Asset Administration Shell meets agent: How AAS and MHS fit together
How Anthropic’s Model Hardware Standard (MHS) fits the Asset Administration Shell: lifecycle vs. runtime, matching submodels and where the two compete.
After my article on Anthropic’s Model Hardware Standard, one question came up more than any other: And what about the Asset Administration Shell?
Fair enough. The AAS is the standard the IDTA positions as the semantic foundation for industrial AI – and now an AI company has published an interface built specifically for AI agents. At first glance, that looks like competition.
At second glance, it does not. The AAS answers the lifecycle question: What is this asset, what does the manufacturer say about it, what has happened to it? MHS answers the runtime question: How does an agent operate the device right now, and how does it stay within safe limits while doing so? Three AAS submodels – Asset Interface Description, Technical Data, Capability – map almost one to one onto what an MHS driver needs. Conversely, the agent is the first consumer for which a well-filled Asset Administration Shell pays off measurably.
The real fault line lies elsewhere: formal semantics with IRDIs versus natural-language descriptions for language models. Who takes on which role will be decided in the next twelve months – not in five years.
I published the full article on LinkedIn: Read the article on LinkedIn
You can find the first part of the series – MHS in relation to OPC UA, MQTT and MTP – here: https://www.linkedin.com/pulse/digital-twin-meets-agent-how-mhs-asset-administration-jan-wokittel-sjogf/