Databricks' Always-On Genie and the End of the Chatbot Era

On November 11, 2026, Databricks announced the second generation of Genie, its enterprise AI assistant, repositioned as an always-on agent that keeps working on tasks in the background without waiting to be asked. The announcement, made at the company's Data + AI Live event in a session between Databricks chief executive Ali Ghodsi and OpenAI president Greg Brockman, marks one of the clearest signals yet that enterprise software is moving past the chatbot.

Genie began in June 2024 as a conversational data assistant: employees asked questions in plain language, and it wrote and ran SQL against governed company data. The new version inverts that relationship. Workers give Genie a goal, and it takes ownership of the work — running jobs on a schedule or in response to triggers, delegating to sub-agents when needed, following up on missing inputs, and reaching out when a person needs to make a decision. Decisions surface through a new priority inbox and email, rather than waiting inside a chat window.

What Actually Launched

The announcement has several layers, and the availability status of each matters. Genie itself and Genie Ontology, the context layer that maintains a company's understanding of its own data, people and processes, are generally available to all Databricks customers. Genie Search, which extends Genie across more than 100 enterprise data sources including Google Drive, Jira, Confluence and Workday, is also generally available. The new always-on mode, along with Ontology Manager, is coming soon to preview and beta programmes — it is announced, not yet shipping broadly.

Genie runs inside a customer's existing Databricks environment, governed by Unity Catalog and Unity Gateway. Every action is logged and auditable, administrators control which tools an agent may use, can require human approval for specific actions, and can set budgets and spending caps. Native plugins bring Genie into ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot, carrying the same governed context and permissions. Genie One, the layer that turns answers into finished work such as slide decks and dashboards, is free for users through January 31, 2027.

Databricks also released a benchmark built with Turing, claiming Genie passed 91% of task runs against 58% for the stronger of two anonymised frontier agents. That is the vendor's own evaluation, not an independent one, and Databricks itself cautions that the gap reflects an established deployment with accumulated business knowledge. Still, the company reports that General Motors has deployed more than 1,000 Genie agents across sales, engineering, supply chain and human resources, and that thousands of customers already use Genie in production.

Why Always-On Changes the Product

A chatbot is a tool you consult. An always-on agent is a colleague you delegate to. That difference is bigger than it sounds. Chatbots are judged on the quality of a single answer. Always-on agents are judged on reliability over hours and days: whether they remember the goal, chase the missing input, escalate at the right moment, and stop before doing something irreversible. The failure modes are different, and so is the buying decision. A company does not adopt an always-on agent because a demo was impressive; it adopts one because it trusts the agent with unfinished work.

That is why Databricks built the launch around governance rather than capability. The Ontology layer exists to give the agent business context — definitions, metrics, people, processes — so its answers are grounded in how the company actually operates, not just in table schemas. The audit trail and approval controls exist so that a compliance team can sign off on an agent that acts while nobody is watching. In enterprise software, trust infrastructure is the product.

What It Means for Founders Building Agent Products

For founders, the Genie launch clarifies where the agent market is heading, and where the openings are. First, the platform layer is consolidating fast. Databricks, with its data lakehouse and governance stack, can now offer agents that arrive with context and controls built in. A startup selling a standalone agent to a large enterprise must now explain why its agent is better than the one that ships inside the data platform the customer already pays for.

Second, the opportunity is shifting to the layers above and beside the platform: vertical agents with deep domain knowledge, agents that work across systems no single vendor owns, and the evaluation and observability tooling that proves an agent did its job correctly. Reliability is the unsolved problem of the agent economy, and the companies that can measure it, guarantee it or insure it will find buyers.

Third, pricing and return on investment are still unresolved. Databricks has not stated a price for the always-on mode, and the free period for Genie One ends in January 2027. Enterprises will want proof that background agents save more money than they cost in compute and oversight. Founders who can show that arithmetic clearly — hours saved, errors caught, decisions accelerated — will sell faster than those selling on the novelty of autonomy.

The chatbot era taught enterprises to ask AI questions. The always-on era will teach them to hand AI the work. The winners will be the companies that make that handover feel safe.

Angeled.com connects founders with investors.