Ask three systems inside a large company what last quarter’s revenue was, and you will often get three different numbers. Humans deal with this the way they always have: someone senior picks the version they trust and everyone moves on. AI agents don’t do that. They pick one, act on it, and keep going.
That gap is the pitch behind Denodo Platform 9.5, the data management company’s latest release. It adds an enterprise knowledge graph and standardized metric views inside the platform’s semantic layer, so that a revenue figure or a customer KPI carries the same definition whether it’s being read by an AI agent, a BI dashboard, or an analyst building a slide at 11pm. One definition, governed at the source, rather than reconciled after the fact.
Why the timing matters
Gartner expects 15 percent of day-to-day work decisions to be made autonomously through agentic AI by 2028. Denodo’s argument, unsurprisingly, is that most enterprises are nowhere near ready for that — not because the models are weak, but because the plumbing underneath them is inconsistent. Business meaning varies between systems, metrics conflict, and governance controls range from partial to nonexistent. The agents don’t wait for any of that to be fixed.
The company is pointing particularly at the Gulf, where it says national AI strategies in the UAE and Saudi Arabia have pushed adoption ahead of the global curve. Organizations there are already handing genuine operational tasks to agents, which makes the trusted-data question less theoretical and more of a live liability.
“AI systems need to understand business context, work with trusted metrics, access live operational data, and operate within clear governance controls,” said Alberto Pan, Chief Technology Officer at Denodo.
The semantic layer land grab
Denodo is not alone here, and that’s worth saying plainly. The semantic layer has quietly become one of the most contested pieces of the enterprise AI stack, with warehouse vendors, BI platforms, and data virtualization players all arguing that their layer should be the one agents read from. Denodo’s historical angle is federation — querying data where it lives rather than copying it into yet another store — and 9.5 leans on that: live operational data, governed in place, rather than a stale replica an agent might happily cite.
The obvious catch is that a knowledge graph and standardized metrics are only as good as the people who define them. A platform can enforce that a metric has one definition; it cannot decide that the definition is the right one. Companies that never agreed on what “active customer” means will discover that 9.5 makes the disagreement explicit rather than making it disappear — which is arguably the point, though it’s a less comfortable sales narrative.
Still, the underlying observation is hard to dismiss. Enterprises spent the last two years buying AI capability. The bill for the data foundations underneath it is now coming due.
