AI WORLD HQ
If your leadership wants AI results this year, but your customer, sales, and operations records live in several different systems, you may feel stuck. The usual advice is to clean up and consolidate everything first. Snowflake's announcement argues that this order no longer has to be fixed.
Why moving data became the default
For years, analyzing data meant going through the software that stored it. Cross-system work usually meant copying everything into one central place first. Every copy adds cost and another point of risk, so centralizing became the starting line for AI too.
Snowflake says that sequence made sense a decade or two ago, but it is no longer a technical requirement. In its view, the newer approach lets you run AI on the data you have now while your modernization work continues on its own schedule.
What changed: open formats
The key idea is an open table format (a shared way of organizing data files so that many different tools can read them). Snowflake points to Apache Iceberg as the example. Data stored this way can be read by any compatible compute engine (the program that does the actual calculating).
According to Snowflake, this lets you publish data products from systems such as Salesforce, SAP, and Workday and use them elsewhere without copying or building complicated pipelines (the automated routes data takes between systems). Snowflake also says the data stays in your own cloud storage, saved as Parquet files (a common, compact file format), so it remains yours.
Governance that travels with the data
Moving data was also the way companies enforced rules. Snowflake says its Horizon Catalog (its governance layer, the place where access rules are defined) can apply one set of policies across data, models, apps, and AI agents, including assets outside the platform.
Two features stand out:
- Shared policies for other tools: Snowflake says its row access and masking policies now extend to external engines such as Apache Spark, Trino, and PyIceberg, in public preview (an early-access stage). Masking hides sensitive values, and row access limits which records a person can see.
- Separate permissions for AI agents: Snowflake's Agent identity capability (an AI program acting on a task) lets administrators decide what data an agent can reach, apart from the person who started it. Actions are logged under both the user and the agent, which gives compliance teams an audit trail.
Snowflake also notes that for European enterprises, where data sits and who can access it are compliance questions tied to rules such as GDPR and the EU AI Act. Its message is that governing data in place can help meet those requirements without forcing everything into one region or system.
Wrap-up
Snowflake's core message is that AI and data modernization can run on separate timelines. Open formats let tools share data without copies, and shared governance keeps access rules consistent. Your next step is simple: pick one business question AI could help answer, identify where the data lives today, and bring that list to your next planning meeting.