Databricks Genie Has Two Costs. Most Teams Only Track One.

AI usage and triggered compute land on your bill separately. Here's how to see both.

Genie One, Genie Agents, and Genie Code each work differently, and each hits your Databricks bill differently. LLM usage runs through Genie budgets. Everything Genie triggers to actually answer a question runs through regular compute, and that part gets hard to trace back once it's mixed in with everything else. This guide covers how the Genie family fits together, why grounding changes the answers you get, and how to set up attribution before the bill arrives instead of after.

See Where Your Genie Spend Is Going

Fill in a few details to get instant access.

Written by

  • Olivier Soucy, Founder & Principal Data Engineer, okube.ai

    Olivier is the Founder and Principal Data Engineer at okube.ai, an official Databricks Partner. He's spent his career building data systems at scale, most recently as Head of Data at Newton and across 13 years at CAE.

One Genie cost shows up in the budget. The other shows up in the bill.

TWO COST STREAMS

LLM usage runs through Genie budgets. Compute runs through your regular Databricks bill, and tracing it back falls on you.

SERVICE PRINCIPALS

Automated Genie usage skips the user free-tier threshold and can scale fast without anyone noticing.

GROUNDING

Same question, same data: Genie got it wrong until the missing context got encoded into an Agent.

ATTRIBUTION

A dedicated warehouse per Genie workload turns "reconstruct usage after the fact" into "read it straight off the bill."

BUILT FOR THE PEOPLE WHO OWN DATABRICKS SPEND

Platform & Data Engineering Leads

Genie compute lands in the same warehouses as everything else by default, so a spike is invisible until it shows up in the bill with nothing pointing back to Genie.

Data & Platform Leaders

Service principals don't get the same free-tier protection users do, so automated Genie usage can scale past expectations before anyone notices.