Soon your AI won't wait for you to ask. Is your data ready?
Databricks announced an always-on Genie. The agent isn't switched on yet, but the map of definitions it will trust is. The shift, the pitch, the catch, and three moves.
02
This week in Databricks
ChatGPT plans in Unity Gateway, Power BI metrics, a corrected deadline, roadmap moves.
03
From brickster.ai
Brickster SQL is on the App Store: learn SQL by building a city under the sea.
01
🧞 This week’s lead
01 / THE LEAD
Soon your AI won't wait for you to ask. Is your data ready?
On November 11, Databricks announced a Genie that watches your numbers and starts work on its own. Most teams can't switch that agent on yet. The map of definitions it will read is live today. Here's the story in five steps.
20,000+
Organizations on Databricks
70%
Of the Fortune 500
91%
Correct, in Databricks' own test
All three figures come from Databricks. The 91% is from a benchmark it built with Turing; more on that in step 4.
Shift
→
Problem
→
Pitch
→
Catch
→
Your move
1. The shift
Your AI is about to stop waiting for questions
Think about how it works today. Weekly signups drop on a Tuesday. Nobody notices until Friday, when someone opens a dashboard, asks a question, waits for an analyst, and gets the answer just before the weekend. The data was there all week. What was missing was someone to ask.
The industry is now racing to close that gap. On November 10, Anthropic launched a spreadsheet app and a feature that lets Claude join video calls, per Fortune. The next day Databricks answered with a new Genie that "stays on the job continuously, tracking progress, following up on missing inputs," and "can run on a schedule or respond to triggers, such as when a metric dips." It also put Genie inside ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot, so it answers wherever people already work.
2. The problem
An agent working alone still has to pick a number
An analyst asked on Friday knows that finance and sales count revenue differently, and says which number they used. An agent working on its own only knows what it has been told.
If three teams count revenue three ways, an always-on Genie will spot a dip, pick one definition, and act on it before anyone checks. The faster the agent, the less time there is for someone to catch the wrong one.
3. The pitch
Databricks says a map of your business fixes that
"Genie has made me a 10x more productive CEO, and now it doesn't wait for me to ask," Ali Ghodsi says in the press release. It works, he says, because it knows "what we mean by revenue, which numbers are official, and who can see what."
That knowledge is Genie Ontology, which "gives Genie a map of your business, including its definitions, metrics, relationships." Agents build most of it from "the dashboards, queries, and notebooks people already use." Teams set the official version in Unity Catalog, and "the OntoRank algorithm gives those definitions precedence." In a benchmark built with Turing, "Genie answered correctly in 91% of task runs, 1.6 times the rate of the stronger frontier agent," "while using less than half as many tokens."
Databricks is not alone. "The entire data platform market is chasing ontologies right now," Mike Leone of Moor Insights & Strategy told TechTarget.
4. The catch
The test is Databricks' own, and the agent isn't on yet
The two rival agents are not named, "correct" is not defined, and Databricks set up the test. Treat the 91% as a reason to run your own trial, not as proof. And the always-on part is not something you can switch on today. Here is what has actually shipped:
Generally available since November 11: "governed definitions of business concepts" that curators can certify.
Genie Search, Genie in Slack and Teams
Generally available since November 9, per the release notes. The press release still calls Genie Search coming soon; we go with the docs.
Always-on Genie, Ontology Manager
"Coming soon to preview and beta programs." No release-notes entry yet.
"...enterprises need to govern what agents do, not just what they read."
Sanjeev Mohan, founder and principal, SanjMo, to TechTarget
Databricks does ship guardrails. "Genie acts with your identity and accesses only information you're authorized to see," and "every action Genie takes is traced and auditable." Customers "pay for what they use, not per seat," and "admins can set workspace and per-user budgets, spending caps, and alerts." No list prices came with the launch.
5. Your move
Use the time before the agent arrives
1
Certify the metrics people argue about
Put your 20 most-disputed definitions, such as revenue, active customer and churn, in Unity Catalog and certify them as Pages, so OntoRank ranks them above what the agents guessed.
2
Cap spending first
Set per-user budgets and alerts in Unity Gateway before anything runs unattended. An agent that works all day also bills all day.
3
Pick one goal and one approver
For the preview, choose one metric that matters when it dips, and name the person who approves what Genie does about it.
The AI that doesn't wait is coming. The definitions it will trust are your job, and you can start today.
Databricks says metric views, its shared definitions of measures such as revenue, can now be published to Power BI in private preview through your account team. The aim is one definition for Power BI, Databricks dashboards and Genie. Publishing runs one way. If Power BI copies the data in (Import mode), Power BI permissions apply, not the access rules set in Databricks. Ask your BI lead to keep the default live connection (DirectQuery) wherever Databricks access rules must still apply.
Dates
A model switch and a corrected deadline
Checked November 16 against the release notes and upcoming-change notices.
Gemini 3.6 Flash is retired. Pay-per-token calls are redirected to Gemini 3.8 Flash at the same price, so answers can change with no release on your side.
Correction: from this date, AI Functions called from SQL that need Runtime 15.4 LTS or above return an error on compute older than 17.3 LTS. Last week we wrote January 2027 and said ai_similarity was excluded; it is not on the exempt list.
Find the apps that call Gemini 3.6 Flash and the queries that use AI Functions, and give each an owner.
November 30 reminder: the Genie One MCP Beta endpoint, an early connection that lets other AI tools call Genie, stops working. Move those integrations to the Genie One MCP in Unity Gateway. Migration notice.
Roadmap
Roadmap moves this week
Generally available: Genie Ontology, Genie Search, Pages, the Genie apps for Slack and Microsoft Teams, the Genie One desktop app on macOS, standalone pipelines on serverless general compute. New in beta: AI Functions Asynchronous API (up to 50,000 requests per batch), AI Function tasks in bundles, Genie ZeroOps, the Genie One Windows app and Chrome extension, artifacts and slides in Genie One, budget alert webhooks. Retiring: Gemini 3.6 Flash on November 19. November release notes · AI/BI release notes
03
🧱 From brickster.ai
📱 Brickster SQL is on the App Store
Learn SQL by building Brickwater, a city under the sea, ten minutes a day. 114 lessons run from your first SELECT to joins, window functions and SQL on Databricks, offline. Free, no account, no ads. iPhone, iOS 16.4 or later; not in mainland China.