideas are outlined, and which knowledge is authoritative. They want what individuals now name context.
