Databricks expands Genie into an always-on AI agent for enterprise workflows

Databricks has announced a new version of its Genie platform, turning the enterprise AI assistant into an always-on agent designed to manage tasks autonomously and support decision-making across business functions.

The second-generation Genie can take ownership of tasks assigned by employees, track their progress, delegate work to smaller AI agents and follow up when additional information is required. It can also notify users when a decision or intervention is needed through a new priority inbox and email alerts.

The announcement reflects a broader shift in enterprise AI adoption, as businesses look beyond tools that answer questions towards systems that can manage workflows and carry out tasks across multiple applications. However, such systems need access to relevant business information and must operate within an organisation’s existing security and governance policies.

Databricks said the new Genie is designed to address these requirements through its Genie Ontology, which maintains an understanding of how a company operates, including its data, business definitions, people and processes.

Genie moves from answering queries to managing tasks

The updated Genie can operate in the background using its own computing resources. Employees can assign it a goal and define the boundaries within which it can work. It then manages the task and contacts the user when it requires additional information or a decision.

For instance, a marketing team could ask Genie to monitor generative engine optimisation (GEO) performance and suggest relevant content. Customer success teams could use it to track changes in customer usage patterns and alert account managers when activity shifts. Executives could also ask it to monitor revenue trends and customer feature requests before preparing regular business updates.

The platform can delegate work to multiple sub-agents, allowing different parts of a task to be handled in parallel. A new priority inbox is intended to bring decisions and pending actions to the user’s attention, while email alerts provide updates when users are away from the application.

Databricks CEO and co-founder Ali Ghodsi said Genie had made him more productive by keeping track of emails, preparing him for meetings and working in the background. He attributed this capability to the platform’s understanding of company-specific information, including business definitions, official metrics and access permissions.

Genie Ontology provides business context to AI agents

A key part of the announcement is Genie Ontology, which Databricks describes as an agentic ontology for enterprise AI.

In simple terms, an ontology helps an AI system understand how information relates to a business. It can provide context about what particular metrics mean, how business processes work, which teams are responsible for them and how different pieces of information are connected.

This is important because enterprise information is often spread across different systems and departments. The same business term may also have different meanings across teams. Without this context, an AI agent may retrieve information but misunderstand its relevance or provide an answer that does not reflect how the organisation operates.

Databricks said Genie Ontology maintains this business context as the organisation changes. It is designed to help Genie interpret both structured data, such as tables and metrics, and unstructured information, such as documents and other business content.

The company is also introducing Ontology Manager, a new interface through which teams can manage and govern business context. It will provide recommendations to identify frequently used concepts that may need further refinement.

Genie Search expands access to enterprise information

Alongside the updated agent, Databricks is expanding Genie Search to help employees and AI agents find information across enterprise data systems and business applications.

The search capability is designed to work across the Databricks lakehouse and external platforms such as Google Drive, Jira and Confluence. Databricks said it now supports more than 100 connectors, offering different ways to access information. These include bringing data into the lakehouse, indexing information where it is stored or querying applications through the Model Context Protocol (MCP).

The aim is to allow AI agents to search across structured and unstructured information without having to check each application separately. By using Genie Ontology, the search system can also interpret information in the context of business definitions and processes, rather than simply returning documents that match a query.

Databricks said the system follows existing access permissions and allows teams to inspect the sources behind the information retrieved. This is particularly relevant for enterprises where employees and AI agents may have different levels of access to business data.

Governance remains central to autonomous AI deployment

As AI agents gain the ability to access enterprise applications and perform tasks, organisations need to ensure that their actions remain within defined security boundaries.

Databricks said Genie operates within a company’s existing environment and is governed by Unity Catalog and Unity Gateway. Unity Catalog manages permissions and access to enterprise data, while Unity Gateway provides controls for AI model and tool usage, including approval rules and budgets.

The company said every action taken by Genie remains subject to these controls and is auditable. This is intended to allow enterprises to introduce autonomous workflows without creating a separate set of permissions and governance systems for the agent.

The announcement also includes capabilities to generate data-driven presentations, dashboards and interactive reports that remain connected to governed data. Genie Agents will additionally be able to work with documents such as PDFs, contracts and presentations stored in Unity Catalog volumes.

Databricks also highlighted customer use cases. HelloFresh said its use of Genie to analyse performance signals had improved programme-level advertising returns and reduced the time taken to identify effective advertisements on Meta by 75%. The company also cited adoption by Headspace, which said governed business definitions had helped expand access to self-service analytics.