Latest innovations in BI and AI for business users

Summary

Databricks is unifying conversational and visual BI into a single integrated experience called Agentic BI, combining Genie Agents, Genie One, and AI/BI Dashboards all powered by a shared Unity Catalog semantic layer.
The Unity Catalog semantic layer—including Ontology, Metrics, and Glossary—provides a single source of truth so Genie and AI/BI Dashboards always return consistent, trusted answers regardless of which interface a user chooses.
Genie One integrates with external systems through MCP read connections, enabling business users to query across their full operational and analytical estate in natural language from a single unified interface.

Latest innovations in BI and AI for business users

Databricks is setting a new standard for BI—combining unified semantics with AI-powered agents to truly deliver on the promise of Agentic Business Intelligence.
Whether through natural language with Genie, AI/BI Dashboards, or a combination of both, users can move seamlessly between conversational and visual analysis in a single, integrated experience.
In this video, we’ll cover:

An end-to-end walkthrough of agentic BI on Databricks
The latest innovations and customizations capabilities to simplify access for business users
How unified semantics power consistent, trusted answers across Genie and AI/BI Dashboards
How agents, Genie, and dashboards work together to accelerate analysis
How conversational and visual workflows come together to accelerate insights and actions

Talk By: Chao Cai, Sr Director of Engineering, Databricks ; Miranda Luna, Director of Product Management, Databricks ;

Chapters

FAQs

What is Agentic BI on Databricks and how does it differ from traditional dashboards?

Agentic BI on Databricks combines traditional AI/BI Dashboards with conversational Genie agents, allowing users to move seamlessly between visual and natural-language analysis in a single integrated experience. Unlike static dashboards, Agentic BI can schedule tasks, automate knowledge work, and take actions based on analytical findings.

What is the difference between Genie One and Genie Agents?

Genie Agents—the rename for Genie Spaces—are domain-specific agents grounded in a particular dataset or business context, designed for deep analytical work within a defined scope. Genie One is a broader interface for business users that can operate over the entire data estate and connect to external systems via MCP read connections.

Why is the Unity Catalog semantic layer essential for trusted BI answers?

Without a shared semantic layer, different tools and users define metrics differently, producing inconsistent answers that erode trust in AI-generated insights. The Unity Catalog semantic layer—comprising Ontology, Metrics, and Glossary—provides a single authoritative definition that both Genie and AI/BI Dashboards use, ensuring consistency across every analysis.

How are customers like Rivian using AI/BI Dashboards on Databricks?

This video references Rivian as an example of a customer using AI/BI Dashboards in production, though specific implementation details are not provided in the available content. The session presents AI/BI Dashboards as supporting natural-language authoring and embedding scenarios, making it easier for teams to build and distribute data products to business users.

Full transcript

[00:07] Well, welcome, welcome. Thank you guys all for joining us today. My name is Miranda >> and I'm Ciao >> and we are super excited to be chatting with you today about some of the latest innovations in uh BI and AI for business users. So, what we're going to cover
[00:20] today um is a couple things. First, if you haven't, you know, believed us before, please hear us and believe us now. We're all in on AgenticBI. Um you should have seen us pop in on the Gartner magic quadrant. Um we say
[00:33] agentic BI because it's not just traditional dashboards and reporting. So of course we're going to have that. We do think that that kind of artifact is also going to evolve a bit. So we'll we'll get into those details. Um but also the Genie suite. So you saw a lot
[00:46] of that in the keynote yesterday whether it was Genie one kind of as the interface for business users operating over the entire estate. Uh Genie agents which is the rename for Genie spaces. And I promise we'll explain why. Um,
[00:59] we're going to start giving out, you know, a nickel every time we rename. So stick with us, but we will try to tell you why that's happening. Um, and then semantics. Ontology was a big announcement. Glossery was a big announcement. We're going to break that
[01:11] down a bit. Um, as well as a uh look kind of across the board. So if you're kind of looking for, hey, what's the session to get a sense of everything end to end, you're in the right place. Um, all of the topics we're going to hit
[01:23] today are going to have much deeper dives available in other areas, but we think there's still some value in kind of bringing it all together for you. And of course, we're here to answer some questions at the end. Awesome. So, everybody has to have the
[01:37] updated architecture slide, right? when we start to think about kind of what does everything look like as you build up at data bricks you know it all starts from the infrastructure everything on kind of the open format uh data lake with delta and iceberg support layering
[01:49] on uh data warehousing lake flow for streaming ETL lake base obviously critical for kind of any of the the new workloads we're talking about unity catalog applies across so I think that's
[02:01] you know something to be really really clear about right unity catalog is the foundation of any sort of unified governance and semantic approach Of course, you can federate to other systems. That hasn't changed, but everything that we're talking about in terms of Genie, in terms of AIBI, um
[02:15] that is going to all be kind of the foundation beneath all of those. We did uh talk about Genie a lot today, but just to level set, you know, we talk about Genie one as that kind interface
[02:27] you saw for uh business users to log in directly to get any sort of knowledge work done. You know, I'll be transparent. We use it quite a bit outside of just the business user group at data bricks because it is so kind of helpful in a number of different areas
[02:40] from kind of not just doing some basic analysis and not having to traverse a million different places to find the right the right area to answer your question but also producing some of those artifacts and next steps like you saw in the keynote to schedule a task or
[02:52] to then take another action after you've done that analysis you know MCP reads are live writes are coming shortly thereafter um and then that of course working handinhand with genie agents right we why do we do the kind of big Genie one
[03:05] announcement if we had Genie spaces and agents? Well, we still think that there is a really important place for those kind of very targeted agent use cases. You know, if you're building a benefits bot on an HR portal, it's going to be a lot faster um and higher accuracy to
[03:18] have something kind of very semantically curated and targeted towards that use case versus pointing at kind of the entirety of the world via Genie1 MCP. Um and then Genie code of course that is somewhere where we're investing across
[03:30] the board and using that to kind of really help anybody who's building uh different artifacts in the data bricks platform and then you'll see those kind of punctuated throughout all that application layer including AIBI for business intelligence.
[03:42] So to start a bit with some of the semantics pieces, you know, certainly I think that this is one of those things we really hit on a lot last year, but uh we did, you know, just because it wasn't the headliner this year doesn't mean
[03:54] we're asleep at the wheel, right? We still believe that unified data, unified governance, moving those definitions of your semantics as far we say left or upstream as possible is going to be what's allows you to be successful in
[04:06] having a single source of truth um in terms of making sure that nobody's coming to meetings with different uh numbers. And when we say nobody, we now mean no agents uh who are empowering us coming to meetings with different numbers. It was hard enough when there
[04:18] was a person who maybe had some extra expertise in their mind and was able to look at it. But now if you really want to kind of actually maximize the potential of agents, they have to have the right context. We talked a lot about that in the keynote today. They also
[04:30] have to have, you know, the right uh the right uh single source of truth and not be in a place where they're kind of coloring outside the lines too much. There's a couple great stories from some of our customers that we definitely
[04:42] encourage you to take a look at. Um, you know, Zando and Mercedes-Benz have done quite an investment in kind of the metrics layer in data bricks. Um, and are definitely kind of articulating some of the benefits that they've achieved.
[04:55] But, you know, what are some of the things we added in the last 12 months? I promised you we were busy. We weren't just sitting on our hands eating bon bonss. Um, big investment in metric views. So level of detail calculations, window functions, parameters just about
[05:07] to come out, relationships. I say about because as soon as we get to data and AI summit, some of the release stuff gets fun. Um, but the idea here is to make sure that kind of both at the BI layer um, and at the Unity catalog layer, you
[05:20] have the same capabilities available, that they're interoperable. You know, Unity catalog should be the source of truth for everything. Unity catalog isn't always the quickest place to iterate. So we want to give kind of folks that last mile iteration ability
[05:32] and then the ability to promote back on the performance side. You know definitely invested in materialization um and the kind of integration with Raiden or Lakehouse RT that we talked about a bit in the keynote yesterday and
[05:44] then ensuring that there's kind of the ability to start from anywhere dashboards Genie agents genie code um as well as use anywhere and of course business glossery and ontology are kind of some of the big big announcements. We're going to walk through those in a little more detail.
[05:58] So all starts with open connectivity. Again, we layer on Unity catalog semantics. Then we have this new thing called geneontology. So I'm going to step through this really quickly. So within Unity catalog semantics, you
[06:10] know, this time last year, we were talking about just metrics, right? Traditional semantics, dimensions, measures. Um, and those were reusable KPIs, super important to make sure folks weren't kind of coming with conflicting information. But this year, we've
[06:23] layered on domains. That was an announcement because it came out a while ago. Um, but that is kind of a cap an ability to kind of organize all of your assets. Um, whether that is kind of the actual semantics, whether that's the objects that use them, whether that is
[06:36] certain uh other data sets that are critical to kind of understanding that space. Um, not just domains but also subdomains to kind of have that be the organizational uh mechanism to make sure everybody's finding the right uh assets.
[06:49] And then glossery, uh, that's the new one. We're going to talk a little bit more about that in detail. But again, that's being able to kind of define concepts and taxonomy. You know, we heard loud and clear that there wasn't really a home for that in a good consistent way. So, we really wanted to
[07:02] make sure that came to Unity catalog this year and that it was going to be able to be kind of that next piece of the puzzle to providing the right context to both humans and agents. Now, the geneontology is not a manual layer. It's inferred. So, I know this we
[07:15] went through this a little bit quickly in the keynote yesterday, but um what we're really doing with the ontology is kind of trying to make sure we're leveraging all the context that exists in the platform to autolearn behaviors that can then inform agents. Now, we
[07:28] showed it in Genie 1. Um but the really important thing to note there is that this is not just operating over data directly in Unity catalog, but anything available via MCP. Um, so what you're able to do is actually get a complete picture uh leveraging that ranking
[07:41] function that's going to say, "Hey, I this is the right way to answer this question because of all the context I've seen of how it's been answered in other places, by other people, by other agents and then of course consumption of those
[07:54] from anywhere. So when we think about kind of the stack, it goes from unstructured to kind of layering on metrics and BI. Uh, and then we start to think about the knowledge side. Now there are both a combination of defined and learned metrics. um in that
[08:07] knowledge stack. So, quickly hitting some of the metrics, you know, one of the things we really want to make sure everybody's hearing is generally available generally available in Unity catalog, generally available at the BI layer in dashboards, you know,
[08:20] measures, dimensions, relationships. So, we call multiffactor relationships. So, anybody who is waiting on that, that's available, you know, ensuring that there's support for uh not just building but maintaining those. We have agentic
[08:32] authoring. uh if you're doing a BI migration uh we not just support that kind of helper in Genie code but we also will autogenerate those semantics for you so you don't have to kind of try to translate them yourselves layering on
[08:45] materialization to ensure we have excellent performance and then making them available from anywhere you want to leverage them. Now one of the things I want to call out here and again I know you guys can all read so I won't read the slide for you
[08:58] but that certification piece is so so critical. So, one of the things, you know, I've had in a lot of conversations with some of you this week has been, well, how do I kind of bubble the right things up to the top? How do I make sure that, you know, all of the behaviors are
[09:11] leveraging the right data sets or leveraging the right assets, etc., etc. And you're going to do that by leveraging certification because that is going to give higher waiting. Now, I mentioned multiffact. That was one of the biggies uh that folks were
[09:24] asking about last year. So, want to make sure you all hear uh already there in private preview for dashboards. It should be out, I think, public preview next week. We're just ironing out a little UI fun. Um, and then UC will be following shortly thereafter. There's
[09:37] just a little bit of a DBR release train we have to catch with UC, which is why it's slightly behind dashboards. Um, but those should both be there and kind of working in concert with each other very shortly. A little preview of kind of how that'll look like to define that. Hopefully,
[09:50] this is quite familiar for some of you folks who might be coming from on some other tools. Um, and you can also see that we note in the data set panel if you're using this in a dashboard. you know when I'm using this data model uh where exactly is local what is kind of
[10:02] coming through from UC as well as anything that has materialization applied with these little check boxes. So the nice thing about this, right, is that you're naming the relationships once, right? You don't have to hardcode
[10:15] every single path as you might be more familiar with with some other tools. Once you kind of define that relationship once, the agent can pick the right path. New questions don't require you to model change and kind of put that extra permutation of the other
[10:27] hard-coded uh path. And again, this is the same graph that's being used across the entire estate. So if you're in an app, it's using the same thing that Genie agents are using, using the same things that dashboards are using, using the same thing uh that Genie1 is using.
[10:42] The glossery is a big one. So this is again one that we're really excited about because if we looked again last year, there wasn't really a home for this type of knowledge. So we saw, you know, you guys all being very creative and doing the things we know people who
[10:55] love to build do, which is try to kind of be creative in where some of this information lands. Some of this was getting squeezed into domains. some of it was getting squeezed into kind of meta table level metadata. Um, but here we really want to make sure that there
[11:07] is kind of like an authoritative like wiki type glossery experience to kind of enhance some things that may have maybe uh otherwise would just be on kind of a a very singular piece of information and
[11:19] metadata. So, the geneontology, I know you saw quite a bit about this, but we're very excited about this. And I have to tell you, uh, I've personally been very impressed by how accurate it is. Uh, you
[11:32] know, we use this heavily in kind of our own internal, uh, stack. And, you know, every time I ask a question in Genie1, I see the ontology off in cooking and then I, it say flashes something, I'm like, ah, that's wrong. I'm going to get it this time. And then it selforrected
[11:45] itself by the time I go and like expand the whole thing to really scrutinize it. I'm like, I'm going to report a bug. I'm going to report the ontology was off. And, you know, uh, our tech lead on that project, BMNet, has always beat me to it. I'm like I'm going to get him one of
[11:57] these days but not yet. Um so we are quite a cool it is quite cool to see uh again you know one of the things I think has been a question about the ontology is well if this is being automatically generated what if I have other
[12:09] information in other systems is it autosyncing over constantly um but the thing I would say is like even if you've defined certain things in other systems if like the bulk of user behavior reflects that guidance you're
[12:22] not going to have to be super pedantic about it. Um, and that would be one of the things I'd say was a little bit more of like a aha like wow this is pretty cool moment. Um, of course you can't just trust you know one bad job or one user behavior one bad user behavior like
[12:35] on its own but when you look at the totality of kind of all of the activity um it's pretty impressive to see how well we can figure out filter out that noise uh that bad signal um and make sure it doesn't become noise for the
[12:48] overall answer. Um, so that's something that's pretty cool and I would definitely recommend giving that a pressure test very shortly. Awesome. Um, child, do you want to talk about
[13:00] dashboards? I realized I might have jumped in on some
[13:12] filtering parameters, drilling, uh, embedded analytics, uh, custom visualizations are out. uh parameterized reporting. So all the things you kind of know and love and expect from kind of some other tools are available. Uh of
[13:25] course powered by Genie. This little GIF is showing the kind of enduser experience in terms of I want to ask some follow-ups. I want to keep insight summary. Let me refine the context of this particular visual and get to
[13:37] another answer. Um, but Genie is also being leveraged super heavily in the authoring experience too. And not just the let me create from scratch or let me adopt this new feature or like hey can you try to optimize my data model. Those
[13:50] are all great use cases. Um, but also for supporting kind of uh migration from other tools via the import BI function. Um, so that's something we've seen folks be able to leverage pretty successfully.
[14:06] folks like Rivian um who recently shared kind of one of their great stories with us around kind of a migration they did from another legacy system and some of the benefits they were able to net right again I think the thing to remember is it's not just like this sheet of visuals being converted to that sheet of visuals
[14:19] it's really kind of a migration to the entire ecosystem um and being able to benefit from kind of all of those stack we just walked through not kind of an isolated well this bar chart is now recreated like yeah that's great we want you to do that but um it's really kind
[14:33] making sure that you can benefit from the totality of the platform, not just individual uh bits and pieces. What else have we done in the last year? Uh lots of bells and whistles. So, we like bells, we like whistles, you know,
[14:46] we like gant charts, we like themes, we like bookmarks, we like doing all the things that are going to kind of let you make this all match your business completely. Um, you know, we're really excited to keep um keep shipping more
[14:59] capabilities out of the box even as we make some other uh other investments. So, you know, we're not asleep at the wheel. Please keep an eye on our release notes. They're fun every week. We spend a lot of time on them. Uh mentioned kind of migration and Genie support already.
[15:13] Uh Genie Code support. So, you can see that now kind of pretty trivially you can describe the exact uh like outcome that you want and trust Genie Code to take you through it. Uh, one of the things I would mention here that's super
[15:25] super critical is that um, you know, Genieode is not just going to kind of say, hey, I'm going to give you the the top layer thing you described. It's going to try to actually make optimized decisions below the se below the
[15:37] surface. So, it's going to try to make good data model decisions for you. It's going to try to guide you uh towards kind of the right combination given what the outcome you want is. It's not going to need you to be extremely literal in terms of saying, "Oh, no. to make the
[15:50] most optimal data model. Take these 75 steps and review these 15 things. It's going to do that already for you. Customization was a big one. So yes, there's I think this came out last week. There's you can round every corner. You
[16:02] can, you know, change every padding. Uh so please, you know, leverage that as much as needed to kind of make sure that anything you're you're building matches your corporate branding that m matches kind of what the look and feel folks are expecting. Um, you know, same as always,
[16:16] this stuff is available at the dashboard level, but it's also available at the workspace level. So, you can set it once. Everybody doesn't have to be a designer. Uh, our designers would like to keep me out of the workspace level theme, but I persist nonetheless.
[16:30] Um, but one thing I do want to give you guys a heads up on, right, we did a lot of bells and whistles. There's that whole rollup slide, but you know, are we ever going to be able to do them all fast enough? There's uh there are some tools that have been doing this for 10 or 15 years. How are we going to catch
[16:43] up on every little configuration? That's a great question, Miranda. Um, quick quiz though. Is this a dashboard or an app? Who thinks it's an app? Yeah. Okay. Half. Who thinks it's a dashboard?
[16:57] Oh, very few. Okay. All right. The doubters are here. Uh, this is a dashboard. Um, but I set you guys up a little bit. Um, I set you up a little bit because I think that one of the big insights we had in the past year was not
[17:10] that the visual layer is what's defining an app or a dashboard. It should be everything below it. So, this is actually something that we've built to kind of showcase what you can do and to match the end experience and dashboards
[17:22] and apps. Um, and you know, historically these have been very very separate. There are different layers to them. The visuals, the front-end presentation layer has been separate. Whether you can read and write has been a different story in either artifact. The backend
[17:36] has been very different in terms of turnkey services versus very bespoke and you know security as anything goes in an application uh pending review oftentimes a monthsl long review. But in dashboards you know we usually see a lot broader
[17:50] access to creating and sharing because they're not going to color outside of a very defined set of lines. And when we were kind of looking at some of the trends, you know, seeing folks that were starting to leverage tools like uh, you
[18:02] know, cloud code and streamllet together to get the exact visual output that they wanted, you know, we kind of had this insight of like that's great. We want them to be able to do that in data bricks too, whether it's a data bricks app, whether it's a data bricks dashboard. Um, but that layer isn't
[18:16] really what we think distinguishes apps and dashboards, right? It's everything below that. It's not the top of the iceberg. It's everything below the water. So once we had that insight, you know, what we really wanted to push on was making sure to that the visual layer
[18:29] wasn't a reason for the tipping point between one or the other because we were starting to hear that like, yeah, it's great everybody can use cloud code. Everybody can use streamllet. Everybody can make the visuals they want, but then they don't have a scheduling service. You know, I have a thousand different scheduling services being spawned over
[18:42] and over and over again. I have no idea what permission model they've included in the dashboard. I have no idea if it adheres to these different policies around data download and xfiltration. Um, so with that insight, one of the things that we're doing and investing in
[18:55] quite a bit and I kind of led the witness with that that little gif is making sure that front-end layer is going to be one wholly extensible in dashboards and then two shared between apps and dashboards so there's much more
[19:07] consistency. like I we we'll play that fun game of is it an app or is it a dashboard a hundred more times and we really want you to not be able to tell um from the UI layer because we think it's important to be able to build exactly what you want as the
[19:19] presentation layer no matter what tool you pick but if you really want to kind of get to a place where you know you can unleash the potential of humans unleash the potential of agents uh we want to make sure that you have everything below the surface like really well controlled
[19:32] um in a dashboard for those cases um and as we see trends where agent uh agent prompted uh agent-driven creation is more and more of kind of the activity. You know, you can't really at least if it was my
[19:45] enterprise, I wouldn't set agents totally loose with no guardrails. And so that's where we think that, you know, that flexibility at the visual layer will get the right artifact um for the consumer, but all that detail that's kind of a little bit messy, but super
[19:57] important um to the administration folks will be taken care of behind the scenes. So you'll see both options are going to stay available. Dashboards are going to be evolving to be incredibly flexible at
[20:10] the presentation layer. We're going to share with our good friends in apps so that it's not kind of a choice driven by can I have a custom layout or not. It'll be a choice driven by what is the right decision for my services backing this
[20:22] presentation layer? What are the right data model and security elements to back this layer? Even I this was a fun one. We made this because one of the the most common requests we were getting is how do I
[20:34] make like a nice fun drag and drop pivot table UIUX like that's a common thing I had in something else. Um so it's one of the test cases we use. You can see that you can do this now with that kind of extensible layer.
[20:46] Awesome. Ciao. Let me give you a few minutes to chat about genie agents. Genie one. >> All right. Okay. So we'll start with genie agents. Uh what are genie agents? You heard about them yesterday but uh if you've played with genie spaces actually
[20:58] show of hands. How many people here have used Genie spaces or deployed one before? Very nice. Okay. When I asked this two years ago was barely a single person. So very nice to see. So Genie agents are what you know and love about
[21:11] Genie spaces but more. What where what do you get more out of? Well, first in addition to structured data, we are adding unstructured data. So you can bring in documents, you can bring in uh whatever sort of unstructured data you
[21:23] have that you would like the agent to be grounded in. Um, but also we're bringing MCPs along the way uh fairly soon where you can also just hook it up to say Google Drive or Alassian etc to bring in
[21:35] even more context or potentially even to take action coming out of what you want to do with these agents. Um, and they're otherwise the same setup and the same performance. We know a lot of you really like the fact that there's a very uh
[21:49] easy to get going, easy to man manage Dev Loop in being able to test this consistently to be able to monitor what users are doing with this. Those are all still going to be here. You just get to do a lot more with it.
[22:02] Um 7-Eleven, lots of really good success stories there. You might have seen some of their stuff out there. uh and uh Virgin Atlantic also lots of uh very
[22:14] happy customers in there as well. So we want to flash those. Um in terms of the features that we've recently released, uh we in in the last 12 months we've introduced agent mode. So not only is Genie agents able to answer your sort of
[22:28] very standard Texas SQL what kind of questions that are retroactive, it can also do forward-looking sort of how and why type questions. You can ask things like, well, how is this business going to do tomorrow? Or, hey, why did this
[22:40] drop so much in the last month? Or, hey, help me think about what I should be doing next. So, those kind of questions are what gets facilitated in agent mode. Um, there's a we we've been listening to a lot of your feedback to try to
[22:52] streamline what that dev loop looks like so that you can go from your idea about what the agent should do to actually deploying out to a few people to deploying out to the whole organization. Uh and of course we've also added a
[23:05] whole lot more visibility to all the sort of all the conversations so that even the consumers can see what's going on with the agents and what assumptions they're making. Uh and uh this is a big one. A lot of folks have been really
[23:18] excited to get agents not just within the day UI but also into Slack into teams into their own websites and their own team portals through an iframe as well as APIs or anywhere else that we
[23:30] haven't taken care of. One big thing we've heard as feedback is, hey, actually APIs, uh, they're great, but this whole five queries per minute thing is really hampering us. It's really
[23:42] keeping us from being able to deploy to a lot more people than we'd like. Well, that's why we introduced this notion of pay where uh, now you can actually uh, really arbitrarily scale these APIs to whatever you want without having to pay
[23:55] this very large fixed cost to always keep the extra uh, tokens around. you can basically pay as you go and only pay for the tokens that uh you're you're actually sort of increasing peak traffic for. So hopefully that will allow you to
[24:08] scale out to thousands tens of thousands of users a lot more easily. Um of course lots of people want APIs, lots of people want dabs. They're like I want to programmatically manage my 10 genie spaces. I do not want to click through
[24:20] them one by one. So we heard you. We built those. And uh as I mentioned unstructured data and MCPS are coming. Uh I mentioned this but uh Genie agents we want to make them available wherever
[24:33] your users are. We do not want them to just stay in the data bricks UI on a laptop. Uh that's why we've introduced the mobile apps. Uh personally I didn't even bring a laptop to the summit because I've just been doing everything on the mobile app in Genie. Uh we wanted
[24:46] to make sure that it's available in Slack and Teams. You can just add the data bricks slack bots to ask questions and of course embed it anywhere else you want. All right. So now I'm gonna hop over to
[25:00] Genie One. Um because uh we've introduced a lot of names at you in the last year and I want to make sure you all know what this is. If you heard of data bricks one last year, we turned data bricks one into Genie and then we
[25:13] slapped the one at the end and so now you have Genie 1. Um and the reason for this transition >> t-shirts got reprinted. >> That's right. That's right. Yes. Um, and the reason why we did this transition is because when we first introduced data
[25:26] bricks one, it was meant to be your one-stop shop for all your dashboards, all your Genie spaces, all your apps. And immediately the first piece of feedback we got from a lot of customers, especially in the last couple of months when a lot of folks are starting to use
[25:39] all these other agents to do their work is I don't want to actually look through any of the stuff. I don't actually want to click on a bunch of dashboards. I just want to start asking questions. Don't make me do anything else. I just want the first thing I see every day when I log into this thing to be a giant
[25:53] box where I can type in whatever I want and you'll just do the work for me. And that's what Genie one is. So the way this all comes together is you've got the bottom layer which is the
[26:06] Unity catalog, business semantics and glossery. The business semantics, the metric views that you might already be curating, the relationships you're defining on top of that that gives you the structure context that you're curating. You're basically ground saying
[26:19] this is what's certified. This is what my organization believes in in terms of how to calculate metrics, how to actually query from tables, how to do joins, etc. Glosseries is the unstructured complement to that. It's basically a knowledge graph of all of
[26:33] the concepts you've got across the organization. Here's what fiscal year means. Here's how I'm organizing my regions. Here's this really weird obscure term that no one else knows about but my team. All of those things
[26:45] will live there. And you don't want to declare them five times across five tables just because there's five tables that all use the same column name. You want to declare once here. And so together those give you the foundation, the building blocks to say this is what
[26:58] we've curated. This is what our team believes in for now. But usually that layer is going to be slow moving. No one wants that to change every day because if it changes every day, your users are not going to keep up. Um and also your
[27:12] decision processes are not pro probably not that fast. So we build geneontology on top of that to keep up with all the real changes that are happening across your organization. All the notebooks that are being written, all the genie
[27:24] queries are being asked and the the the sort of the clarifications that are being done in there, all the dashboards that are being built where they've defined a new version of churn that you haven't quite approved yet. uh all those
[27:37] things are actually happening in your data bricks workspace today and they're not going to make it into Unity catalog fast enough. So, ontology will infer all those things for you so that the next
[27:49] time someone asks about, hey, um, can you calculate churn for last week, it's going to say, hey, actually, you have a churn definition in Unity catalog, but you've also got three more in these three dashboards that all say different
[28:02] things based on different departments. Which one did you mean? Which one do you want to believe in? And it may be that because you're working with a particular department that doesn't want to use the official definition, that might be the right one. So we'll give you the option
[28:14] to do that and then building on top of that the dashboards and AIBI side of things will give you the visuals so you can do the monitoring so you can do the rich reporting. The Genie agent side
[28:26] will give you the curated domain specific chat bots. A Genie one sits on top of all of that so that you can actually just get one place to point all of your users to and they can just start
[28:38] doing whatever knowledge work they want. So, let's go a little deeper here. Um, in terms of the things that we've added on top, one
[28:50] of the things I want to call out is MCPS. Um, previously you were only able to use whatever is sitting in Unity catalog, all the data that's sitting in there, all the metadata that's sitting
[29:02] in there. But what if you want to like, you know, ask a question like, "Hey, schedule uh given all of the meeting customers I'm meeting next week. Give me a summary of how much they're spending
[29:14] on us." Well, that would have to go into Google Calendar or whatever Outlook or whatever system you're using, pull all the meetings, find out what the customer names are, go and join with your data, etc. Well, now you can actually do it
[29:26] all in Genie1 because you can just connect it to the Google calendar MCP, read in all that data and the rest of the analysis is inside of data bricks. We'll build search indices on top of
[29:38] that to make all of that fast. And of course, we still have connect. We still have custom pipelines, which means if you happen to have some other data that's sitting in some very bespoke
[29:50] system that you'd like to bring in and you'd like us to house and analyze and serve, we can always do any of that. I want to mention skills. How many people have used skills in various other tools? All right. So, these are super
[30:04] useful because uh if you happen to be doing an analysis and you're saying, "Hey, actually, uh I would like all my decimals to be always be formatted to two places or I would like uh if I ever
[30:18] say fiscal year um I would like the marketing team's definition of that or something." um all your various personal preferences is sort of going to be where skills are
[30:31] most useful and there's an additional benefit to that which is if you save a particular analysis as a skill the next time Genie does that analysis it won't go through all the exploration and fail
[30:43] paths and all those things it'll just jump straight to here's how I did the analysis in the right way last time and so your answers are going to come back faster as well so skills are already available they are um currently
[30:55] available for you to sort of save and and revise on your own and use. Uh we're also in the middle of introducing a skill registry so that you if you have a really cool skill you want all of your friends to have you can also share that
[31:08] through Unity catalog as well. Um and uh and then on top of that what we're actually introducing is this concept called documents. Now we've already got dashboards. So you're asking why are we introducing another thing that looks
[31:21] halfway like dashboards? Well, the reason for that is because dashboards tend to be this sort of highly curated uh artifact that's meant for monitoring, for always updated refreshes of data,
[31:34] all of those things. But what if you just want this snapshot that's a little better than a PDF where you can sort of share that out to your colleagues? They get to actually interact with the charts. They get to actually look at uh
[31:47] all the sort of rich reporting. And that's what we introduced natively inside of Genie1. So that every time you actually do an analysis, we'll generate a summary that says here's the results, here's some interesting graphs, etc. You
[31:59] can save that to a document and then you can share that out to all your colleagues for them to also sort of explore as well. And we are exploring what else do you want to export that too. So if you have ideas, let us know.
[32:13] I mentioned Genie agents in the last section. I'm going to mention it here as well. One of the reasons why we call this agents is because a lot of folks are basically now using Genie1 and
[32:25] saying, "Hey, I just did this analysis, but what if I want everyone else in my team to be able to do the same analysis?" Well, that starts to look very similar to how Genie spaces started coming about. Basically is the same
[32:39] pattern of I have a set of tables. I have some context behind it. I may have some specific assumptions. Can you just package that all up and make it easily sharable? And then can you make it easy for me to actually see how people are
[32:51] using this? That's a genie space where I guess now it's a genie agent. And so you can literally inside of Genie ones once you've done analysis say turn this into an agent and we will bootstrap a genie agent for you. And then from that point
[33:04] forward you get to manage it just like you would with any existing Genie space. And this is a private preview. We are working on how to make sure that this is as safe and as usable as possible so
[33:17] that you don't accidentally go and delete all your emails with it. Um we want to make it possible for you to not just read and do analysis and have things stop there and then having to open up three other tabs to go and
[33:30] actually take action on it. We want you to be able to just take action in Gen1. And so we are building integrations to all these systems to do things like you know do an analysis on which customers
[33:43] are likely to churn and then go schedule meetings with them in Google calendar for me or uh send a slack message every morning with a summary of the top gains and losses that I've seen from the last
[33:55] week or hey actually uh the folks that uh I would like to share this out to are this team that I've never met before. So why don't you go and send an email to them and introduce me first before I go
[34:08] bug them with a million questions. So all those things you could actually do through Genie1 once we've introduced the right MCPS in there which is fairly soon. And of course hopefully everyone here is
[34:21] as lazy as I am. I never want to do the same work twice. And so I would like Genie to do all the work for me. And now you can basically say, "Okay, I've done this analysis once, but I want to do it once a week or once a month. So, hey,
[34:35] Genie, just schedule this as a task. Monday morning, 8 a.m. Send it to me." And it's done. Putting all together, these are the kind of things that you can now do with Genie
[34:47] One. For example, give me a summary brief of all the customers I'm meeting with this week. I literally did this Monday morning because I realized that I had no context on half the customers I was meeting with and I wanted that context quickly. So I basically said,
[35:00] "Tell me about all the customers from this week, including how much they're spending on each product line, recent emails, meeting notes, quarter over quarter growth patterns, and honestly anything else that you might find interesting." And sort of Genie spent a,
[35:13] you know, a few minutes thinking about that, going through all the relevant information and gave me a pretty good report. And then I asked a few follow-up questions and got more details and off I went. Made my meetings a lot easier. Uh
[35:25] or you can do things like schedule a task to send me the updated version of the monthly business review doc at this particular Google doc link. I we literally pasted a Google doc link there that had a bunch of numbers in there and
[35:38] said go update these numbers, refresh them for the next uh month and it just did it. And you can schedule it to do that every month and send it to you. um what used to be a giant data science exercise is now a few minutes of having
[35:51] the agent think about it and do it or create a group DM with owners of these three conflicting definitions of churn so I can actually figure out which one's the right one done so I think the challenge to this
[36:05] group is what are you going to use gene one for and what do you think is going to be the most interesting

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.