Snowflake releases studio to build AI functions from plain-language requests
Cortex AI Function Studio combines task creation with automated evaluation, using configurable metrics to tune prompts, models and workflow strategies.
Listen to this story
The audio brief
Story brief
3 key pointsSnowflake made Cortex AI Function Studio generally available on November 16, 2026, giving teams a way to create and tune AI functions around specific tasks and their data. The studio automates comparisons of prompts, models and workflow strategies using configurable evaluation metrics, aiming to reduce experimentation while improving task-specific results. It is not a universal configuration, and media workflows still depend on the selected model’s capabilities.
- 01
Users can describe a task through Cortex Code CLI or the guided, no-code experience in Snowsight AI Studio.
- 02
The studio supports workflows involving text, documents, images, audio and video, subject to the selected model’s capabilities.
- 03
Snowflake lists enterprise uses including document extraction, summarization, sentiment analysis, classification, compliance and structured generation.
Snowflake users can now describe an AI task in everyday language and use automated tools to test and refine how it runs. The company made Cortex AI Function Studio generally available on November 16, 2026, bringing function creation, prompt tuning and model selection into a workflow designed for enterprise data tasks.
Two ways to describe the task
The starting point is a description of the work a user wants an AI Function to perform. Snowflake supports that natural-language creation through Cortex Code CLI, its command-line interface, and through Snowsight AI Studio. The release provides an engineering-oriented route alongside a guided, no-code experience for analysts.
Testing the configuration against the task
The automation covers prompt engineering—the instructions given to a model—as well as model selection, benchmarking and optimization. Snowflake says the system can compare prompts, models and workflow strategies against a customer’s particular task and data, rather than rely on a generic configuration.
That comparison uses configurable evaluation metrics and managed optimization workflows. Prompts are only one part of the process: the studio also evaluates which model and workflow strategy to use. Snowflake positions this task-specific approach as a way to improve quality and reduce the work involved in experimentation.
Media support depends on the model
The studio supports text, documents, images, audio and video workflows. That list comes with an important qualification: support depends on the capabilities of the model being used. The studio’s media coverage is not a promise that every supported model can handle every input type.
Snowflake’s intended enterprise uses include extracting information from documents, summarizing material, analyzing sentiment and classifying content. It also lists structured generation, compliance workflows and reasoning pipelines—placing the release’s focus on repeatable business tasks that teams can create, evaluate and tune.
Sources
- docs.snowflake.comNov 16, 2026: Create, evaluate, and optimize Cortex AI Functions with Cortex AI Function Studio ( General availability - Snowflake Documentation
Reader comments
Newest comments first. Replies stay oldest first.