For decades, business software has depended on people defining the structure of information in advance. Databases organise information into records, fields, categories and relationships, while applications provide the means to enter, retrieve and process it.
AI changes this relationship. Large language models can interpret documents, conversations, images and other largely unstructured information without requiring everything to be organised into predefined fields.
But interpreting unstructured information is not the same as maintaining an understanding of it.
The next development may be a richer layer of information structure, created and continuously maintained with AI.
Consider an email advising that an equipment installation will be delayed because a design has not been approved. Beyond recognising the message as a schedule update, AI can identify the installation activity, supplier, revised date, outstanding approval, causal relationship and potential impact on dependent activities.
These are more than categories. They describe relationships between events, entities, processes and evidence.
A system could retain those relationships in a persistent semantic representation, updating them as new information becomes available. Crucially, it would distinguish established facts from reported claims, inferred relationships and uncertain consequences, while retaining the original evidence.
Traditional databases would remain important for authoritative records, transactions and controls. An evolving semantic layer could add a richer representation of what the information means and how it relates to everything else.
This has implications beyond information retrieval.
Today, multiple AI agents may repeatedly interpret the same documents to perform different tasks. With a shared semantic model, specialised agents could instead operate against common entities, relationships and context. A scheduling agent, risk agent and reporting agent could work from the same underlying understanding without independently reconstructing it each time.
Applications could increasingly become different ways of interacting with that shared information rather than independent systems maintaining separate interpretations of reality.
None of these ideas requires abandoning existing database technology. Knowledge graphs, ontologies and semantic information models already provide foundations. AI introduces the possibility of creating, enriching and maintaining such structures more dynamically, with less manual classification.
The challenge is making that interpretation dependable: preserving provenance, resolving contradictions, managing uncertainty and preventing inferred relationships from becoming accepted facts without appropriate validation.
The opportunity is not simply to make unstructured information searchable. It is to make its meaning persistent, structured and reusable.
The next generation of information systems may be defined less by how data is stored, and more by how its meaning is continuously maintained.