Open Pragmatics: Foundations
Your enterprise AI systems are failing for data work. The solution may already exist within your walls.
Executive Summary
Most enterprise AI transformations never make it to production, especially in operations that depend on data and require reliable outcomes. The usual explanations: Maybe it’s the wrong model, weak deployment, or data that is too messy. However, before AI, many enterprises were able to get reliable answers from their data systems, just not ad hoc nor at scale. So if the old system worked, the cleanliness of the data itself, while important, probably isn’t the core problem.
Our thesis: The infrastructure to make AI work with enterprise data exists within your workplace, it just hasn’t been fully productized yet. That infrastructure, Pragmatics, sits between your data and your data outcomes as an operating system for context. It already has a reference architecture, too: it is applied by subject matter experts every day, every time they take a question and provide a relevant answer (considering not just the dictionary meaning of a request, but also who said what, why, and in what setting).
We believe Pragmatics is what the industry is really reaching for today when it talks about needing a “context layer” to make AI actually work for the enterprise. It comes from the science of language, the same field that the data industry borrowed the term “semantic layer” from decades ago. So, if semantics is what something means, then pragmatics is that meaning in context of use. The closest attempts seek to swap the SME with an AI agent, but that’s a common anthropomorphic error. Our perspective is that the data expert today is an entire abstraction layer across social understanding, tribal knowledge, political stakeholder management, learned data instincts and judgement, execution behavior, and even how they adhere to protocols.
In “Open Pragmatics: Foundations,” we define the traits of a pragmatics system, and how it is structurally different from legacy frameworks that were built for a different technology paradigm. Perhaps most importantly, we discuss why accuracy is an incomplete measure for successful implementations and suggest an alternative, SME parity, as a more appropriate evaluation framework. We also offer a benchmark to understand where your company stands on its journey toward SME parity, and provide field notes on how conformance is evolving across the industry.