Data warehousing giant Snowflake is holding its annual user and partner conference, Snowflake Summit 2025, this week. As with most infrastructure software vendors, the company emphasized the proliferation of artificial intelligence (AI) across its platform.
Given Snowflake’s focus on the enterprise, with almost 12,000 customers, the pitch of all the announcements had a singular message: Business analysts, the individuals who primarily work with the Snowflake database to get work done, can be the driving force behind both developing AI models and making predictions with those models.
Also: Snowflake’s new AI agents make it easier for businesses to make sense of their data
Among the new features, ZDNET’s Sabrina Ortiz relates that the chat mode lets one speak with the data, if you will, using natural language prompts. It is powered by OpenAI and Anthropic LLMs, along with Snowflake’s own Cortex models. Data prep and analysis are made less burdensome by a new Data Science Agent feature that can automate some tasks.
(Disclosure: Ziff Davis, ZDNET’s parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
As ZDNET’s Webb Wright relates, a new service called Openflow is Snowflake’s take on the classic data integration pipeline, known by the industry rubric “extract, transform, and load,” or ETL. The Openflow functions will aid the production of AI agents, the company claims, by streamlining the complex process of cleaning up the data that has to be used by the agents.
In addition to those two announcements, the company doubled down on its efforts to make its programs where Gen AI development happens.
A function called Cortex AISQL allows business analysts to wrap the output of AI models inside standard SQL query language commands. For example, a “JOIN” command, one of the most basic ways of manipulating a relational database table, instead of being hard-coded to certain tables, can take a variable value based on what the AI model says about, say, a person’s resume in relation to open job offerings in a company.
Also: Snowflake launches Openflow to help businesses manage data in the age of AI
The company claims that this makes it easier to create complex, “multi-step” queries on data with less coding.
It also raises the profile of the business analyst, says Snowflake. “This unified approach transforms what would traditionally require data science expertise and weeks of development into straightforward SQL queries that business analysts can build and modify in minutes.” And that, it says, “turns analysts into AI developers.”
The company says that by plugging SQL into the AI “pipeline,” the tool “uplevels data analysts into AI superheroes who can work with all types of data.”
The traditional DevOps or DevSecOps domain of observability is being applied to large language models to let Snowflake customers continually evaluate how an AI model is performing relative to criteria of trustworthiness, etc.
The company says the tool has “evaluation data sets” to measure the model’s output and logging capabilities to facilitate debugging, prompt refinement, and governance.
Also: Snowflake customers eke out early gains from Gen AI applications
In a sense, Snowflake is making a statement with this tool, namely, that Gen AI training and maintenance is in some sense the province of the business analysts rather than the traditional IT folks who carry out DevOps or even AIOps.
There are innovations as well regarding the engineering of AI models, innovations that Snowflake claims make the process of building the models more tightly integrated with its tools and also broaden what can be served in production.
One is the ability to run machine learning (ML) code from a development environment or notebook context with what’s called ML Jobs inside a Snowflake container service. That means the AI model tasks of training and the rest can be spun up within the rest of Snowflake’s development work. ML Jobs is expected to be “generally available soon” on Amazon AWS and Microsoft Azure.
Also: Nvidia teams up with Snowflake for large language model AI
There’s a new way to grab the best-performing AI models during the training process, called experiment tracking, that lets a developer share and reproduce that individual model. That function is currently in a private preview.
For serving trained models, Snowflake has added to its Model Registry the ability to serve those developed and staged on Hugging Face. Says Snowflake, any model on Hugging Face can be brought into the Snowflake container service “in one click without downloading any client-side model … by just pointing to the model handle and task for logging and serving in Snowflake.”
Want more stories about AI? Sign up for Innovation, our weekly newsletter.
Leave a Reply