cd /news/ai-agents/databricks-launches-always-on-genie-… · home › topics › ai-agents › article
[ARTICLE · art-168056] src=startupfortune.com ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Databricks launches always-on Genie agents to take on Anthropic and Microsoft

Databricks on Wednesday, November 11 announced a second generation of its Genie AI assistant that runs always-on in the background on dedicated compute, with Genie and a new Genie Ontology generally available now while Genie Search, an Ontology Manager and the always-on mode remain in preview or beta. CEO Ali Ghodsi told Fortune "The agents are running in the background all the time. So it's always on," and Databricks priced the product on usage with no per-seat fee, a day after Anthropic's own office-tools push. Databricks, which has surpassed $5.4 billion in annualized revenue growing 65% year over year and is eyeing a $175 billion valuation, also released Genie plugins for ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot built on MCP.

by read5 min views4 publishedNov 11, 2026
Databricks launches always-on Genie agents to take on Anthropic and Microsoft
Image: Startupfortune (auto-discovered)

Who should run the agents that work while you sleep: the AI lab, or the company that already holds your data? Databricks is betting on itself, and the bill will be usage-based.

On Wednesday, November 11, Databricks announced a second generation of Genie, its AI assistant, that runs in the background on dedicated compute and works through long jobs without being asked. According to the company's press release, Genie hands pieces of a task to sub-agents that work in parallel, then reaches out through a priority inbox and email when it needs a human decision. CEO Ali Ghodsi told Fortune: "The agents are running in the background all the time. So it's always on."

Fortune framed the launch as coming a day after Anthropic's own push into office tools, which the outlet described as including a spreadsheet product and a version of Claude that joins video calls. Databricks wasn't at that event, but its timing reads as a reply. The pitch is that the agent should live where the company's data already sits, not in a chat window bolted on top of it.

Genie and a piece called Genie Ontology are generally available now, per the press release. The ontology is a living map of a company's metrics, definitions, people and processes, which the agent uses to ground what it does, and it updates as the business changes. Genie Search, an Ontology Manager and the always-on mode itself are listed as preview or beta, coming soon. So the headline feature isn't fully in customers' hands yet.

Genie Search covers more than 100 data sources, including Google Drive, Jira and Confluence, and can index them in place, ingest them or query them live. Databricks says that cuts the number of tool calls an agent makes and the tokens it burns.

Databricks is closing in on Snowflake and the numbers make that case plainly Databricks has surpassed $5.4 billion in annualized revenue growing at 65% year over year, with its Snowflake-competing data warehousing product more than doubling in run rate. As the company eyes a $175 billion valuation in a new funding round, its IPO could become the defining enterprise software listing of 2026 and a benchmark for late-stage AI... - Databricks revenue growth compared to Snowflake enterprise - how Databricks SQL is competing with Snowflake

Pricing is usage-based with no per-seat fee. Neither Fortune nor TechCrunch gave a figure, and the press release doesn't either.

Databricks also released Genie plugins for ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot, built on MCP. That's a telling move. The company isn't asking you to drop your favorite model, only to let its governed data layer sit underneath whichever one you use.

The CEO as customer zero #

TechCrunch's piece gives the launch a face. Databricks runs Genie internally, and Ghodsi uses it to read employee satisfaction surveys, flag problem areas, test proposed fixes from executives, summarize sales calls with prospects and prep for meetings. He says it has made him far more productive as a leader.

His reasoning rests on Dunbar's number. Databricks has 12,000 employees, far past the roughly 150 people a person can keep stable relationships with, so staff don't tell the boss everything. An agent that reads the surveys and the sales notes will. "Anyone who has more than 150 employees, which is a lot of companies, a lot of people can get value out of this," Ghodsi told TechCrunch.

Frankly, a CEO praising his own product is a sales line, and you should read it as one. Still, an agent that tells the boss what the survey data says is a sharper demo than another meeting summarizer.

Why Databricks thinks it has the edge #

Databricks has the money to make this a real contest. Fortune reports the company raised $5 billion at a $190 billion valuation in August, and that it runs at about $7 billion in annual revenue growing around 80% a year. The edge is the one it keeps repeating: permissions. Genie operates inside a customer's existing Databricks environment and obeys Unity Catalog access rules, so an agent can't read what the person who asked isn't allowed to read.

That matters more for an always-on agent than a chatbot, because nobody is watching each step. Per the press release, Unity Gateway sets which tools an agent can call, which actions need approval and how much it can spend, and every action is auditable. TechCrunch adds that the gateway lets companies cap usage and swap in cheaper open-source models when the work doesn't need a frontier one.

Databricks is choosing patience over a 2026 IPO Databricks CEO Ali Ghodsi says 2026 is not the right year for an IPO, even as the company sits at the center of enterprise AI demand. Its private funding gives it room to keep building without public-market pressure. - why databricks delayed its 2026 IPO plans - AI company postponing public market debut strategy

The model labs have the better-known models. Microsoft has the inbox and the calendar, and Salesforce has the customer records. What they mostly don't have is the warehouse where a company's numbers are reconciled, and Databricks is wagering that an agent doing analysis on those numbers will be trusted sooner.

What to check before you sign #

An agent that never sleeps also never stops spending. Usage pricing with no per-seat ceiling means the cost of a background agent depends on how much it chooses to do, so ask for a spending cap per agent before the pilot, not after the first invoice. The approval rules deserve the same scrutiny: decide which actions a Genie can take alone and which wait for a person.

The press release names two customers on the existing Genie product. Headspace expanded self-service analytics across its business teams, and HelloFresh cut the time it takes to spot winning Meta ads by 75%. Both describe the earlier, ask-and-answer version, not the always-on one. Nobody outside Databricks has yet shown results from an agent left running unattended for weeks.

Databricks hasn't said when the always-on features leave preview.

Also read: BloombergNEF finds AI data centres plan 181GW of their own gas power • Google's Gemini Decision 1 destroys the so-called system one models • United Airlines bans AI agents from booking flights ahead of the holidays

This article is posted in AI News, check it out for more related stories.

Join the discussion #

Open in the community → Almost there. Sign in and your reply posts straight away.

── more in #ai-agents 4 stories · sorted by recency
── more on @databricks 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/databricks-launches-…] indexed:0 read:5min 2026-11-11 · —