It organizes the data AI can see. Someone still has to sense the physical world.
Operations software has always assumed a person on the other end. Dashboards, reports, and interfaces all expect someone to read them. Cory LaFleur of Thinaer presented at a recent defense industrial base (DIB) forum. He described the mismatch:
“Everything has been built for people, but now we have AI doing everything, so we’re kind of optimizing for the wrong thing.”
Cory LaFleur, Thinaer
The industry’s answer is the ontology layer, an architecture that organizes data so AI agents can use it without a person interpreting a screen first. It gives agents the context people carry without thinking about it. It also depends on something that has to come first. That something is data from the physical world, captured by a system that can sense it.
What the Ontology Layer Does
People read data in context without noticing. For example, a technician who sees a work order knows which bay it refers to, which crib stocks the parts, and who signed off the last step. An AI agent, however, has none of that background. It has the data in front of it, and that data usually comes from several systems, in different formats, with no shared meaning.
The ontology layer supplies that meaning. It takes data from many sources and organizes it around the things an operation cares about, such as assets, locations, people, and orders. It also maps the relationships between them. As a result, an agent can query that structure and act on it directly. Standards like Model Context Protocol (MCP) give agents a common way to reach it.
Cory explained why structure matters for agents in particular. A browser interface is inefficient for an agent to scan. Give the agent structured data instead, and it can work out what is happening and act on it quickly.
Why So Many Companies Are Building One
AI agents now do work that people used to do, and the systems they read from were never built for them. Defense contractors, enterprise software vendors, and data platforms are each racing to build the layer that closes that gap. The ontology layer was a major point of discussion at the forum.
The stakes rise as agents move from reading data to acting on it. An agent working from a stale or incomplete picture acts on that picture with the same confidence it would have in a complete one. The quality of what sits underneath the ontology layer sets the quality of every decision built on top of it.
These builds share an assumption, though. They assume the physical operation they describe already exists as data. In many facilities, it does not.
Someone Still Has to Sense the Physical World
“They’re all pivoting to this ontology layer … they still miss the physical layer, which is us. So it all starts with someone having to sense the physical world.”
Cory LaFleur, Thinaer
A tool sitting in a classified bay or a part waiting between workstations has no digital presence until something captures it. The physical world doesn’t log itself. An ontology layer can model that tool and every relationship it has. However, if nothing reports where the tool is right now, the model has nothing current to work from.
Thinaer describes the full stack as Capture, Learn, Act. Capture senses the physical environment and produces structured data. Learn is where models, digital twins, and the ontology layer make sense of that data. Act is where agents and autonomous systems do the work. As a result, Learn is only as good as what Capture hands it. Our post on the Capture, Learn, Act framework covers the whole stack.
Someone still has to sense the physical world.
The ontology layer lives in Learn. Learn is only as good as what Capture hands it.
Where Thinaer Fits
Thinaer is the Physical AI capture layer. We make AI work by capturing the data it runs on. That data is open and customer-owned, and it goes to any cloud, any model, and any ontology layer a customer chooses. The two layers do different jobs and fit together.
The capture side is built for hard environments. Thinaer has deployed more than 150,000 sensors across 33 locations and over 12M square feet. With 40+ sensor types available, the environment decides the technology. That includes classified bays, where Thinaer’s coverage is patent pending, and ordnance environments where Thinaer is HERO ZERO certified. Deployments are DISA approved.
For more on how a physical operation becomes usable data, read How Unstructured Data Becomes AI-Ready Data and AI Can’t Fix What It Can’t See.
What Does Ontology Mean in Palantir?
Palantir describes the Ontology in its Foundry platform as the layer that connects an organization’s datasets and models to real-world counterparts such as equipment, facilities, and orders. It also covers the relationships and actions that link them. The company publishes the details in its Foundry documentation. It is a product-specific name for the same idea described above.
The same dependency applies. An ontology layer, Palantir’s or anyone else’s, represents the operation using the data it receives. In practice, capturing that operation on the floor happens upstream.
Give the Ontology Layer Something to Work With
Bob Donnelly of Thinaer is looking for the organizations that have already noticed the gap:
“Those are the places we need to get to, the ones that understand they’re missing this piece.”
Bob Donnelly, Thinaer
People carry physical context everywhere without effort. Agents get it only from what sensors have captured. If your team is building an ontology layer or deploying agents, ask what feeds it from the physical side. That answer sets how much context the agents have. Talk with Thinaer about capturing it.