The New Genie: From Data Q&A to Enterprise Agent

Summary

Genie 1 is introduced as a unified, multi-agent AI co-worker for business users that goes beyond answering data questions to taking governed actions—sending Slack messages, creating Jira tickets, opening GitHub PRs, and co-authoring documents with visualizations—powered by the Genie Ontology that auto-builds business context from enterprise data.
Genie connects to enterprise knowledge stores and unstructured content through data connectors including MCP, search connectors, Lakeflow, and text-to-SQL, and outperforms generic frontier models on business-specific questions where domain context matters.
Recurring business workflows can be automated through Genie Skills scheduled via Lakeflow Jobs or triggered by events, with proactive Genie Insights delivering push notifications to the Genie mobile app so users stay informed without polling.

The New Genie: From Data Q&A to Enterprise Agent

n this video, we'll introduce the new Genie, a unified, multi-agent AI that doesn't just answer questions about your data, it acts on it. We'll show how Genie connects enterprise knowledge stores and unstructured content to deliver richer insights, how it orchestrates across Genie Agents to complete complex work, and how it can take governed actions on your behalf, including automating recurring tasks and skills. Attendees will leave understanding why Genie is not just a data Q&A chatbot, but an enterprise AI that connects your people, your data, and your systems — and gets work done.
Talk By: Elise Georis, Senior Manager, Product Management, Databricks ; Rhetta Nadas, Product Manager, Databricks ;

Chapters

FAQs

What is Genie 1 and how is it different from a data chatbot?

Genie 1 is Databricks' AI co-worker for business users that goes beyond answering questions about data—it connects enterprise knowledge stores, computes results rather than reciting stored documents, and takes governed actions on behalf of users. Unlike a typical data Q&A chatbot, Genie 1 can orchestrate across multiple Genie Agents to complete complex work and automate recurring tasks through Skills. This video explains how Genie 1 is benchmarked against frontier models specifically on business questions, where enterprise context makes a decisive difference.

What is the Genie Ontology and how does it work?

The Genie Ontology automatically builds and ranks business context from an organization's data, learning the semantics, metrics, and relationships that matter for specific business questions. It surfaces this context through ontology snippets that help Genie answer questions accurately without requiring extensive manual configuration. The ontology can also detect deprecated tables and suggest updated alternatives, keeping business context current over time.

What kinds of actions can Genie 1 take on behalf of users?

Through MCP write actions, Genie 1 can send Slack messages, create Jira tickets, and open GitHub pull requests based on the results of data analysis. It can also co-author documents with embedded visualizations, such as courier earnings analyses, and automate recurring business workflows through Skills. Skills can be scheduled on a calendar basis or triggered by events via Lakeflow Jobs, enabling Genie to complete work proactively without user prompting.

How are Genie 1 capabilities governed at scale?

Genie 1 operates within constraints and guardrails set by the organization, meaning any actions taken—writing to external systems or running automated skills—are bounded by defined permissions. The video covers governance controls that prevent Genie from acting outside its authorized scope while still allowing business users to take meaningful action on insights. Proactive Genie Insights push notifications allow users to stay informed without granting broader system access than necessary.

Full transcript

[00:08] Hi everyone. Thank you so much for joining us today. My name is Elise. I'm on the product team here at Data Bricks. >> I'm Rea, another product manager at Data Bricks. And we're here to talk about Genie 1, which is our AI co-worker for
[00:20] business users. Uh some of you may have known this as datab bricks one which was an initial code name but what we are launching this week is a lot bigger. Uh Genie 1 goes beyond your lakehouse to
[00:32] actually answer questions about all of your data. It goes beyond answering questions to actually helping you take action on what you learn. And it's also powered by the Genie ontology which
[00:45] you've hopefully heard about this morning. uh we are going to recap the context from the keynote uh earlier today, but then we're also going to dive into several of the new features in quite a bit more depth and we're going to show you a couple things that we
[00:58] actually didn't show this morning. So specifically, we're going to introduce Genie 1 and share what makes it different. Then we're going to again talk about those features. We're then going to cover how you can keep this governed at scale. So we're going to
[01:11] dive into that a little bit more than we did today earlier. And then if you have questions after this talk, we might have time for some live Q&A, but if not, you can just find us in the hallway afterwards. We will stick around. Uh just as a reminder, session is
[01:23] forformational purposes only. The road map just continues to shift as we continue to learn from customers like you. Now, I want to start with a quick definition. Again, Genie1 is data brick's AI co-worker for business users.
[01:37] Uh anyone can connect their data and their sources. they can take action on the answers that Genie1 gives them within the constraints and the guardrails that you've set. Uh, and along the way, a Genie is answering
[01:51] pretty deep questions about your business, including computing results rather than just kind of reciting what's already in your documents and your sources of truth. Now, I think when we say something like this, you might just
[02:03] be thinking, "Oh, great." Like, he had another agent. Uh, but what we're going to try to prove to you is that Genie1 is is really nothing like the other agents on the market. And if we're successful, you might want everyone in your organization to try it.
[02:17] After all, a generic model is really good at answering questions that require searching the web or maybe a specific code base. But when you start to ask questions about your business, it starts
[02:30] to fall down. Uh, for example, if you ask questions about your team's revenue, it might grab some random revenue table rather than the one that the finance team actually uses. Or if you ask for
[02:42] your turn rate, it probably has no idea that your team excludes free trials from that metric. Or maybe even the concept of free trials uh itself. And if you ask maybe which campaign performed best, it
[02:54] has no idea that the campaigns table is super laggy. So if you want the latest data, you actually have to go straight to HubSpot for that. More broadly, every company runs on this type of institutional knowledge that's stuck in
[03:07] docs and dashboards and tables and people's heads. And to make matters worse, that knowledge moves. Tables get renamed, docs go stale, people leave the company. And so no matter how
[03:20] sophisticated a model is, it's missing a lot of that really critical context. But Genie was built for exactly those types of questions and it helps that we've been working on this context problem for
[03:32] a long time and we will dive into how we're solving that problem. But first Red is going to show you this in action. Awesome. Thank you Elise. So Genie has made Elise and I so efficient and effective. We have have had some time
[03:45] for a side hustle. Lake Lunch, our delivery app. It takes you from the data lake to the dinner plate. So let's see how that's doing. So, I've asked it a very simple question about daily active users at Lake Lunch over the last 90
[03:58] days. And this is the response it gave me. So, if we dig in a little here, I'll zoom. I know it's hard to see back there. Um, it didn't just give us a daily active user number. It gives us one, but it gives us an interactive chart where we can see how it's changed
[04:11] over the last 90 days. It shows key patterns like week weekly seasonality for days of the week that there's more orders. And something's really cool here. Look at this. It's pointing out that there was a dip, which you can see here. And it's attributing it to an iOS
[04:24] outage. So, it's already figuring out something I'm going to want to know information about and telling me it ahead of time. And I didn't even bring that up. I just asked for daily active users. It also shows the methodology, which is really important here. So, it
[04:36] shows us that it is counting distinct user ID per calendar day. So, it is not triple counting multiple counting a single user in one day if they order multiple times, which is something I do a lot. um and it's pulling it from a
[04:48] certified source of truth table which is app events and they're dduplicating across iOS, Android and web. Now this might seem simple but it's ve really it's not easy for all agents to do. And if we click into app events where it's
[05:01] getting its data from let's take a look at that raw app events we are in. Let's look at this table. Yeah, this isn't this is a pretty raw table. It's got raw event ids, raw user
[05:13] IDs. It's pretty messy and um there probably are different freshness levels in here and there's definitely no daily active user section. So, Genie Dit figured all of that out on its own. Back
[05:25] to you, Elise. Sweet. Now, of course, that is one question, but more broadly, when we benchmark Genie against some of the strongest models available, it
[05:37] consistently comes out as one of the top performing agents for these types of business questions. And the gap starts to really show when we start asking hard questions where being right depends on picking the authoritative data source
[05:50] and applying your business's rules. And you're also seeing some impact on cost from that. A general agent has to make up for the fact that it doesn't know your business. So if it's not going to
[06:02] hallucinate, then it explores. It searches your schemas. It samples your tables. Runs a bunch of excessive queries. pulls in a bunch of docs and then it loads all of that into the model's context in case any of this
[06:15] turns out to matter later and you pay for every single one of those tokens. But Genie has to do a lot less work because it knows the sources that matter. So it can go straight to the few
[06:27] that do and ignore the rest. And for the same reason, it can actually be a lot faster on these questions too. agents get really slow when they have to traverse data. But because Genie knows the right source and goes straight
[06:40] there, you can get results in a fraction of the time and we are really seeing a lot of customers benefit from that in practice. So the proof is really in the pudding. One of our customers, GM, said
[06:52] that they have over 1,000 Genie agents in their production environment. It is changing the way they surface information from over 9 million data assets and data bricks. And what used to take them hours of digging now happens in moments. And that has been a total
[07:05] gamecher for them. So how does Genie get these questions right? Well, because it knows your data, your definitions, even who's most authoritative when it comes to drafting those definitions. And it does that through the Genie ontology.
[07:19] The ontology is a map of your data and your business that's automatically generated and maintained by Genie. So Genie is going to automatically extract snippets of knowledge from your
[07:31] dashboards, your tables, your metric views, your connected apps. And for example, you know, a snippet might say something like our fiscal year starts in February. So FYQ1 is actually February through April rather than January
[07:44] through March. Or, you know, it might tell you that headcount is authoritative in workday, but the table's actually quite laggy. or you know a qualified lead only counts if it's you know a meeting is booked and the budget is
[07:56] confirmed and all of that information lives in Salesforce and as Genie accumulates these snippets it assigns each of them an authority that's a function of things like how often the snippet is used what asset it was drawn from you may have heard the term onto
[08:09] rank when Ali and Ken talked earlier today that is what this is referring to so when you ask a question Genie is actually able to rank the relevant snippets to see which one might be most helpful to you. It's kind of similar to
[08:22] how we might rank results in a search algorithm. And that includes resolving conflict between two snippets that are talking about the same concept. So some random query that I wrote a couple years ago is going to be outranked by a
[08:35] certified dashboard when they're talking about the same thing. And that is a lot more powerful than just grinding through an agentic loop and trying to figure out what matters and what doesn't. And just as importantly, the genie ontology
[08:48] enforces your permissions. So if Rea has access to a dashboard that I do not, she can see snippets from it, but I certainly can't. So now Reddit is going to show you that in practice. Awesome.
[09:00] So we're back in Genie. We're still thinking about my awesome app, Lake Lunch. I'm going to ask it another question and see if we can invoke some of the ontology. So, I'm going to ask
[09:12] now, show me monthly active users over the last 90 days. And it's important to note that when we were putting this together, I tried a very similar monthly active user question um for one of Data Brick's products across three very, very
[09:24] popular agents. And true story, they all failed. One of them looked at a blog and then gave me a query for system tables. Another just looked at a document and gave a really old data uh really old
[09:37] number. and another did connect to the MCP data bricks but it still um got the number completely wrong. It quoted a number that was double counting. Um so this isn't that easy to do. So if we look in here I again asked about monthly
[09:50] active users over the last 90 days at Lake Lunch and off the bat it's doing something great. It knows that March is partial and instead of giving me an incomplete number it's pointing that out. Um again here we have the methodology. It knows that monthly
[10:03] active users is counting distinct user ID um over each full calendar month instead of just taking those daily active users we had before and multiplying it by 30. Um it's dduping across platforms. And another thing
[10:16] that's really cool here is you'll see it's calling out that this differs from one of my existing dashboards. And I might think, hey, I thought I just saw this number in another dashboard. This looks completely wrong. It's telling me that that dashboard is using a deprecated stale table. It's again
[10:29] pointing out this thing where we are recounting. And so this is super helpful when sometimes you might see different numbers from a different source. Jeanie is going to proactively tell you why it's different. Um, okay. So let's see
[10:41] if we've got any ontology snippets. I thought I saw some while it was thinking. Ah, here we go. Perfect. We've got some. So it one of the snippets is that active you let me zoom in a little. Active users is a distinct dduped user.
[10:54] It's looking at this certified source of truth, which is great for active users. And it knows that monthly active users must be recomputed as that count distinct user ID. Something that a lot of the other agents got wrong. Let's
[11:06] keep looking at these ontology snippets. Let's click into one. Everyone can see. So, when you first click in, you'll see you've got a bit of a bigger description on what it's actually telling you. If we
[11:18] look down here, we'll see the author is Elise. She definitely knows what she's doing. Um, and then there's this chart down here where you can see where it comes from. It looks like it is coming from there we go that app events table
[11:30] and Elise, you can see that Ely contributed to it. I want to see what Elise has been up to. So, let's click into her. Wow, you've been very busy. You've contributed to a ton of domains and let's see how many Wow, a ton of
[11:42] ontology snippets have come from some of the objects you've created. So, you're definitely a great source of truth here.
[11:54] So where does that context all come from? There are a couple different ways that you can get data into Genie. Each of them is suited to a slightly different type of data and a different type of query. So we generally recommend enabling all of these and then letting
[12:07] Genie auto route. You're going to get the most optimal performance and results that way. Uh so when something changes by the minute and you really want kind of that live status uh like the status of for example a Jira ticket genie
[12:20] connects through MCP it queries the source the moment you ask the question so you're getting that fresh information but when you have unstructured content especially at higher volumes or maybe
[12:32] you're asking harder questions across multiple data sources or you want to have those alles uh indexed alongside the data that you're ingesting testing into your bronze tables. We are
[12:44] introducing search connectors which are going to funnel into a new genie search index. This is probably the first time we're actually talking about this publicly. Uh so these search connectors,
[12:56] they're fully managed. There's no need to manage the pipelines. There's no need to manage data processing, chunking, vectorizing. You don't have to think about the vector search index. And you also don't have to think about enforcing
[13:08] permissions. They simply plug straight in to your platform. You say enable SharePoint, enable Teams, and you can feel confident that Genie is only going to show the information to the end user
[13:20] that they actually have permission to in the data source. We are entering private preview with a couple of connectors and then there's several more that are kind of hot on their tails. So, if you are interested in this private preview, find
[13:32] me after the talk and we can get you signed up for that. And finally, for structured data that's kind of sitting in a system of record like a Salesforce or a SQL server, LakeFlow Connect and Lakehouse Federation make it available
[13:44] to Genie through text to SQL. And we have dozens of connectors and that list is continuing to grow really quickly. This is already more logos than we had just a couple months ago. And that text to SQL piece is actually quite critical.
[13:57] Genie isn't just reciting the information that already lives in your documents. the the analyses that Rea was just showing you, there's no document that has that information yet. That's why she's creating these documents. So,
[14:09] Genie is able to query results on your live operational data uh in tandem to doing this more traditional search. So, altogether, we think this forms a really powerful combination. You know, Genie
[14:22] isn't just for unstructured rag, but it also isn't just doing MCP, and it even isn't just doing text to SQL, at least no longer. You know, it combines the power of search with the
[14:35] freshness of live queries across both your first party and your third-party data. So, you can actually ask a much wider spectrum of questions. So, Genie can use all of these data sources, but
[14:48] you may have a few concepts that you do want to deterministically codify. For example, a definition of what an active user actually is. And you can supplement the Genie ontology by contributing to
[15:00] the Unity catalog glossery. The UC glossery is your business's dictionary of terms. Every definition in the glossery is certified. It's owned by
[15:12] someone. It's version controlled. It's stored right next to the data. And the UC glossery is actually symbiotic with the Genie ontology. So because the glossery is authoritative, you told us
[15:24] this definition and you've told us it's certified, that's going to be prioritized by the geneontology. And similarly, if you find something interesting in the geneontology that Genie has inferred, you'll be able to
[15:36] promote it back into the Unity catalog glossery. So you don't really have to manually curate everything. We are offering this form of curation but we really think that together these are
[15:48] going to help actually scale to your entire organization. This combination of of curation with with automation. Now we understand that you may have already invested in a modeled ontology from another vendor and we do not want
[16:02] to waste that work. So the UC glossery can read from your existing ontology providers, your existing definitions. So the semantics that you've already put together become the ground truth in Genie. And in the other direction, the
[16:15] Unity catalog glossery can be exported to a thirdparty provider. So whatever tools you've chosen to standardize on, Genie essentially wants to build on them rather than forcing you to start from scratch. And just to give you a sense of
[16:29] the product in action, here's what one of those UC glossery pages actually looks like. Of course, getting an answer is hard, but it's actually really just step one. You actually probably want to act on what you've learned. And that's
[16:42] why Genie has evolved from a reactive Q&A tool into a proactive AI co-worker. And we want to go through quite a few of the new features in some depth, and we're going to demo them as well. We're
[16:54] going to start with documents. Genie can now co-author documents with you. You can have it start from scratch or maybe hand it a template and ask it to fill it out. That's how we like to do our reviews these days. And you can then send your doc as an email, as a PDF,
[17:08] maybe a link in Slack, Teams, Copilot, and inline collaboration is also in development.
[17:21] Awesome. So, I have spent so much time preparing for this presentation that I haven't had that much time to do a deep dive into how our couriers are doing and look into courier earnings. So, I'm going to have Genie do all that work for me. So, I've asked it to build a doc that performs a
[17:34] deep dive into courier earnings. I'm gonna look at the last seven days. I'm really interested if we should explore a different incentive structure, maybe during certain dates, in various regions, different times. I want to make
[17:46] sure to include new KPIs and trends outside of what I know my team and I are looking at every day in dashboards. So, I'm giving Genie a pretty big task here. It is, as you can see, thinking. I'll zoom in here as well so everyone can see. It
[17:59] should be done in a second. We can even click into what it's thinking. Almost done. Come on, Genie.
[18:18] this is a big question. It's doing pretty well here. As you can tell, it's thinking a lot. We got a lot of vis. It's givingven us a lot of visualizations. Here it comes. Okay, perfect. Thanks for bearing with me. I'm sorry. Here we have our courier earnings action plan. So, let's take a look at
[18:31] what Genie did. I'm going to zoom in for the folks in the back. Great. Um, so first off, great. I love this immediate actions. It's telling me what I need to do. Um, looks like there is a big
[18:43] emergency in Chicago. And so, not only is it telling me what I need to do, it's telling me the owners. It's telling me when I need to get this done, and it's telling me how much it's going to cost me. Um, let's keep breaking this down. So, yeah, something's going on with
[18:56] Chicago here. It's pointing out that this $13 per hour minimum wage is pretty much being hit by almost every city but Chicago. So, this is a very helpful visualization to see. But I guess, you know, is Chicago that important of a
[19:09] city? And it's even giving me this interactive breakdown of all the different cities and what percentage of deliveries they make up. And Chicago makes up a pretty big one, so I should focus on it. Oh, wow. It's it's really telling me Chicago is doing pretty
[19:21] poorly. I don't know what's going on in Chicago. So, it is worse on every axis. If you look at the late rate percentage by city, it is super high relative to the other ones that are in the mid20s, high 20s. Goes into a doom loop. And oh,
[19:35] this is pretty cool. This is showing me time of day um where there are issues. So, if I look at Chicago, it doesn't look like there's just certain hours that are an issue. It looks like I need to do some surge pricing across the
[19:47] entire day. And something's obviously going in in Miami late at night. Um, and this is amazing. I love this, but you know, before I do anything, I kind of want to know how much this is going to cost me. So, I'm going to ask Genie
[20:01] to include a costbenefit analysis on the instru incentive structure change it has recommended. Um, and while it does that, you know, I want to take some credit for this because I've put a lot of hard work in obviously. So, going to make myself
[20:14] the author. So, even while it's writing, I can write in it. Elise, I'll give you some credit here. Um, and let's see, it's already done. So, it added a costbenefit analysis section. Um, and the big
[20:27] takeaway is that it's going to cost I'm going to make money. $2.3 million annualized. You know what? I'll take that. Let's see where it is. Usually, it is outlined in green. Here it is. This looks great. I'm going to accept.
[20:41] But, you know, I don't know if I love the tone. I This is going to go out to some of my investors. So, I want it to be a little more professional. Okay. I asked Genie what kind of tones it offers up. Perfect. Board deck investor. Let's go
[20:55] with that. That sounds like me. Click that and let's see what it does. Um, and you'll see these kind of interactive questions throughout, which is u super helpful to ask you follow-up questions, and it's always in these little modals
[21:08] that are very user friendly. All right, so it restyled the dock in the board deck investor tone. It serviced the capital deployment thesis up front, and it reframed the doom loop section as compounding unit economics. That makes me sound pretty smart. Um, I can go
[21:21] ahead and accept those changes. Awesome. Back to you, Lee. >> Sweet. Now, documents live inside of Genie, but we know that a lot of your work doesn't. So, Genie now uses MCPs to
[21:35] write to the systems you already use. For example, commenting on a Jira ticket, creating a PR in GitHub, updating a record in Salesforce. And Genie uses your credentials. So, source
[21:47] permissions are always respected. And every action contains the same audit trail as if you'd done it manually.
[22:00] Cool. Let's see this in action. So, I would like to send a Slack to the owners up here. Let's go look at them. Who are our owners here? I want to send a Slack to all of these owners and I want to let them know that they need to take this
[22:13] action, what their deadline is, what the estimated weekly cost is. But you know what? I'm super busy. I want you to do all that for me. So, I've gone ahead and I've asked it to send a Slack to the owners of the action table and do all those things. Let's see how it does.
[22:27] It's thinking. It's done. It sent the message to Beet, Will, Jalet, and Meline, each with their action deadline and estimated weekly cost. Um, awesome.
[22:41] Now, a lot of those actions that you're taking are probably repeatable. And to that end, you can also now train Genie on skills. So it repeats that same action the same way every time. Once you've defined a skill, you can either keep it to yourself or you will be able
[22:54] to publish it to a Unity catalog skills registry so that others can take advantage as well. All right, let's see a skill in action. So get my skill ready.
[23:06] Um, you know, I've spent a lot of time on this document. We've um nudged the tone. We've made manual updates in it. I've even put together this like costbenefit analysis. And I'm going to want to send this out every week, but in
[23:19] seven days, I don't want to have to do this again. So, I'm going to create a skill that'll do that for me. So, I'm going to ask it to create a skill that uses this doc as a template to perform a career incentives deep dive over the last seven days. And here we go. It is
[23:32] done. Easy. In seven days, I'll just call this skill and be ready to go. So, what is the step up from a skill? Well, a skill can perform a task. An
[23:45] agent can perform a whole job. Genie agents bundle skills, instructions, context, and memory to create these expert co-workers that can answer questions and take action within a
[23:58] particular domain. And all this, you know, comes together so that you can really start to specialize what Genie can do. For example, you might create a renewals agent that helps you prep for customer QBRS, or you might create a
[24:12] launch agent that helps you track feature statuses, or maybe a finance agent that answers questions about your product's performance. And most customers are not starting from scratch here. And that's because Genie spaces
[24:24] are now Genie agents available in Genie 1 with a couple extra bells and whistles. We now support volumes. Um, and we are also adding MCP reads and
[24:36] writes to these agents and sub agent support is also in development. Awesome. So, let's see an agent in action. So, if we go down here, get my agent ready.
[24:49] Okay. So, zoom in. I I just sent a Slack to all those owners that have a bunch of action items for Lake Lunch and I don't want to have to stay on top of them. I don't have
[25:02] time to keep tracking the the deep dive action items. I don't have time to check how Chicago's going with the doom loop. So, I want Genie to do all that for me. So, I have asked it to create an agent
[25:14] using the cur incentive rollout tracker. I wanted to track the four deep dive action items and owners. I don't even need to tell them who the owners are. It's smart enough to know. Um, it's going to watch live courier tables to see whether each fix fix is actually
[25:27] moving its target. Um, it's going to flag anything past its deadline or not improving as a blocker. And this is even better. It's going to do the it's going to bug everyone instead of me having to bug them. It's going to send a Slack nudge to the owner with their Jira
[25:39] ticket linked if they're behind. And it's going to remember each item's owner and status across conversations. So next week it picks off where it left off. And look at this. It's it's asking me, you know, see what's blocked. Okay, I know I
[25:51] just sent this Slack out, but our team works fast. So let's see what's blocked right now. is checking the progress on the four action items. Let's see how they're doing. Some of the people are in this room. They better be working hard.
[26:04] Here we go. Uh oh, Chicago. Been I don't know if he's here. Okay, so looks like Bennett's behind. It's saying there's one blocker right now. Chicago still needs to have um the emergency surge
[26:16] done by end of day. And so it went ahead and it sent beet a slack nudge with a ticket link saying, "Hey, you need to get this done today. everyone else still has some time or is progressing along.
[26:31] Now, once a task works the way you want, you can set it and forget it. You can put it on a cadence, daily, weekly, whatever you pick, and you'll get the result right in your inbox. Event-based triggering is also in development. And
[26:43] behind the scenes, all this works on the LakeFlow jobs platform. So, you know, it's tried and tested and highly reliable. Awesome. So, I want to
[26:56] send this. So, I've worked hard on this report and I want to send it out. Hold on. Sorry. I want to schedule I'm forgetting my
[27:08] shortcut. I want to Okay, perfect. We've worked hard on that skill and I want to run the skill every Monday at 8 a.m. and send it in an email to all of the GMs. And again, I'm so busy I don't have time for this. So, I want this
[27:20] skill to run every Monday morning. I'm going to enter it. Let's see how it does. Uh, it needs to know who the GMs are. Let's tell them I've got I've got a shortcut for everything. Um, great. It's
[27:33] thinking about it. And it created the scheduled task. Done. The current incentives deep dive is now going to run automatically every Monday at 8 a.m. Pretty cool.
[27:49] Oh, sorry. This is still me. So, of course, that's great if you know what to ask, Genie. But what about the things that are happening without your knowledge? After all, most BI is poll based. You have to know to ask a question. You have to know what to ask, and you have to know where to look for
[28:01] the answer. Genie insights will flip that, servicing a feed of important metrics, a number drifting off its forecast, a spike in a particular segment, and it's personalized. So, for example, an account exec sees their own
[28:14] customers and regions. A product manager sees their key products and so on. And just as importantly, insights aren't telling you just what changed. They're also allowing you to click into the why. Which segment is actually behind a trend
[28:26] spike? And are there any leads I can go chase down? So, ultimately, Genie is no longer just a reactive tool. It's a truly proactive co-orker. And look at this. I just got a push notification from Genie. Hopefully, everyone can see
[28:38] it. Um, it says Lake Lunch. Something's happening in San Francisco's order and S orders in SF are up 2.7x over the last three days. It's the highest of any city. So, whoever's in town this week has expensive taste and a serious thing
[28:51] for pizza. Okay. So, here I am pushing in clicking into this push notification and it's opening up Genie in my pocket on my mobile device and it is it lands
[29:03] me right on the insight in mobile app and it picked up that there is an event driven demand occurring in downtown San Francisco and it's even attributing this to all of you to data and AI. Pretty cool. And then
[29:16] um finally, none of this is really all that useful if people have to leave their favorite tools to use it. So, Genie shows up where you already work. So, for example, there's a web app which you can now make your own. You can put
[29:29] your logo, your colors, a welcome message, and the assets that you want pinned up front. There's also support, like I just mentioned, for mobile on iOS and Android, so you can ask Genie questions on the go. This actually went out to public preview yesterday. So,
[29:43] please go check out your Google Play Store or App Store. If you feel like giving it five stars, we'd also love that. Um, and because Genie Mobile Genie mobile app is built on the existing data brick services, it inherits your exist
[29:55] existing security settings, but please note that the Mabel app may be the mobile app may be subject to your organization's MDM policies. So, please speak with your IT team if that's necessary. Um, and then of course there's
[30:07] integrations with Slack, Teams, and Copilot. And you can at mention Genie in a thread. You can also spin up skills and schedule them all right there. Um, and if your team happens to live in an
[30:19] entirely other agent, you can plug in Genie via MCP powered by Genieontology behind the scenes. And we're also working on support for Genie MCP apps, which will give you embedded charge,
[30:31] deep links, and so on within that thirdparty agent. Ultimately, the proof is in putting this into production. So, we talked to our friends at Unilever and they told us that today there are over 100 finance users at Unilever using Genie to reduce
[30:44] risk an analysis cycles from days to minutes. This has resulted in faster and better decisions and already delivered around $3 million in annual savings. And that's not all. Unilver is just getting
[30:56] started. It's currently developing an external data layer to complement all the data we are already providing. It'll contain economics, weather, competitors, commodities, and many other data points. And they will scale these solutions to
[31:08] more countries. Genie soon will be the key to delivering efficiencies to over 500 users at Unilever. So, as you think about putting this into production on your own data, you're probably going to want to make sure that
[31:20] you can manage it. So, we're going to talk a little bit about governance. We're going to start with controlling who can see what. Uh, permissions. Genie doesn't have its own permission system. It enforces the permissions of the data
[31:33] sources across you know Google Drive, Jira, whatever you have connected. So two people at the same company could ask the same question and rightfully get different answers each grounded in in the data that they do or don't have
[31:45] access to. And we keep those permissions fresh as the organization continues to change. So for MCP we do check the permissions live at query time and for the search index we are storing those
[31:57] ales alongside the data and keeping them fresh at a cadence that's essentially reasonable within the constraints of the sourc's API limits. We also offer robust source management so you can control what Genie can do. We are in preview
[32:11] with a unified console where you can connect, configure, monitor and disable all your different data sources for Genie and that includes granular control over which meta stores enable tools and even which tools you enable or disable.
[32:25] So for example uh the company might be okay with commenting on Jira tickets but not creating new ones and individual users can also constrain within that. So even if an admin says like yes genie you can send email an individual user might
[32:38] restrict it to drafting email behind the scenes all of this integrates with the Unity AI gateway which is a single even more powerful governance layer in UC that includes control over every tool
[32:50] that every agent is able to call among many other things like agent monitoring. So definitely be sure to check out the talks on that if you're interested in this topic. You might want to get even more granular with controlling what
[33:02] Genie can do with instructions.
[33:15] controls. Starting in July, Genie is moving to a pay as you go model. So there are no per seat charges. You're simply paying for what people actually use build in DBUs. and we are removing
[33:27] the old query rate limit so Genie can really scale as your organization continues to use it. Along with that, we are giving you a little over $10 of usage per user per month at no cost. I
[33:40] think you heard this morning that that covers a pretty large percentage of uh of a of a typical user. Uh and then for usage beyond the free allowance, admins can set alerts and hard caps via again the Unity AI gateway. So you can grant
[33:54] higher limits to power users or to power groups and then you can track all of that spend via system tables. So that wraps up what's new in Genie 1. If there are three things to take away, it's really context, outcomes, and
[34:07] governance. Uh Genie knows and understands the business through the Genieontology, and you can connect it to hundreds of data sources at this point. It also turns answers into work with docs, MCPs, skills, agents, scheduling,
[34:20] and more. And you can really do that work wherever you want. And it's governed by design using the same governance structure that you may already have in your lakehouse today. So we'd encourage you all to try it and we are super eager for your feedback as we
[34:32] continue to build this together.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.