By 2026, AI brokers will eat 10x extra enterprise information than people, however with not one of the contextual understanding that forestalls catastrophic misinterpretations.
On this presentation I shared yesterday, that is the principle argument.
Traditionally, our information pipelines have served individuals. We’ve architected complicated pipelines to ingest, filter, and remodel info in several programs of file: cloud information warehouses, safety info and occasion administration programs (SIEMs), and observability platforms.
We then interpreted these outputs and acted upon them.
However in a short time the tip shopper gained’t be individuals. So, we have to basically rethink the interface between these programs of file and their remodeled information.
Individuals thrive in ambiguity as a result of we’re nice at contextual interpretation. One VP of Gross sales mentions income, a CFO understands the demarcation between bookings, billings, GAAP income, or contracted ARR. People navigate these nuances effortlessly, machines don’t.
What occurs when your AI agent pulls “buyer acquisition price” information however doesn’t acknowledge that advertising and marketing measures it by marketing campaign spend, gross sales calculates it primarily based on AE + BDR prices, & finance contains fully-loaded worker prices?
The consequence: costly nonsense masquerading as intelligence.
To fight this disinformation, the groups that had been previously chargeable for sustaining and monitoring pipelines will develop into cultivators of a continually evolving assortment of cross-domain semantic layers that feed the questions from AI brokers through MCP or one other protocol layer.
The main query in all that is the right way to ship the semantic layer. Traditionally, it’s been troublesome to promote a semantic layer as a standalone product. Looker was profitable with its LookML language, and different firms have developed their very own question language, which to some extent has enforced a unfastened semantic layer.
The approaching years will see a serious shift as enterprises notice that their most useful digital asset isn’t their information lake or their AI fashions—it’s the semantic layer that makes these investments significant.
Software program is the enterprise of promoting promotions, and nobody has been promoted for implementing a semantic layer. Nonetheless, many individuals will likely be promoted for massively bettering the accuracy of AI programs and throughout information safety and observability.
The semantic layer is the keystone to that mission and consequently, probably the most strategic a part of any information pipeline immediately.
