01Genie One. Databricks turned Genie into an AI coworker that sits over all your data. Here's how it actually works.
02This week, in brief. CustomerLake, Unity AI Gateway, and OpenSharing from the rest of Data + AI Summit 2026.
03From Brickster.ai: we built a sourced Databricks vs Snowflake comparison.
01
🧞 Genie One
Genie One wants to be the front door to your data
Databricks built an AI coworker that answers in Slack, ranks its own data sources, and can take action. Engineers are already poking holes in it.
Genie One was the headline of Data + AI Summit 2026. The pitch is blunt: it's an AI coworker anyone in the business can talk to, not another dashboard tool for analysts. You ask a question in plain English, in Slack or Teams or on your phone, and it answers from your governed data, with no SQL to write. Databricks keeps repeating one comparison: 84.5% first-try accuracy on its own 28-question test set, versus 52.4% for the best general-purpose coding agent it tried, at about twice the speed. It's a vendor benchmark, so take it with a grain of salt. What's under the hood matters more.
What's new isn't the chat box. It's a knowledge layer Databricks built underneath, the Genie Ontology. It learns what your business actually means by a term like "active customer," then ranks the competing definitions the way a search engine ranks pages. Genie One sits on top, reaches across your data through Lakehouse federation and Lakeflow Connect, and runs inside the tools your team already uses. It can also act: write documents, set alerts, schedule tasks, and write back into connected tools like Gmail, Slack, and Teams. It's available now.
⚙️ How Genie One actually works
Start with the hard part, the piece Databricks spent the most words on. The Genie Ontology is a living map of how your company works. It pulls bits of meaning from your tables, queries, dashboards, pipelines, and connected apps, then links them into a graph of what your data means and who relies on it. The goal is simple. Stop the model from guessing across scattered fragments, and make it look up the real answer in trusted data through SQL.
The clever part is how it settles arguments. Ten dashboards might define "active customer" ten slightly different ways. The Ontology scores each definition the way Google's PageRank ranks pages. It weighs where the definition came from, who wrote it, how many people rely on it, whether it sits in certified and widely-used assets, and how fresh it is. A definition in a certified, heavily-used table beats one buried in a stale notebook. That ranking is what's supposed to keep answers consistent instead of confidently wrong.
Reach is the second piece. Genie One reads across sources through Lakehouse federation, pulls data in through Lakeflow Connect, and gathers context from over 100 connectors, including Google Drive, Jira, Slack, Confluence, and SharePoint. Then it meets people where they work. It lives inside Slack and Microsoft Teams, where you @mention Genie in a thread, and there are new iOS and Android apps. If your team already runs its own AI agent, an MCP app hands it Genie's features without changing how they work.
The last piece is action and governance, and they go together. Genie One doesn't just hand back a chart. It writes documents and reports, sets alerts, schedules tasks, saves reusable skills, and writes back into other systems. All of it runs under the Unity AI Gateway, which governs not just your data and models but the live calls between models, agents, MCP services, and tools. Your existing Unity Catalog permissions still apply, so a person only sees what they were already allowed to see. The Gateway is also where you watch activity, set guardrails, and control spend.
Under the hood
Genie Ontology Living map of your data, ranks definitions by authority and freshness
Lakehouse federation Queries across data sources without moving the data
Channels Native Slack and Teams @mention, plus iOS and Android apps
Actions & skills Writes documents, alerts, schedules, and back into connected tools
Unity AI Gateway Governs live calls, permissions, guardrails, and spend
84.5%
First-try accuracy
52.4%
Strongest rival
2x
Faster than rival
1M+
Genie Spaces built
🎯 Where it earns its keep
The marketing lead who stops filing tickets
A marketing or merchandising manager @mentions Genie in Slack: which promotions lifted basket size in the Northeast last quarter? Genie picks the right metric definitions through the Ontology, queries the governed tables, and answers in seconds, showing its work. Albertsons uses Genie to let merchants explore complex merchandising data in plain English, the kind of question that used to start with an analyst.
The exec who self-serves the morning number
Instead of waiting on a standing analyst report, a regional leader asks Genie on the iOS app and gets a governed answer before the meeting starts. Foot Locker rolled out Genie Agents to give executives across its North American banners one place to get AI-driven insights.
The ops manager who automates the follow-up
Genie One isn't read-only. An operations manager can tell it to watch a metric, raise an alert when inventory drops below a threshold, draft the summary, and ping the right team in Slack. The query, the alert, and the message happen in one flow, not three manual steps across three tools.
The analyst whose job changes shape
The most interesting case isn't a new user. It's the analyst's job changing. Once colleagues answer the repetitive questions themselves, the analyst stops being a query desk. The work moves to curating the Ontology, certifying the trusted definitions, and building the Genie Agents and skills everyone else then uses. Bad definitions in, confident wrong answers out, so curation now carries real weight.
🏢 What it changes for your business
The promise is a different kind of demand on your data team. Today, someone who needs a number files a ticket, waits, and an analyst writes a one-off query. Genie One is built to soak up that whole category of repetitive ad-hoc work, so people get answers in seconds and the data team stops answering the same five questions. When analysts aren't a human query service, they move up to forecasting, modeling, and pipeline work. That's a real productivity win if it holds.
The catch is that Genie One is only as good as the foundation under it. The Ontology learns from what you already have, so messy lineage, sloppy definitions, and loose access rules don't go away. They get amplified, and handed confidently to people who can't check the SQL. The boring prerequisites matter most: Unity Catalog governance in good shape, a clean set of trusted metrics, and access rules that are actually correct, because Genie inherits them. Most of the work to make Genie One trustworthy is work you'd want done anyway.
On ROI, be honest about the order of things. The price is friendly: it's available now, with $10 of free use per person each month and no per-seat fee, so a pilot is cheap to start. The real cost is getting your governance in shape. Teams that already invested in Unity Catalog and clean definitions will see Genie One pay off fast. Teams hoping it covers for a messy setup will get fast, fluent, sometimes wrong answers, which is worse than a slow ticket queue.
1. The Ontology is the product. Genie One's accuracy claim rests on how well it ranks your own metric definitions. Cleaning up and certifying those definitions is now the data team's most important work, not an afterthought.
2. Governance is the gate, not the garnish. Genie inherits your Unity Catalog permissions and runs under the Unity AI Gateway. Your access rules and certified data are what stand between self-serve answers and confident mistakes.
3. Pilot cheap, trust slowly. Free usage makes a trial easy. Build trust by checking Genie's answers against queries you already trust before you let non-experts act on them on their own.
🧪 The honest part: it's only as good as your metadata
Genie One is days old, so the real stress test is just starting. But the early read from people already running Genie Code, its developer-facing sibling, points straight at the catch, and it's the same one this whole story keeps circling back to. In SunnyData's two-week production review, the verdict was "production-ready, but with conditions." Its effectiveness is directly proportional to the quality of your Unity Catalog metadata, and with poor naming or empty descriptions, the agent hallucinates. Feed it clean, certified definitions and it works. Feed it a messy estate and it confidently makes things up. The same review logged a 77.1% success rate on real-world data science tasks, and called it "a costly toy" for teams that skip the data-quality work.
You can watch that failure happen. A community thread titled "Genie Code hallucinates CLI commands" reports Genie suggesting databricks sql-statements execute, a command that doesn't exist, built by gluing the real SQL Statement Execution API onto invented CLI syntax. Plausible, confident, wrong. That example is Genie Code, not Genie One, but the same model and the same risk sit under both, and Databricks' own docs note Genie can return hallucinations or errors and lose the thread when you switch topics mid-session. Reviewers also flag answers that shift with phrasing, which is a problem if you need an audit trail. None of this is a reason to avoid Genie One. It's the reason to get your metadata in order first, and to check its answers against a query you already trust before you let it act.
Beyond Genie, Data + AI Summit 2026 kept shipping across the rest of the stack. Three threads stood out in the brickster.ai archive this week. Full breakdown at brickster.ai/digest.
Applied AI
Marketing gets a CDP built for agents
In his keynote, Ali Ghodsi reframed the customer data platform as agentic, and Databricks announced CustomerLake to back it up. It's a native CDP that lives inside the lakehouse, governed by Unity Catalog, with Infinity Campaigns that use LLMs and agents for continuous 1:1 personalization instead of batch cycles. It's newly announced, so watch for concrete benchmarks before you rip out your existing stack.
Every Genie agent and coding tool now needs governing, and the Unity AI Gateway is the answer. It extends Unity Catalog to models, agents, MCP services, and skills, with hard spend caps, per-team cost attribution, and beta runtime policies that allow, deny, or require approval on specific actions. Managed MCP services for Slack, GitHub, Jira, and others ship alongside guardrails against PII exposure and prompt injection.
Delta Sharing is becoming OpenSharing, an open-source protocol moving to the Linux Foundation that shares models, agents, and AI skills, not just data, across clouds and vendors. SecureConnect drops the manual IP allowlisting and global replication skips egress fees. Separately, Apps on Databricks Marketplace hit public preview with 20 launch partners, installing third-party apps that run on serverless inside your own account under Unity Catalog.
Not a hot take. A sourced, side-by-side comparison where every cell links to the vendor's own docs.
The most-Googled question in this ecosystem is some version of Databricks vs Snowflake. So this week we built a real answer. /compare puts Databricks, Snowflake, BigQuery, Microsoft Fabric, and Dremio side by side across 70 measures, from architecture and pricing to ML, governance, and streaming. The part that makes it different: it doesn't crown a winner. Every cell links to the vendor's own product, pricing, or docs page and carries the date we last checked it. It's a citation index, not an opinion.
For the head-to-head everyone actually searches, there's a dedicated page. Databricks vs Snowflake opens with a plain-English verdict, then walks the trade-offs section by section with a sourced feature table and a pricing breakdown. The open-source momentum under the platforms, Iceberg, Delta, Spark, and the rest, refreshes every morning from the GitHub API. Have a look, and tell us what we got wrong.
