TL;DR
The Databricks Assistant still exists, but since March 11, 2026 it carries a new name, Genie Code. Everyday tools also carried over, including autocomplete, Diagnose error and slash commands such as /fix and /explain. Its behaviour changed most, because Genie Code now plans a task, runs the code and fixes its own errors. It can only reach data your Unity Catalog permissions allow, and an AI classifier approves its actions unless you pick Ask every time. Since July 8, 2026, each user gets 150 free DBUs a month, and usage beyond that is billed pay-as-you-go. Admins should check the approval mode, set Genie budgets and keep a review habit before usage grows.
Key Takeaways Databricks Assistant has been Genie Code since March 11, 2026, and the old documentation and product links now redirect to Genie Code pages. The tools people use daily survived the rename, including autocomplete, Quick Fix, Diagnose error, /explain, /fix, /doc, /optimize and /findTables. Genie Code now plans and runs code, and by default an AI classifier approves its actions, though you can switch your default to Ask first. It only touches what your Unity Catalog permissions allow, yet prompts, code, table names and descriptions are sent to the model as context. Billing started July 8, 2026, with 150 free DBUs per user per month and no free allowance for service principals. Teams that standardised on the Assistant should review approval mode, budgets and code review habits before usage grows. Watch on YouTube
Databricks Genie Code: Agentic AI for Enterprise Data Teams
Kanerika explains what Genie Code, the renamed Databricks Assistant, does for data teams and how its agent behaviour works inside the workspace.
Where Did the Databricks Assistant Go?
Open the old Assistant FAQ in the documentation and you land on a page titled Genie Code. The old product URL does the same thing. Yet internal wikis, training decks and community threads still talk about the Assistant, and that mismatch leaves teams with a fair question.
Was the tool they rolled out in 2024 retired, replaced or repriced? Renamed and repriced, but not retired. Databricks renamed it on March 11, 2026, gave it agent behaviour that plans and runs work, and started billing for it on July 8, 2026.
The name is the smallest part of that change. For a data team, four questions matter more. Which features stayed, what can the agent now do alone, what does it send to the model, and what does it cost? Each answer sits below, in the old name’s terms.
What Is the Databricks Assistant?
The Assistant was a context-aware AI helper built into the Databricks workspace. Databricks announced its public preview on July 31, 2023 , with support in notebooks, the SQL editor and the file editor. It then reached general availability on June 27, 2024, according to the Databricks release notes for June 2024 .
What the Assistant Was Built to Do
Its purpose was practical. First, the Assistant generated, debugged, optimized and explained code, wrote and edited SQL queries, and built visualizations. It also answered questions from the documentation and knowledge base, so a user could stay in the notebook instead of searching in another tab.
That made it a coding helper for people already working on the Databricks data intelligence platform , not a general chatbot. Databricks says the same about Genie Code today, describing it as not intended to be a general-purpose programming assistant.
Where You Will Still See the Old Name
For example, the old Assistant FAQ address in the documentation now redirects to the Genie Code page. Likewise, the former Assistant product page now serves the Genie Code product page. However, release notes from before March 2026 keep the original name, and so do older blog posts and forum answers.
From Databricks Assistant to Genie Code: A Short Timeline
The rename makes more sense once you see the steps behind it. Each date below comes from official Databricks release notes or announcements.
Table 1: Databricks Assistant to Genie Code, Dated Milestones
Date What happened July 31, 2023 The Assistant enters public preview in notebooks, the SQL editor and the file editor June 16, 2024 The Assistant starts diagnosing notebook code errors automatically (Diagnose error) June 27, 2024 The Assistant becomes generally available September 3, 2025 Assistant Agent mode, the Data Science Agent, enters Beta October 8, 2025 The Data Science Agent can use Anthropic models on Databricks when partner-powered AI is on March 11, 2026 Databricks Assistant is renamed Genie Code, and Genie Code Agent mode becomes generally available July 8, 2026 Genie Code moves to pay-as-you-go billing with a per-user free monthly allowance
The March 2026 release note puts the change in one line. It says Databricks Assistant is now Genie Code, with expanded agentic capabilities for autonomous, multi-step data tasks.
So if you searched for Assistant Agent mode, this is where it went. Agent mode entered Beta in September 2025 as the Data Science Agent. On the day of the rename, Genie Code Agent mode reached general availability for data science, data engineering and dashboard authoring. Current documentation also describes Genie Code as an agent by default, rather than as a mode you switch on.
What Stayed the Same: The Assistant Features You Already Use
The everyday commands survived the rename almost untouched. In other words, a developer who learned the Assistant in 2024 can still use the same shortcuts today.
Slash Commands That Still Work
Slash commands are the fastest way to give Genie Code a precise job. The Databricks guide to Genie Code in notebooks and the SQL editor lists these commands as of November 2026.
Table 2: Slash Commands That Carried Over to Genie Code
Command What it does /explain Explains the selected code in plain language /fix Proposes a fix for an error, shown in a diff view /doc Adds comments to your code, shown in a diff view /optimize Suggests performance improvements for SQL queries /findTables Searches for relevant tables using Unity Catalog metadata /findQueries Searches for relevant queries using Unity Catalog metadata /prettify Formats SQL queries for readability
Two details matter in practice. First, code you accept from /fix or /doc does not run automatically, so you stay in control of execution. Second, you can reference a table in any prompt by typing the @ symbol, which saves describing it in words.
Autocomplete, Quick Fix and Diagnose Error
Inline autocomplete still suggests code as you type. It also uses a Databricks-hosted model, even when partner-powered AI features are switched on.
Quick Fix handles basic errors that a single-line change can fix, with a fix you can accept and run. Diagnose error, by contrast, goes further. On a code error it runs /fix behind the scenes, and on a failed library install it runs /repairEnvironment instead.
AI-Generated Comments in Unity Catalog
People searching under the old name also ask what AI-generated comments do in Databricks. In short, they are suggested descriptions for tables and columns in Unity Catalog. An owner, or anyone with modify permission, requests them per object in Catalog Explorer, then reviews, edits or accepts them before saving.
The AI-generated comments documentation adds two cautions. Saving a comment runs an ALTER command, which can disrupt running pipelines. The comments are also general descriptions built from the schema, so someone who knows the data should check them for accuracy.
Good descriptions pay off twice. First, they help people browsing the catalog. They also give Genie Code better context, because table and column descriptions go to the model with each prompt. There is no workspace-level or catalog-level toggle for the feature, because comments are generated per object only when someone requests them.
What Changed When the Assistant Became Genie Code
The rename also changed how the tool behaves. The table below compares the Assistant at general availability in 2024 with Genie Code as documented in November 2026.
Table 3: The Assistant (2024) vs Genie Code (2026)
Area Assistant, 2024 Genie Code, 2026 Behaviour Suggests, explains and fixes code that you then run Plans the task, runs code, reads the output and fixes errors itself Where it works Notebooks, SQL editor, file editor Notebooks, SQL editor, Lakeflow Pipelines Editor, AI/BI dashboards, MLflow, jobs and the file editor Approvals Not applicable, because you ran the code yourself AI classifier by default, or Ask every time, Allow in current chat, Always allow Billing No pay-as-you-go model yet (it started July 8, 2026) Pay-as-you-go since July 8, 2026, after 150 free DBUs per user per month Extensibility Prompts and slash commands Adds custom instructions, memory, agent skills and MCP connectors
It Runs the Work Instead of Suggesting It
The Genie Code documentation calls it an autonomous agent. It plans a solution, retrieves context, runs code, uses cell outputs to improve the result and fixes errors automatically. That is the line between a copilot and an agent, and the guide to AI copilots versus AI agents explains it in general terms.
Sometimes you only want an answer, though. In that case, say so in the prompt. For example, write “Just explain this to me, don’t run any code.” Genie Code then replies in conversation instead of executing anything.
Approvals: The Classifier or You
By default, an AI classifier approves the tool actions that carry out your request and is designed to block actions beyond it. Alternatively, you can click Allow or Skip on a single request. You can also set a default approval mode, as the Genie Code navigation guide describes.
Each approval prompt gives four choices, Ask every time, Allow in current chat, Always allow and Auto-approve, which Databricks marks as recommended. The default setting has two options, Ask first and Auto-approve. However, Databricks also calls auto-approve a productivity feature rather than a security boundary. It says to keep it off for production data, sensitive workspaces and shared resources. For those workspaces, we suggest Ask first as each user’s default until the team trusts how the agent behaves on its data.
Where It Works Now
Genie Code also follows you around the workspace. It runs in notebooks, the SQL editor, the Lakeflow Pipelines Editor , AI/BI dashboards and MLflow , and a chat persists as you move between pages.
Newer additions include scheduled tasks, web search with citations, memory, agent skills and MCP connectors. For the agent side in depth, see Kanerika’s Databricks Genie Code guide , which covers agent workflows, use cases and limits. This post, instead, stays with the questions people still ask under the old name.
Databricks Assistant vs Genie: Which One Do You Need?
“Genie” now names a family of products, which is partly why the Assistant-versus-Genie question keeps coming up. Each one, however, serves a different person, as the Genie Agents documentation lays out.
Table 4: Genie Code, Genie One and Genie Agents Compared
Product Who it is for What it does Genie Code (formerly Databricks Assistant) Engineers, analysts and data scientists who write code Writes, runs and fixes code in notebooks, SQL, pipelines and dashboards Genie One Business users A simple chat to ask data questions, explore dashboards and run apps Genie Agents (formerly Genie Spaces) Data teams Configures trusted data, metrics and business rules that business users then query
If you write code, you need Genie Code, the Assistant’s direct successor. If business users want answers in plain English, though, look at Genie One and Genie Agents. Kanerika’s overview of Databricks Genie covers that side. Genie One can also run apps built with Databricks Apps .
The wider shift toward plain-language analytics for business users is the subject of the guide to generative BI . Finally, for a comparison with GitHub Copilot or Cursor, see the Genie Code guide linked above.
Watch on YouTube
What is Genie? Databricks’ AI Platform Explained #shorts
A short Kanerika explainer on the Genie family, useful when you need to tell Genie Code apart from the business-user tools that share the name.
Security and Privacy: What the Assistant Sends to the Model
Two separate boundaries decide what Genie Code can do with your data. The first is about access, while the second is about what leaves the workspace as model context.
Permissions Decide What It Can Touch
Genie Code acts with your Unity Catalog permissions. It can only access data and perform operations that you are allowed to. So a user without read access to a table cannot reach it through the agent either. Clean Unity Catalog grants become the first line of control once an agent runs code on a user’s behalf.
Context Decides What Reaches the Model
To answer, Genie Code sends your prompt plus relevant context to a model. The Databricks AI trust and safety page lists what that context includes.
Table 5: What Genie Code Sends to the Model
When What is sent Always Your prompt, code in the current cell or SQL editor tab, table and column names and descriptions, previous questions and favorite tables Partner-powered AI on Also cell outputs and data samples read from tables, so the agent can check its own work Never Your data, prompts and responses are not used to train foundation models Databricks offers to third parties
Model partners do not retain submitted data, even for abuse monitoring, and Databricks describes these as zero data retention endpoints. Partner-powered features, meanwhile, run on models hosted by Azure OpenAI, OpenAI on Databricks or Anthropic on Databricks. For the controls around them, see Kanerika’s broader guide to Databricks security .
Residency and Compliance Profiles
Genie Code is a Designated Service that uses Geos for data residency. On March 13, 2026, Genie Code chat and cell actions became available through models hosted in-country in Brazil, Canada, Japan, Korea, Singapore and the UK. The change covers workspaces with partner-powered AI on.
Workspaces with a compliance security profile, however, need a closer look. There, partner-powered AI is off by default, and Genie Code availability depends on the compliance standard the workspace follows.
Checklist
Generative AI Checklist (Secure Adoption & Governance)
A practical checklist for adopting generative AI securely, covering data exposure, access controls and governance before an AI tool reaches production data.
Get the Checklist → How to Enable, Limit or Turn Off the Databricks Assistant
Enabling Genie Code is mostly about one account setting and your region. In particular, agentic Genie Code needs partner-powered AI features switched on for both the account and the workspace, plus a supported region.
Check partner-powered AI. Account admins manage it in the account console under Settings, then Feature enablement. It is on by default for workspaces without a compliance security profile.Check data residency. If Genie Code still cannot be enabled, an admin may need to disable “Enforce data processing within workspace Geography for Designated Services” to allow cross-Geo processing.Agree the approval mode. The default is a per-user Genie Code setting, so tell users which one to pick, such as Ask first for production work.Set the compute policy. Genie Code settings include a serverless usage policy for running the code it generates.Pick the effort level. Auto is the default and aims for the highest quality, while Low lowers cost on simpler tasks.Changes to partner-powered AI are logged as account audit events. Databricks is also removing that toggle, but workspaces keep their current value when it goes, according to the partner-powered AI features page .
Can You Turn Genie Code Off?
Turning it off is less clear-cut. As of November 2026, the documentation describes no single switch that removes Genie Code from a workspace. Disabling partner-powered AI removes the agentic capabilities, yet Genie Code stays available on a Databricks-hosted model. Where that setting is already on, its toggle is locked in the settings page. Only the Settings API can switch it off until Databricks removes the setting.
For a full opt-out, therefore, confirm the current options with your Databricks account team. Until then, the approval mode and Genie budgets control what the agent does and what it costs. Your AI governance policies then decide who may use it where.
Kanerika Service
AI Governance Consulting Services for Enterprises
Kanerika helps enterprises set the policies, controls and review processes that govern AI tools and agents across their data platforms.
Explore AI Governance Is the Databricks Assistant Still Free?
Not entirely. Since July 8, 2026 at 00:00 UTC, Genie Code has been billed pay-as-you-go, with a free allowance for each user. Microsoft Learn’s Azure Databricks Genie Code page gives the same date.
The Genie cost monitoring page sets out the allowance. Each user gets 150 DBUs of free Genie usage a month, and the allowance then resets on the first of each month. Service principals, by contrast, get no free allowance, which matters if you schedule tasks or automate Genie Code under one.
Billing Beyond the Free Allowance
Usage beyond the allowance is billed in DBUs based on the underlying LLM usage, tracked through Unity Gateway. Free usage also appears in billing data under the SKU name GENIE_FREE_USAGE. You can then query it alongside the rest of your spend in Databricks system tables .
Two promotions run through January 31, 2027. First, billed Genie usage carries a 25% discount. Second, Genie One and Genie Agents usage by users is free, although service principals are still billed.
Budgets are the main cost control. Account admins create Genie budgets in the account console, as the Genie budgets guide explains. Each budget uses the Unity Gateway resource type and the tag databricks-product: genie.
A budget can target the account, workspaces, groups or users, and it alerts before a limit is reached. It cannot remove a user’s free allowance.
One cost sits outside Genie billing. The compute that runs generated code is billed separately, whether that is serverless or a Databricks SQL warehouse . For that reason, treat Genie Code as one more line in your Databricks cost optimization plan, not as a separate budget nobody watches.
Getting Good Results: Prompts and Review Habits
Genie Code works best with prompts that state the grain, the join keys and what to exclude. Vague requests, on the other hand, leave a code-running agent to act on its own guess.
Prompts That Work in Notebooks and the SQL Editor
The first example below is a SQL prompt that leaves little to guess. The table names are illustrative, so swap in your own with the @ symbol.
Example prompt (SQL editor, illustrative table names):
Using @sales.orders and @sales.customers, write a query that returns
monthly revenue per customer region for 2026.
- Grain: one row per region per month
- Join on customer_id
- Exclude orders with status = 'cancelled' and test accounts
- Just write the query, don't run itSimilarly, the second example asks Genie Code to debug PySpark without rewriting the whole cell. It is also illustrative.
Example prompt (notebook, PySpark):
This cell fails with an AnalysisException on the join.
Explain the cause first, then propose the smallest change
that fixes it. Keep the existing column names and
don't change the output schema.Prompt quality is only half of it, though. Context matters just as much, which is why the guide to context engineering versus prompt engineering is useful reading for teams that rely on agents.
A Review Checklist Before You Accept Generated Code
Databricks itself recommends reviewing generated code before you run it, because it can make mistakes. A short four-point check, which we suggest for any agent output, covers the main risks.
Permissions. Does the code touch only the tables and schemas the task needs?Accuracy. Do the joins, filters and grain match what you asked for, checked on a small sample?Cost. Will it scan far more data than needed or start heavy compute?Traceability. Can a colleague see what changed and why, through the diff, comments or version control?For errors the agent cannot fix, the Databricks troubleshooting guide covers the common failure patterns. Teams that want to score AI output at scale, rather than by eye, can also borrow ideas from LLM as a judge evaluation methods.
AI Assessment
How Ready Is Your Team for AI Agents?
Take Kanerika’s AI maturity assessment to score your data, governance and skills before agents like Genie Code run code on production data.
Take the Assessment → A Transition Checklist for Teams That Standardised on the Assistant
Teams that rolled out the Assistant before billing began now run a billed agent under a new name. The checklist below therefore covers the changes worth making on purpose rather than by accident.
Update internal docs, training and support macros from “Databricks Assistant” to “Genie Code.” Agree which default approval mode users should set in their Genie Code settings, with Ask first for shared and production work. Confirm the partner-powered AI setting and the region and cross-Geo settings each workspace needs. Create Genie budgets per group or user, with alerts set below the limit. Give service principals their own budget, since they get no free allowance. Import the downloadable Genie Code events dashboard to track adoption and engagement. Refresh Unity Catalog table and column descriptions so the agent gets good context. Agree a review rule for agent-written code before it reaches production. For item six, Databricks documents how to measure Genie Code impact through system tables plus an internal survey, with a downloadable events dashboard. Beyond Genie Code, the wider habits that keep a workspace healthy are in the guide to Databricks best practices .
Datasheet
Redefining Enterprise Data and AI Success with Kanerika and Databricks
Kanerika’s datasheet on how we deliver data and AI programmes on Databricks, from platform setup to governed production use.
Download the Datasheet → How Kanerika Helps Teams Adopt Genie Code on Databricks
Kanerika is a certified Databricks partner , and our Databricks consulting services cover the platform work an agent depends on. Applied to Genie Code, that work covers five areas.
Assess. Review current Assistant usage, workspace settings and the quality of Unity Catalog metadata.Design. Set the approval policy, budgets and the data classes the agent may touch.Pilot. Run Genie Code with one team on real tasks and measure the results.Govern. Put permissions, lineage and review rules in place through data governance services .Enable. Train users on prompts, slash commands and the review checklist.Clean metadata is where the payoff starts. For example, Kanerika’s data engineering team refactored an AI-powered sales intelligence platform’s document workflows from JavaScript to Python in Databricks. The results were 80% faster document processing, 95% improved metadata accuracy and 45% accelerated time-to-insight. Accurate metadata of that kind is also what gives an agent good context.
One pitfall is treating the rename as cosmetic. Skipping the approval mode, budgets and review rule leaves two gaps open, unplanned spend and unreviewed agent writes to shared tables.
Case Study
80% Faster Document Processing with Databricks Workflows
Kanerika re-architected an AI-powered sales intelligence platform’s document workflows on Databricks, with 95% improved metadata accuracy and 45% accelerated time-to-insight.
Read the Case Study → What to Do About the Databricks Assistant Now
The Databricks Assistant did not disappear. Since March 11, 2026 it has been Genie Code, which keeps the old commands but now plans and runs work. Since July 8, 2026 it also costs money past 150 free DBUs per user each month.
So check three things this week. Set the approval mode, put a Genie budget on every group and service principal, and agree how your team reviews agent-written code. Do that, and the old Assistant habits carry over safely to the new name.
Frequently Asked Questions
Is Databricks Assistant still available?
Yes, under a new name. Databricks renamed it to Genie Code on March 11, 2026, and the old documentation and product pages now redirect to Genie Code. The everyday tools, such as autocomplete, Diagnose error and the slash commands, still work in notebooks and the SQL editor, alongside newer agent features that plan and run code.
What is the purpose of Databricks Assistant?
The Assistant was built to help people write and fix code inside the Databricks workspace. It generated, explained, debugged and optimized SQL and Python, built visualizations and answered documentation questions. Under the Genie Code name it does the same jobs and can also plan a task, run the code and correct its own errors.
Is Databricks Assistant the same as Genie Code?
Genie Code is the Assistant’s direct successor, renamed on March 11, 2026. It keeps the Assistant’s commands and autocomplete while adding agent behaviour, approval modes and pay-as-you-go billing. It is a different product from Genie One, which serves business users, and Genie Agents, formerly Genie Spaces, which data teams configure for trusted answers.
Is Databricks Assistant still free?
Not entirely. Since July 8, 2026, Genie Code has been billed pay-as-you-go after a free allowance of 150 DBUs per user per month, which resets on the first of each month. Service principals get no free allowance. Through January 31, 2027, billed Genie usage also carries a 25% promotional discount, according to Databricks documentation.
How do I enable Databricks Assistant?
Account admins switch on partner-powered AI features in the account console under Settings, then Feature enablement, for the account and the workspace. The workspace also needs a supported region. If Genie Code still will not enable, an admin may need to allow cross-Geo processing for Designated Services in the workspace settings.
Can I turn off the Databricks AI assistant?
As of November 2026, Databricks documentation describes no single switch that removes Genie Code from a workspace. Turning off partner-powered AI, which needs the Settings API where it is already on, removes its agent capabilities, but a basic version stays available on a Databricks-hosted model. Approval modes and Genie budgets limit what it does and costs. Confirm full opt-out options with your Databricks account team.
What happened to Databricks Assistant Agent mode?
Assistant Agent mode, also called the Data Science Agent, entered Beta on September 3, 2025. On March 11, 2026, Genie Code Agent mode became generally available for data science, data engineering and dashboard authoring. Current documentation describes Genie Code as an agent by default, so there is no separate mode to switch on.
What data does Databricks Assistant send to the model?
Genie Code sends your prompt, the code in the current cell or SQL tab, table and column names and descriptions, previous questions and favorite tables. With partner-powered AI on, it can also read cell outputs and data samples. Databricks says this data is not used to train foundation models it offers to third parties.
What is the role of AI-generated comments in Databricks?
AI-generated comments suggest descriptions for Unity Catalog tables and columns. An owner or user with modify permission requests them in Catalog Explorer, then reviews and edits them before saving. Saving runs an ALTER command that can disrupt running pipelines, and the descriptions should be checked for accuracy because they are built from the schema.
Can Genie Code access data I do not have permission to see?
No. Genie Code works within your Unity Catalog permissions, so it can only read data and run operations that your own account is allowed to. If you cannot query a table directly, the agent cannot query it for you either. That makes accurate catalog grants the first control to check before teams adopt it widely.
