The advent of Large Language Models (LLMs) and Artificial Intelligence (AI) more in general has triggered a new sense of urgency around data as a critical asset for organizations. A steady stream of optimistic, sometimes heavily biased announcements from both hardware manufacturers and software vendors suggests that AI adoption at scale is straightforward and imminent1. This narrative often goes further, implying that entire categories of jobs will be disrupted or replaced within months2. At Agimus, we think that such claims, while attention-grabbing, are not well grounded. In practice, many early AI initiatives inside organizations are underperforming or failing to move beyond pilot stages3,4.
Crucially, these shortcomings are not due to limitations in the models themselves, nor to a lack of competence among analysts and engineers. On the contrary, we believe these professionals are destined to become central actors in the AI era if they are equipped with the right tools. The core issue lies at the level of data: most organizations operate on fragmented, poorly structured data systems that lack consistent relationships, shared definitions, and contextual meaning. Without these elements, even state of the art LLMs struggle to deliver meaningful outcomes. What is often framed as an AI problem is, in reality, a data problem5,6, one that has been accumulating for years, but is now becoming impossible to ignore as companies attempt to adopt AI at scale.
Until recently, this weakness in data remained largely hidden because enterprise tools such as ERPs, CRMs, and other transactional platforms were built around well-defined and static use cases, each with its own simplified interface and internal logic. Within these boundaries, data inconsistencies could persist without causing immediate friction, allowing companies to scale and operate effectively for decades. However, the emergence of AI fundamentally changes the expectations placed on data. Unlike traditional software, AI systems require a unified, coherent view of information across domains in order to reason, automate, and allow faster and better decision making. This shift reveals the limitations of legacy architectures, where data is fragmented across dozens of disconnected systems.
At Agimus, we are sure that a solution exists: an AI-native data access layer, an "ontology" that can reconcile these discrepancies and provide a consistent foundation for intelligent systems. Only by addressing the structure and semantics of their data can companies move beyond experimentation and begin to unlock the full potential of AI.
Why Enterprise Data Breaks Under AI
For years, organizations have operated with fragmented definitions of their most fundamental business concepts, without immediate consequences. A "customer" in a CRM system, for instance, is rarely the same object as a "customer" in an ERP: the fields differ, the identifiers do not align, and even the meaning of the term can vary depending on the team or context. These inconsistencies were tolerable as long as systems remained isolated and use cases were narrowly defined. But the moment data needs to be interpreted by AI models, or analysts need to start a project using data from different systems, these discrepancies become critical failure points.
In many organizations, a small number of experienced employees act as the informal glue holding this fragmented reality together. They carry institutional knowledge that is rarely documented: they know that "qty_cs" refers to "quantity in cases", or that a specific airline's orders arrive through an entirely different pipeline than another. This knowledge allows day-to-day operations to function somewhat well, but it does not scale. AI systems cannot infer this context on their own, and neither can developers who are new to the organization or working across unfamiliar systems. As soon as a company tries to build something programmatic that spans these boundaries, e.g. an internal tool, a cross-system analysis, or an AI agent, the limitations and inconsistencies quickly become apparent.
The scale of the problem only amplifies its impact. Large enterprises commonly operate hundreds of SaaS applications, each possibly introducing its own data model and semantic layer. The result is measurable financial loss: poor data quality alone costs organizations millions annually7, even before accounting for AI initiatives that fail to progress beyond experimentation. In this context, the current industry focus on models, tooling, and frameworks is misplaced. The central question is not which AI system to deploy, but whether the underlying data is structured, consistent, and meaningful enough to support intelligent behavior in the first place. AI's potential will remain constrained by the environment it is asked to operate in, until it can rely on a single, unified, and reliable data access point, the ontology.
Ontology: Structure That Scales
An ontology provides a structured and shared way to understand a company's data by defining core business concepts, their properties, and the relationships between them. Instead of relying on fragmented schemas tied to individual systems, it introduces a unified layer of meaning that reflects how the business actually operates. Concepts such as Customer, Order, Product, or Inventory are defined once and used consistently across all data sources. This ensures that data is not only standardized, but also interpretable in the same way by every team, application, and AI system.
This unified structure fundamentally changes how data is used in practice. Rather than manually stitching together tables and reconciling conflicting definitions, users can navigate data as a coherent system where relationships are explicit and reliable. Analyses become more trustworthy because they rely on consistent semantics and business knowledge. As a result, the time and effort typically spent on data cleaning and interpretation is significantly reduced, allowing teams to focus on higher-value work.
Importantly, an ontology does not require organizations to move or replace their existing data infrastructure. Instead, it acts as a semantic layer on top of current systems, mapping raw data into a consistent structure without disrupting underlying storage. This approach allows companies to preserve their existing investments while resolving long-standing inconsistencies. The ontology becomes the point of alignment across systems, ensuring that different sources of data can be used together without ambiguity.
This abstraction also transforms how software is developed. Developers no longer need to work directly with low-level database structures or rebuild data models for each new use case. Instead, they can interact with the ontology as a stable interface that already encodes the logic of the business. Applications are built on top of shared definitions and relationships, which reduces duplication, limits errors, and ensures consistency across projects. Because this foundation already captures the core needs of the organization, teams are also less dependent on external SaaS tools that only partially fit their requirements. Over time, this leads to lower costs, faster development cycles and systems that scale more reliably as new requirements emerge.
In practical terms, this means that different parts of the organization can operate on the same foundation without additional integration effort. A supply chain application can seamlessly connect orders, inventory, and logistics data across multiple clients and suppliers. Instead of manually assembling information from disconnected systems each time a new project begins, teams can immediately work with a complete and consistent view of their data. This allows organizations to continuously evaluate, adjust, and improve their operations on a shared and reliable foundation.
Giving AI the Context It Needs
Large language models are exceptional at coding and automation but, just like a new employee, can not understand your company's data and common practices without context. When you point an LLM at a database, it sees tables and columns but it doesn't know what they mean, which ones are related, or what "date" column is the correct one to use for a certain task. It fills in the gaps with guesses, and often those guesses are wrong. An ontology fixes this directly by giving AI a model where entities, properties, and relationships are all explicitly defined.
In addition, our MCP server connects directly to AI coding tools like Claude Code and Cursor, so that when a developer is building on Agimus' ontology, the AI already sees the full picture. It writes correct code against the data model without needing separate documentation or context files, which is how our clients end up building production applications in weeks rather than months.
The ontology is the data layer, so you don't need to design a database schema for every new application. A supply chain tool, a risk dashboard, and a forecasting model can all build on the same structured foundation using the SDK. Businesses change: new product lines, new regions, new data sources. The ontology grows with those changes. You add new entity types, new properties, new relationships, and everything that's already built on top keeps working.
Build the Foundation Now
The pattern across every company we work with is the same: fragmented data, inconsistent definitions, and teams spending more time wrangling than building. The ones that structure their data first, giving it real meaning and relationships, are the ones where AI actually works, where internal tools get shipped in weeks, and where the business compounds its advantage every quarter.
Just a few days ago, Gartner projected that semantic layers will be treated as critical infrastructure by 20308, on the same level as data platforms and cybersecurity. The industry is converging on something straightforward: AI needs structured, relationship-aware data to work properly. The companies that build that structure now will compound that advantage for years, and the ones still trying to make AI work on top of disconnected tables will keep wondering why their projects don't make it past the pilot stage.
An ontology is how you get there. Agimus is who brings you there.
References
[1] NVIDIA, "CES 2025 Keynote" — blogs.nvidia.com
[2] Dario Amodei, "The Adolescence of Technology" — darioamodei.com
[3] Gartner, "GenAI Project Failure" — gartner.com
[4] McKinsey, "The State of AI" — mckinsey.com
[5] Fortune, "Microsoft's Mustafa Suleyman: Give It 18 Months" — fortune.com
[6] Time, "Sam Altman on Superintelligence and AGI" — time.com
[7] Gartner, "Data Quality" — gartner.com
[8] Gartner, "Top Predictions for Data & Analytics 2026" — gartner.com
