Products in the News

Tech and AI products in the news, read for the go-to-market angle

Databricks wants Genie to stop waiting for the prompt

Databricks wants Genie to take a goal and keep working without being asked. Analysts like the ontology underneath it. The always-on mode is not something customers can switch on yet.

By
Eric Wright
Product
Genie (second generation), Genie Ontology, Genie Search
Company
Databricks

The short version

On November 11, Databricks announced the second generation of Genie. The headline feature is an always-on mode. A user hands Genie a goal and some limits, and it keeps working in the background. It delegates to subagents, chases missing inputs, and raises decisions in a priority inbox or by email. Databricks also made Genie Ontology generally available and announced Genie Search. Read the fine print, though. Genie and Genie Ontology are generally available today. The always-on mode, Genie Search, and Ontology Manager are listed as coming soon to preview and beta programs.

What it is

Genie launched in June 2024 as a natural language way to ask questions of data in Databricks, according to TechTarget. It has since grown into a family with Genie One, Genie Agents, Genie Code, and Genie Ontology.
The new release has three parts.
Always-on Genie runs on its own compute. Databricks gives examples like keeping a quarterly forecast on track, watching customer usage for shifts, and sending executives a weekly brief on revenue trends. Each new alert arrives as its own email thread. Actions stay inside Unity Catalog permissions and Unity Gateway approval rules and budgets.
Genie Ontology is a knowledge layer. It captures a company’s definitions, metrics, people, and processes, and keeps them current as the business changes. Databricks first introduced it in June.
Genie Search indexes content where it already lives, including Google Drive, Jira, and Confluence. Databricks says it reaches more than 100 enterprise sources without copying the data.

What the coverage says

TechTarget’s Eric Avidon talked to two analysts. Both liked the direction. Both had caveats.
Mike Leone of Moor Insights & Strategy called the release significant “because Genie stops waiting to be asked.” He said the ontology is where Databricks pulls ahead, since it learns definitions from how people query the data. He also said the whole data platform market is chasing the same idea.
Sanjeev Mohan of SanjMo said always-on Genie “will get the headlines, but it only works if the agent understands the business.” His concern was governance of actions. He said enterprises need to govern “what agents do, not just what they read.” He listed per-agent identity, limits on actions, human approval for consequential steps, and a full audit trail. Leone wanted a way for Genie to say when it is unsure of an answer.
The AI Journal interviewed a Databricks co-founder. That piece reported a Databricks benchmark where Genie answered correctly in 91 percent of runs. Two unnamed frontier agents scored 45 and 58 percent. The article also noted that Genie ran inside Databricks for the test, while the rival agents could not reach Genie’s tools. It added that customers cannot turn on the always-on mode yet.
The coverage got the governance question right. It mostly glossed over the availability gap. Several headlines read as if autonomous Genie shipped this week. The press release says preview and beta programs are coming soon.

Opinion

The go-to-market read

The interesting choice here is the inbox. Databricks is selling a coworker that comes to you, rather than a smarter chat box you visit. That changes the buyer. A chat tool gets bought by the data team for analysts. An agent that emails a sales leader about a forecast slip gets judged by that sales leader.
That is a big opportunity and a real risk. Leone’s point lands here. A wrong answer delivered unprompted to a busy executive spreads faster than one someone had to ask for. The first bad alert will shape trust for months.
If I were on the Databricks field team, I would lead with the ontology and the governance story. I would sell always-on as the reason to get the ontology right now. Customers who build clean definitions today will be the ones who can turn on autonomy safely later.
If I were a buyer, I would ask three questions. When will always-on reach my workspace? What does it cost in compute when it runs all day? Who gets paged when it sends something wrong?
The benchmark is a vendor number from a vendor-friendly setup. Treat it as a claim to test on your own data.

Who should care

  • Data and analytics leaders already on Databricks.
  • RevOps and FP&A teams who live in recurring reports.
  • Field teams at Snowflake, AWS, Google Cloud, and Microsoft, who will hear about this in deals.
  • Security and governance owners who will have to approve agent actions.

Sources