Snowflake’s SAS migration agent is AI-only—and that changes the review burden

Snowflake has added a SAS migration path to its AIM Agent for Data Warehouses, giving teams a natural-language interface for assessing SAS estates, converting programs to Snowflake SQL and loading SAS dataset files. The feature is in preview and available to all Snowflake accounts, according to Snowflake’s SAS migration documentation.

What the preview actually does

The agent accepts requests such as “Assess my SAS codebase for a Snowflake migration” or “Convert this SAS program to Snowflake SQL.” Its assessment reports cover codebase complexity, volume and dependencies between programs. Its conversion flow produces Snowflake SQL and includes a validation step, while its loader can ingest .sas7bdat files into Snowflake tables. Snowflake documents all three capabilities.

That makes the preview broader than a code translator: it can help inventory an estate before conversion and move SAS datasets as well as programs. But Snowflake’s exclusions are just as important as the feature list. SAS conversion is AI-based only; it does not use the deterministic, grammar-based conversion layer available for other source systems. The SAS path also lacks integration with Snowflake’s cloud data-migration and validation framework, its testing framework and Power BI report repointing, according to the same product page.

Why the AI-only boundary matters

For SQL Server, Redshift and other supported warehouses, Snowflake describes a two-layer process: SnowConvert first performs repeatable grammar-based translation, then AI handles unresolved constructs interactively. Snowflake says correctness should be established by comparing source and Snowflake outputs rather than by compilation alone, and its main workflow can capture reusable fixes for propagation across a project. The code-conversion documentation explicitly says SAS does not use that deterministic first layer.

Practitioners should therefore treat the SAS preview as an accelerator with a larger review obligation, not as the same conversion contract offered for the core warehouse dialects. A sensible pilot should preserve representative SAS inputs and expected outputs, review generated SQL for semantic differences, and run independent comparisons before production cutover. That recommendation follows directly from the documented absence of deterministic conversion and testing-framework integration; Snowflake does not claim the preview eliminates those checks.

How teams can try it

The migration agent runs through Snowflake CoCo CLI. Snowflake’s overview lists Python, Git, CoCo CLI, a configured Snowflake connection and access to the source environment as prerequisites for the broader migration workflow. After launching cortex, users can ask the migration agent directly to assess a SAS codebase, convert a SAS program or load a SAS dataset, as shown in the SAS-specific instructions.

The preview’s useful dividing line is clear: assessment, AI conversion and dataset loading are available now, while deterministic conversion, the full migration-validation framework and integrated testing remain outside the SAS path. Teams evaluating it should measure the labor saved by generation against the manual evidence still required to prove behavioral equivalence.