Databricks Launches ai_decide: Fast AI-Powered Decisions on Governed Data

Databricks Releases the ai_decide Decision Function

Databricks has recently launched a new AI Function — ai_decide — designed for fast, scalable decision-making directly on governed data. This function continues Databricks' approach of embedding large language model capabilities natively into SQL and data pipelines, enabling data teams to execute AI-based judgment logic inside the data platform itself, without the need to build complex external inference services.

The name says it all: ai_decide is fundamentally about decision-making, not just text generation or classification. It targets a common business pain point — how to make rule-based yet semantically aware judgments across massive volumes of structured and unstructured data in a consistent, auditable way.

Why Bring Decision-Making Into the Data Platform

In traditional architectures, business decision logic tends to be scattered across the application layer, ad hoc scripts, or manual review processes — making it difficult to scale and hard to govern consistently. Moving decision-making capabilities down into the data platform layer means the judgment process can directly inherit the platform's existing access control, data lineage, and audit mechanisms.

For enterprises that prioritize governed data, this is especially critical. Data access controls, compliance requirements, and traceability can all extend seamlessly to AI decision workflows within the same framework — preventing the governance blind spots that typically emerge when external model services are introduced.

How It Fits Within Databricks AI Functions

ai_decide isn't a standalone feature — it's the latest member of the Databricks AI Functions family. Databricks has previously released built-in AI functions for text summarization, classification, sentiment analysis, and more. ai_decide fills in the "decision" gap, allowing developers to wire AI-powered judgment into existing data workflows with something close to writing plain SQL.

Databricks AI Functions is a suite of built-in functions that invoke large language models directly at the SQL layer. Users can call AI capabilities within standard SQL queries using syntax like SELECT ai_summarize(text), powered by Databricks' managed Model Serving infrastructure. The key advantage of this design is that data never has to leave Databricks' security boundary — all calls are permission-checked through Unity Catalog and automatically tracked for data lineage.

By contrast, traditional approaches require exporting data to external API endpoints (such as OpenAI or Anthropic's public interfaces), which creates significant friction in highly regulated industries like finance and healthcare, where data residency and privacy audits are major concerns.

"Data Governance" in an enterprise context typically covers three core dimensions: access control (who can read or write which data), data lineage (where data comes from and how it's been transformed), and audit trails (when, by whom, and for what reason data was accessed or modified). Databricks' Unity Catalog serves as its unified metadata and governance layer, enforcing these policies consistently across clusters, workspaces, and cloud accounts. When an AI decision function runs as a SQL operator, every model call's inputs and outputs are naturally captured within this framework — making AI decision behavior as auditable as any ordinary data operation. This level of traceability is nearly impossible to achieve with traditional external model service approaches.

Use Cases and Value

Based on its positioning, ai_decide is well-suited for scenarios that require instant judgment during data processing — such as content moderation approvals, transaction risk tiering, or automated customer request routing. These tasks demand both semantic understanding and high throughput.

The biggest benefit of delegating such decisions to a platform-native function is reduced engineering complexity: teams no longer need to separately maintain model serving infrastructure, manage API quotas, or navigate the compliance risks of moving data outside the platform. Decision outputs also integrate naturally with downstream data tables, dashboards, and pipelines.

Conclusion

ai_decide reflects a broader trend among data platform vendors: making AI capabilities truly native — treating large language models not as external add-ons, but as standard operators within the data processing chain. For teams already using Databricks to govern their data, this means a lower barrier to incorporating AI-driven decisions into everyday workflows.

One caveat: this article is based on limited information from the official announcement. Specific performance benchmarks, pricing details, and supported model options for ai_decide should be verified against the official Databricks documentation.