Incident sheet · /enterprise-ai

Filed
Read
4 min · 806 words
Severity
Routine

Databricks adds always-on Genie agents and ontology

Databricks has made Genie Ontology generally available and added always-on agents and governed search. Analysts see an ontology lead but say agent actions need tighter controls.

Databricks turns Genie into always-on agent for complex work
FeedDatabricks turns Genie into always-on agent for complex work — AI-generated

Event log

  1. Three connected capabilities were announced: always-on Genie, Genie Ontology and Genie Search.
  2. Genie Ontology, first previewed in June 2024, reached general availability.
  3. Genie now spans five named components, including analysis, domain-specific agents and pipeline coding.
  4. Snowflake, AWS, Google Cloud and Microsoft are also developing autonomous-coworker tools.
  5. Analysts call for agent identities, action limits, human approvals and audit trails.

By Amara Osei

Databricks unveiled always-on Genie on Wednesday, giving enterprise users agents that can accept a goal, own work and assign tasks to subagents without waiting for another prompt.

Moor Insights & Strategy analyst Mike Leone said: "Taken together, I'd call [the new capabilities] significant because Genie stops waiting to be asked." He added that handing Genie a goal can put answers in front of people who would never open a dashboard.

What did Databricks release?

The announcement covered three connected products:

  • Always-on Genie turns on-demand assistants into autonomous workers that take assignments and delegate tasks to subagents.
  • Genie Ontology, previewed in June 2024, is now generally available. It captures company definitions, metrics and relationships across structured and unstructured data, wherever that data sits.
  • Genie Search gives people and agents a governed route to information across a company's entire data estate, grounded in Ontology.

SanjMo founder Sanjeev Mohan said: "Always-on Genie will get the headlines, but it only works if the agent understands the business." Ontology provides that context by learning definitions, metrics and relationships from data and usage.

Takeaway: Databricks has moved from answering questions to accepting objectives.

How does Genie learn the business?

Databricks senior director of product management Ken Wong said customer feedback shaped features already in the plan. Ontology Manager, developed in response to requests, lets data teams govern business context and recommends refinements to frequently used ontologies that guide agents.

Customer feedback also drove support for external business intelligence platforms, where much analytical logic still lives. Wong said customers wanted context held in documents, tickets and chats, beyond Databricks and traditional data systems.

The company had federated connectors at launch, but built Genie Search to index those systems and improve performance.

Without shared definitions, an agent cannot tell what an "active customer" means at a particular company or which revenue number the finance department trusts.

Takeaway: Genie's accuracy depends as much on company context as on its model.

What has Databricks built since June 2024?

Since its introduction in June 2024, Databricks has expanded Genie into five named components:

  • Genie One provides natural-language data analysis.
  • Genie Agents helps users build domain-specific agents that reason over curated datasets and business logic.
  • Genie Code helps data engineers and scientists write and debug pipeline code.
  • Genie Ontology supplies the governed business context used by agents.
  • Genie Search extends that context to information across the data estate.

Other additions since Ontology entered preview include live dashboards and interactive reports rooted in governed data. Agents can now work across data sources alongside governed tables, and users can reach them from any workspace, including web browsers and mobile applications.

Takeaway: The product now spans analysis, code, agents and shared definitions rather than a single chat tool.

What does Search change?

Search gives agents one governed entry point across the data estate, rather than separate access to each application. Moor Insights & Strategy analyst Leone called it one of the most valuable additions because it can set permissions in one place and reduce the number of external calls agents make, lowering their cost.

Mohan described most enterprise AI as reactive, with a user typing a prompt and waiting. Genie Search and Ontology expose relevant data together with its meaning, allowing users to hand off recurring analytical work instead of doing it by hand.

Takeaway: Unified search turns scattered information into one repeatable route for analysis.

Is Databricks ahead of its competitors?

Ontology's general release puts Databricks among the first vendors to offer an ontology layer. Snowflake, AWS, Google Cloud and Microsoft are also developing tools for autonomous coworkers, so the category remains crowded.

Leone said Databricks matches the market's overall direction but pulls ahead on ontology because the system learns a company's definitions from how people query data, rather than requiring the team to write every definition first. He also noted that the wider market is developing ontologies that AI can build and manage while people approve changes.

Mohan described the shared destination: a semantic or context layer, unified search across structured and unstructured data, and agents that act on both.

Takeaway: Databricks holds an early ontology advantage, not exclusive ownership of enterprise agents.

What still needs governing?

Mohan said Databricks has made real progress on governed access. Enterprises, he argued, must also govern what agents do in production. That requires a clear identity for each agent, limits on permitted actions, human approval for consequential steps, and a full audit trail of their reasoning and actions.

Leone agreed that Databricks keeps roaming agents mostly within bounds, but wants always-on Genie to signal when it lacks confidence. A wrong answer delivered without prompting can reach many people before anyone thinks to question it.

Takeaway: Databricks has built the autonomy layer; enterprises still need to define its limits.

via techtarget.com (Source)

Filed under

  • databricks
  • genie
  • ai-agents
  • ontology
  • data-governance
Share this article:

More from Amara Osei

Amara Osei

Show full bio

Correspondent covering regulation, policy and their impact on enterprise information technology and the leaders who run at CIO Monitor.

188 articles

The newsroom digest

One email a day with the stories that matter. No noise.

By subscribing you agree to the privacy policy.

Cross-references

‹ Previous articleNext article ›