Snowflake today in collaboration with Salesforce, BlackRock, dbt Labs, and RelationalAI launched an open source initiative that promises to make it simpler for semantic models to interoperate.
Josh Klahr, head of data warehousing for Snowflake, said the Open Semantic Interchange (OSI) initiative will make it simpler to both share semantic models across multiple applications and migrate to another application environment without losing previous investments made in building their semantic models.
Semantic models provide the conceptual framework that defines the meaning, relationships, and context of concepts within a specific business domain. That capability enables organizations to better understand the data they collect, facilitate more consistent analysis and enable integration across different systems. The overall goal is to eliminate the need to rebuild a semantic model for every application deployed, said Klahr.
It’s not clear how many organizations have built semantic models but those that have are generally more agile than those that have not, noted Klahr. The challenge is that building and maintaining a semantic model takes time and effort and can also be difficult to integrate with third-party applications. The OSI initiative addresses that issue by providing a vendor-neutral semantic model specification that standardizes how semantic metadata is defined and shared.
That capability is going to be especially critical as organizations begin to deploy artificial intelligence (AI) agents that will be able to more easily interoperate if there is standard semantic model specification, noted Klahr. In effect, the rise of AI has elevated semantic model interoperability to the point where it is now a pressing issue, said Klahr.
In all, 17 providers of IT platforms are now pledging to support OSI and eventually governance of the project will be transferred to a consortium that will provide transparency into how the project is being governed, he added.
In the meantime, organizations that hope to operationalize AI agents at scale should be looking to define semantic models that align with the terminology used to describe business workflows in not just their own vertical industry sectors but also adjacent ones. An AI agent developed by a pharmaceutical company, for example, is eventually going to have to interoperate with an AI agent that has been trained to automate IT tasks. If semantic models are in place, the amount of time and effort required to automate workflows on an end-to-end basis will be considerably reduced. Hopefully, there might even come a day when AI agents themselves might be used to help build and maintain those semantic models.
Regardless of how semantic models are constructed, the important thing is to balance rigor and flexibility. Constantly changing the way workflows are described will defeat the purpose of building a semantic model in the first place. At the same time, organizations do need to be able to flexibly extend semantic models as additional processes are extended and added. The challenge, as always, is achieving that goal in a way that maintains a level of consistency that both humans and AI agents naturally crave.
