From Search to SQL: How Glean Tap Databricks Genie for Everyday Analytics

Summary

This video demonstrates how Glean's AI assistant and agents connect to Databricks AI/BI Genie and SQL actions, routing natural-language prompts to live governed warehouse queries and combining the structured results with unstructured context from emails, documents, and wikis.
The integration is permission-aware, ensuring each employee sees only data they are authorized to access, and supports indexing Databricks artifacts such as dashboards, tables, and Genie spaces for enterprise-wide discoverability.
Concrete architecture and prompt patterns are shared for scaling self-serve analytics without adding data team workload, using Databricks as the governed data layer and Glean as the enterprise knowledge and search layer.

From Search to SQL: How Glean Tap Databricks Genie for Everyday Analytics

Modern enterprises need both the “what” from governed metrics and the “why” from surrounding context. We’ll show how customers use Glean Assitant and Glean Agents with Databricks AI/BI Genie and SQL actions to turn natural‑language questions into live warehouse queries that any employee can run. A single agent can route a prompt to Genie, execute the generated SQL on Databricks, then combine resulting tables with unstructured context from email, docs, tickets, and Wikis in one answer, all while respecting app‑level permissions. Attendees will leave with concrete architecture and prompt patterns they can use to scale self‑serve analytics without adding more load on data teams. Audience takeaways: Pattern for building Databricks-powered agents that run live, governed queries without extra indexing Examples of combining Databricks results with unstructured work knowledge to deliver a complete story Practical guidance on permissions, debugging Genie output, and tuning Genie spaces for enterprise copilots
Talk By: Joel McKelvey, VP of Product Marketing, Glean ;

Chapters

FAQs

How does Glean route a natural-language question to Databricks Genie?

A Glean agent receives a natural-language question and routes it to Databricks AI/BI Genie, which generates and executes the appropriate SQL against the governed warehouse. The resulting data is then combined with unstructured context from sources such as email, documents, tickets, and wikis to produce a complete answer within a single response.

How does the Glean and Databricks integration enforce data permissions?

The integration is permission-aware at the app level, meaning Glean respects both the access controls defined in Databricks and the user's permissions within Glean's enterprise corpus. Employees only receive data in their answers that they are authorized to view, preventing sensitive information from appearing in cross-functional responses.

What Databricks artifacts can Glean index for enterprise search?

Glean can index Databricks artifacts including dashboards, tables, and Genie spaces, making them discoverable through natural-language search. This allows employees to find existing analytics assets rather than recreating them, reducing rework and speeding onboarding for new team members.

Why does more agentic capability demand more accurate context according to this video?

As AI agents take more autonomous actions based on their answers, the cost of a wrong answer increases significantly. The speaker argues that context efficiency—feeding the AI only the precise, relevant information from both structured and unstructured sources—becomes critical to avoiding errors that cascade through automated downstream workflows.

Full transcript

[00:08] Hello everybody. Uh we'll get started in just a minute here. Uh about on time. Thank you for taking a few moments before your lunch. I hope on uh on a Thursday at this event if uh you've been here all week. I can imagine you're a
[00:20] little weary at this point. So we'll try and make this as lively as possible, give you the information you need, and send you off to lunch uh quickly and effectively. Um, we have a couple of disclaimers which hopefully you've seen
[00:32] in other sessions and have read. We'll give you a few seconds on that one. Um, and they've also very much encouraged me to ask you to fill out surveys for the sessions that you're in. So, please take some time to do so.
[00:44] Thank you. And now we're going to talk a little bit about from search to SQL. Uh what we're going to talk about today is uh how an AI native system like Glean can interact with with data bricks and what data bricks has come out with to
[00:57] unify the data across your entire organization and provide some insights which are otherwise not available. Oh wow, it's it's close captioning as I talk. That's excellent.
[01:11] We live in a we live in a beautiful future, don't we? All right. So, today we're going to talk a little bit about data access is hard. Um, I've been working in data off and on for uh probably 30 35 years now, which
[01:23] is longer than I care to admit. Um, it's always been hard. We'll talk a little bit about that that reality. We'll talk about what Glean does, particularly around context. We'll talk a little bit about data bricks plus Glean and how that works in the context of of uh
[01:37] enterprise corpus or enterprise data set. And then we'll show you show you some real world use cases. This is me. I'm very attractive. But um what you should probably know is that I have been
[01:49] working in data for for quite a long time. I was a product manager on BigQuery. I was one of the early people at Looker. Um I've worked in streaming data and Kafka and Debzium and Flink and
[02:02] um a lot of different technologies over the years around data. So I'm really excited to be here at Data Bricks um to talk with you here. If you've got questions about any of those things um hit me up afterwards. So be happy to chat with you about them. Historically,
[02:15] data access has been hard. If you've been working in data, this comes as no surprise. We know for a long time that we've been struggling with data engineering. We've been struggling with this this idea of uh getting all the
[02:28] data into one place, which is not a trivial task. Big data is big. Data has gravity. These are things that we've been talking about for decades. But defragmenting data can be very hard
[02:41] and very complex and where you get to that data is a challenge of the surface that people interact with data in. Expertise is a blocker. We cannot assume everyone speaks SQL as much as SQL is the canonical way to access data
[02:54] particularly in a relational database. SQL is not a thing that we can expect everyone in organization to have particularly our executive team. So how do we get them access to the data they need? And the last thing we know or at least we believe is that data is very
[03:08] valuable. That there are nuggets hidden in that data which can be extremely useful. These are truths that we believe which is why we are here at this conference which is why we're talking about data today. People don't know
[03:20] where the data is, who owns it, how to find it. So we have this problem of the surface, we have the problem of skills and we have the problem of discovery. And these are not new. These are decades old. these problems.
[03:34] But the problems persist and that is people make decisions slowly and they make bad decisions because they don't have the data they need. And the reality here is we're all going to make the best decision we have with the data we have
[03:46] in the time frame we're allotted. And if we can't get good data in a convenient and consumable way within that time frame, we skip ahead and make a decision that may or may not be the best one. And that's just the way human beings work.
[03:58] fighting that is uh something that we've tried for a long time to get everybody super data literate and have them spend more time in it. But uh really I think it's incumbent upon us to put the data in front of people so they can make the right decisions right at the time they
[04:11] need to. So what is Glean doing about this here? So one of the fundamental problems is that even in a uh lakehouse type environment, even in a combined
[04:24] structured and unstructured data environment, data remains isolated across many places. There's a thousand SAS apps. There's data in multiple databases, multiple schemas,
[04:37] multiple warehouses, multiple locations, multiple tables. This is the reality of corporate America today. data is in many places and a lot of the SAS applications hold on to that data pretty tightly and
[04:51] it exists inside them and it's a little bit hard to export out into a structured environment. We've tried to do this a lot in this industry where we take the unstructured data we push it through some form of machine learning to try and get something like sentiment analysis
[05:05] which can then be numerically identified so that then we can do some type of machine learning or data analysis against it. It turns out that is a good process but is not sufficient for all of the context that are people need in
[05:17] order to make the decisions that they make. So what Glean is doing is trying to use AI and I think we've succeeded in in combining structured and unstructured data in a way that helps answer the questions that people have in real time
[05:29] right where they need it right where they work and that's a very very powerful capability. So this is how we're solving these three big problems right with with data bricks
[05:41] glean can query consolidated data in any interface and that is with natural language that's very very important that you don't have to speak SQL to do it AI
[05:53] should be able to use that it's its power to build the SQL to get the answers back to a person in a timely manner so they can do it you can talk to it in real time you don't have to
[06:05] You don't have to wait for someone in data engineering to build you something or somebody to cube something for you so you can put PowerBI against it or something along those lines, right? Like this is the ability to directly query and ask questions of data in natural
[06:18] language in real time. And that seamlessness of being able to find stuff to solve the discovery problem to make sure the surface works to make sure you don't need any obscure skills to get to it is that last step. So where do things
[06:30] sit? Um I have like I said I worked at Looker millions of dashboards which is the right dashboard which is the accurate dashboard which one's been updated. Um if you work in Salesforce which dashboard should I look at which
[06:42] is the right report knowing about all those things remains a challenge and with data bricks and glean together we've been able to solve a lot of that discoverability challenge and I'll talk to you about how we do that. So you might ask yourself what is Glean?
[06:55] And as an employee of Glean it is incumbent upon me to tell you what Glean is. And um I will promise I won't spend too long on here. If you want a long detailed uh uh doctoral thesis about it, um meet me afterwards. We can talk about
[07:08] how cool is. But really fundamentally what we are is an AI assistant that communicate with someone in real time uh in natural language right on their desktop to build the types of uh artifacts or to give the types of insights that people need in order to do
[07:21] their jobs more efficiently. It's combined with an a platform to build agents so that you can um automate so that you can make consistent certain workflows across the organization um so that they're repeatable and more
[07:33] efficient over time. And that is all based on something that is very very important and I'm going to harp on this one um today quite a bit and that is the idea of a system of context. AI is like a new hire in your business.
[07:48] They don't know what your business does when you've hired them. If you ask them to do a task and they don't know what your business does, the task will be badly done. It's as easy as that.
[08:00] Context is what AI needs in order to do a good job at being AI. And Glean has built unparalleled system of context based on knowledge across the organization that allows AI to do two
[08:13] things. And this is really important. First, answer your questions well. and second, not have to sift through stuff it doesn't need to sift through. So it you can answer the questions more efficiently. Between 30 and 50% fewer
[08:26] tokens are consumed when you feed AI the right context instead of feeding it random information or data it doesn't need an in order to answer the question. So Glean is a work AI platform. You can
[08:38] see across the top we have an assistant, we have agents, but we also have APIs. So you can put Glean anywhere. You can um use Glean as an NC MCP server so that uh your other AI tools can access glean
[08:52] system of context as necessary or access glean's capabilities or chat capabilities through whatever interface you need. We have a number of customers who who use us with their own front ends and our backends and our backend is very
[09:04] very powerful. We have this agents layer. This agents layer is a layer of intelligence that not only does agentic reasoning but gives you a choice of models automatically routes to the the correct model. So you're not necessarily
[09:17] routing to the expensive model, you're routing to the one that give you the best answer and it allows you to um ask questions in a very intelligent and and meaningful way so that you can get to um
[09:29] um my clock isn't running by the way in case you're wondering. Right. Um um I'll I'll follow the time I think. Excellent. Well, we'll keep going. Thank
[09:42] you everybody. Um so what Glean can do is is not only be intelligent about how we use intelligence, which is to say be more efficient about using the right model, give that model fewer pieces of
[09:54] information to get to the right answer, but also u evolve and adapt as its reasoning so that you can get a better answer that's more accurate out the other side. And it runs on top of this system of context. The system of context
[10:06] I'll go into in a little bit more detail in a moment. But it consists of knowledge of your company and knowledge of the individuals as well. Um individual personal graphs, private personal graphs for each person, all
[10:18] based on the ability to index and search across the entire corpus of data in your company. So again, if you were a new hireer in a business, the most important thing to be able to execute a task on behalf of your new company would be to
[10:31] know where that information is, know how people work, and then be able to use that information to work in that manner to come to the right answer. That is fundamentally what Glean is doing in a combination of context and reasoning
[10:43] with our agent layer. Last but not least, we have a protection layer that's incredibly important. Um, Glean is uh permissions aware. So when you access data, if I were to ask a question of glean, I will only be able
[10:56] to receive answers which are based on information for which I have permissions in those host systems. So if you think across a hundred SAS tools in your company, you have permissions in each one. If you ask question of glean, it
[11:08] has indexed your identity in addition to the data. It combines those two so that you only receive information for which you already have permission. So essentially that is how Glean's platform works. Again, I don't want to
[11:20] beat you to death here on that. So, um, we can talk afterwards. You can come down to our stand afterwards. We could show you how it works in action. The point here for this session and why we're at data bricks is to talk about
[11:34] how this works because questions evolve very, very quickly. Datab bricks and a data layer, a semantic layer, an analytics layer is very, very good at telling you what is
[11:46] going on. And the closer you get to real time, the more powerful it can be. That is to say, I can know in real time what sales are like on the east coast. I can know in real time where my supply chain is at any given time based on shipment
[11:58] data. I can know in real time how well my sales team are doing based on close data. But what is happening is only the first question to answer. The second question to answer is why. And why is hidden
[12:12] outside of most structured data systems, even in a data lake, the why is hidden elsewhere. It's hidden in chat conversations, gong calls, call
[12:24] recordings and transcripts, um comments on the sides of documents like uh like product development documents or um or or comments or notes inside Salesforce
[12:36] for uh from one rep to another. This unstructured data tells you not just what a metric has done, not what the changes are in the dimensions and measures inside your data set, but why, who's done it. Let's say you have an
[12:49] outage because you've developed some product and it's an outage. Knowing who's chatting about changing some of the code on the back end the day before that outage occurred will tell you why
[13:01] it occurred, not just how big that outage is or how many people are impacted. So while data bricks and the the data which you can store inside your data repositories particularly structured data can tell you what happened the why is where glean comes
[13:15] in. And the next the third most important thing that occurs after you have what and why is a now what because the next thing you have to do is take that insight that you've created that insight
[13:28] which you have uncovered and why it happened and figure out what to do next. In my example where I talk about well we see three people talking about checking in some new code the day before the outage and one of them has checked in
[13:40] the new code the what next could be can I take that out of the code and recompile it so that we don't have that change so the outage goes away that could be a step you could take and it's a steps that you can automate if you
[13:52] know what happened and why it occurred in an accurate way and so this is where glean and data bricks work very very well together data bricks can give you the what we can give you the why and our agenda authentic capabilities can do some of that. Now what or at least
[14:05] suggest them to a user or an employee in your organization on what to do about what you have uncovered and we do it with three really important capabilities. One is this knowledge graph. So we understand what is going on
[14:18] on inside the business. When we are indexing um our data we're indexing who's commenting on what documents, who's accessing what documents, what is in the documents, who the people are commenting on and where they sit in your
[14:30] organization. These types of really rich pieces of data about the data, data about the people who are interacting give you a good insight. So you probably all have projects at your company and
[14:43] lists of people who are in those projects. What I can tell you is the people who are really engaged in those products in those projects are the ones who are in the docs commenting on them, reading them, writing new docs and
[14:55] adding them to that project. And that is an actually a much more accurate way of getting to who is in a project at any given time rather than a static list. And so Gleam can do some of that around our knowledge graph. We also do that
[15:08] permissions aware retrieval. So we retrieve data out of the corpus that's specific to the question you ask but also to the permissions of the person who's asking. So that security that compliance the ability to be permissions aware is super key. And last but not
[15:22] least is this idea that we are not just searching one app. We don't just look in teams or slack. We don't just look in Jira or a sauna. We look across all of the applications which are connected
[15:35] into glean and they can be all the applications and we find the relevant data the most relevant data and provide that either to the user or to AI to take uh action
[15:47] building an insight or maybe taking action on beh on behalf of the user and relevance is what's key here in a world where we have terabytes of data the most relevant data is not always obvious and that's a really important differentiator
[15:59] glean was founded on the idea of finding ing searching and retrieving data and putting it in front of users. Now we search and retrieve data and can put it in front of AI as well. And that's how we deliver answers which are preferred
[16:13] let's say roughly two and a half times to answers um that don't use Glean's context layer on the back end. So, we're the leader in enterprise context. And I will say this, whether
[16:26] you decide to go with glean in your product or you're trying to go somebody else or whatever you're building out a system, think about context as one of these top core components that you really need. AI is just like a person.
[16:38] It cannot do good work unless it knows the information it needs to know to do good work. And it needs to have relevant information put in front of it in order to be successful. And the more agentic you get, the more
[16:51] AI you get, the more context is required. If you're just trying to find something, then you don't need a lot of context. You can ask for a file name and we can do a boolean search and we can return a file name to you. Glean can do
[17:04] that. Many things can. Um, it's not so easy inside a corporation as you'd think because the permissions problem, but you can do that, but it doesn't require a lot of context. But if you get into a complex search, and this is me all the
[17:17] time, Kelly sent me something in Slack. I don't remember really what it was about, and it was like three days ago or maybe four days ago, but it seemed really important. Can you find it for me?
[17:30] Glean can do that for you, too. And that requires a lot of context. Needs to know who I am. It needs to identify which Kelly I'm talking about. And I work with like four or five Kelly's. One of them's back there. Hi, Kelly.
[17:43] >>
[17:55] important when you're answering a vague human question like what was that slack I missed? And when you get to a complex agentic workflow where it's something along the lines of a ticket came in, I need to respond and escalate it, but I
[18:07] don't know who I should escalate it to. Can you build me the ticket, put the right information in it, and escalate it to the right person? Help. That requires a ton of context. You're updating a system of record. There's a lot of of
[18:20] nuance in what a ticket looks like, what a best practice for writing a ticket looks like, who you're going to escalate to, knowledge of the organization, knowledge of the product, knowledge of the ticketing workflow. That requires a ton of context. And so context is the
[18:33] king of what we need when we're building agentic systems. And the more agentic it is, the more context is required. Glean understands context at a very deep and rich level. We build, I mentioned
[18:45] earlier, two types of graph. There's an enterprise graph that's based on what's going on inside your company. And then there's a unique one for every person, a private personal graph. I have one. I
[18:57] can show you the embarrassing things it knows about me afterwards. But yeah, it does things like it understands who I work for, who I work with, what things I've opened and used recently, my writing style. It can emulate my writing
[19:09] style when I ask it to send a Slack, which will be a lot less formal than I ask it to write a white paper, which might be a lot more formal. And so Glean understands how I work, but it also understands how my organization works.
[19:22] It knows for example if I ask a question like tell me about the latest PR that if I'm in engineering it will assume that's a pull request. If I'm in marketing it might know that that's a press release
[19:34] because PR is a non-specific term. It can change who you are and where you sit in the organization. It might also change if you are an engineer who's involved with marketing day by day what I mean when I say PR. And so that
[19:47] context is really really key and it needs to be based on metadata and permissions which are indexed across your organization too. I mentioned this earlier but um if you're indexing an HR system some of that information needs to
[20:00] stay private. So how do you make that happen? Glean reflects the permissions of the systems to which it is connected. If I do not have permission in that system, that data is not retrieved or fed to the AI. It is not a part of my
[20:13] answer. Only data I'm allowed to get is included in the answer that I receive. And only data that I'm allowed to access or change is data that I'm allowed to access and change through our agentic systems. That is a non-trivial problem.
[20:27] If you have a 100 SAS apps, I guarantee you have a hundred different permissioning systems which are difficult to reconcile with one another. It's a very complex challenge. And we go and we look at data, structured and unstructured data. We look in your
[20:40] applications. We look at the tools you might have available to you. uh maybe you're connected to a bunch of systems via MCP and we're going to be doing some type of federated search across tools that are connected agentically into glean we can do that as well and then we
[20:52] look at the people their activity your identity system to know who everyone is that all comes together in these into this overall knowledge graph which is how we go about context and again
[21:04] context is really important because context helps you deliver the right data to AI at the right time to get the right answers All right, I can go on and on about context and I probably will not, but
[21:18] this you an idea of some of the entities and the components that we put together into an enterprise graph or a personal graph. And I think it's really interesting to look at this, right? Like somebody updates an opportunity field and often they generate a call
[21:31] transcript afterwards because they're going to get on a call. That's a thing that's really interesting to know because we might use that later for asking questions about I saw an updated opportunity. Can you summarize what the call was that came afterwards? If you
[21:43] write a first call follow-up email, you want to make sure that you're adhering to the correct style, that it's the right type of an email for that time of a conversation with a prospect. If you're a salesperson, these types of
[21:56] contextual um decision making we do as human beings all the time because we have at retrieval in our brains the ability to get to some of the somewhat obscure
[22:09] information that makes it look human, feel human, and be effective. And that's what context brings to you here from a from a personal graph in an enterprise graph. I mentioned this as well, right? we can see who's often editing what
[22:21] components, which apps are connected to which apps, who's referencing what. And that gives you an idea of how entity relationships work and how projects are executing. And it really gives you a
[22:33] good idea again of that not just what happened, but that why it's occurring, how it is progressing, and what to expect coming later, how to take action if you want to do that agentically.
[22:48] Okay, let's get into the meat a little bit here. Glean and data bricks are integrated in four ways. So I'm going to start all the way to your left and this is datab bricks genie
[23:00] in glean assistance. This is one of our integration paths. So glean can call databicks gening via mcp and we can get the right information out of datab bricks as necessary. So it's a
[23:15] nice native integration where you can easily from glean assistance or elsewhere in glean get to data bricks information. It'll route you to the right uh genie space. It'll provide you
[23:27] with access to the insights that have been generated inside data bicks. All the way on the other side you have data bricks genie and genie code plus glean. So that's glean available inside datab
[23:40] bricks. So it's the reverse, right? And this is one of the things we're seeing now in AI. And this is why it's really important that we have an open and interoperable world. So I'm glad we have that with data bicks. But not only can you get data bricks inside Glean, you
[23:52] can get Glean's context layer inside data bricks. And that allows you to to not only extract warehouse type or lakehouse type analytic answers into
[24:05] into Genie, but also the context surrounding it. And so again, if you are doing code in Genie code, you might want to know what the conversations about that code are that are going on in Microsoft Teams, that's the sort of
[24:17] thing that Glean can supply into Genie. And so that's a really powerful cross integration. Now, there's a couple of things in between the two that are also important, and I'll talk about this, too. You can actually have Glean call
[24:31] SQL and just call datab bricks SQL and and run a query inside data bricks from glean. Pretty easy. And that's just literally dropping SQL right into the the SQL API data bricks and pulling out
[24:44] uh structured data. Pull a table do a join and pull a table. Very easy. Um and we also have a connector which is just to say that we can also index any uh
[24:56] artifactor object that exists inside data bricks so that when someone is doing a search or trying to combine information about uh completed analyses or completed objects inside data bricks
[25:09] they can find it through glean as well so that data bricks objects become part of the corpus of data that's indexed inside glean. So this is uh these four ways are really important and I'm going
[25:21] to talk through them at some detail here. Um I actually have a technical colleague Jess in the back who can help us if we get really super deep and um but I think that's important to think about whenever you're building one of these systems is I don't believe in AI.
[25:37] We have yet decided what the front door that everyone will be using will be. Some people are going to want to work inside data bricks but need glean context. Some people are going to want to work in the glean AI assistant and need data bricks context and we aim to
[25:51] make sure that both of those people get access to the context that they need so that you can answer the questions you need or take the agentic action that you need to do. So let's talk about data bricks genie in glean assistance. So
[26:03] I've got a workflow here for you. But if if the user is in Glean assist and they ask a question, that question requires Genie information. We have indexed metadata from Genie. We can um find the
[26:15] best fit Genie space, translate into the governed SQL that's necessary, return results into Glean, combine it with the Glean context, and deliver the correct answer. And so that means you can use natural language in real time. We route
[26:27] to the right models to make sure we get the right insights. And of course, we're secure as well. But the idea here is context plus framing around a data bricks warehouse type answer is is
[26:39] really uh simple and easily uh easily addressed with uh data bricks behind the glean UIUX and our agentic capabilities.
[26:51] We can also do that direct SQL. I spoke a little bit about this just a few minutes ago. There's two ways this can occur. One of the ways is where you actually just type SQL right into Glean Assistant and tell it to go send that
[27:03] SQL to data bricks. It will do so. Um, it's actually pretty interesting. If it doesn't speak the right SQL dialect in these contexts, it will know where you're trying to put it and will tune your SQL dialect for you before it sends
[27:15] it over, which is nice if you're a SQL head like me and you don't know what the dialect is you're trying to talk with. Um, but yeah, you can type SQL right into Glean and send it off and get an answer back. You can also from glean agents, the agents can call predefined
[27:28] SQL capabilities, run those, get the answers back and add that into their agentic work. And so you can just simply send SQL. Um, for those of us who speak SQL, this might be the easiest route and the most predictable route. If you don't
[27:42] speak SQL, that might might be better just use the natural language way to try and get to the answers that you're looking for. every dashboard, every genie space, these objects inside data bricks can also be indexed and be made discoverable
[27:54] inside glean. So if you're looking at a metric or a topic or owner or description, if you've got a data dictionary that you're using somewhere that we need to be able to index to understand what metrics are, how they're described and defined, um you can index
[28:08] that stuff inside glean and that gets you to this really rich idea of what has already been done so you avoid rework and that is a constant challenge. um we've seen in the data space for a long time which is I don't know if there's a
[28:20] dashboard can you make me a dashboard what is the dashboard when is the latest dashboard updated etc etc and then here's the first part of our fourth integration right which is um glean behind data bricks so if you are
[28:34] trying in databicks genie to answer a question that is not in the data lake the lakehouse u from data bricks how do you find that additional context um with a with set up properly, you can actually
[28:48] ask that question in Genie. Genie can call glean MCP and pull in that context which is not natively inside data bricks. And so that gives you again like a genie native way if Genie is your
[29:00] interface of choice to access the rich context that Gleam can provide. And this is true also in Genie code. Um, oftentimes if you're using Genie code, you're going to want the context around
[29:12] what's going on in Jira, what's going on in in in other tools that you might use of the Atlassian suite or your coding tool set or GitHub or something along those lines. Um, Glean can index all of
[29:24] that and provide that to people as they code right in Genie code. So that context can be super rich and super useful um to folks who are interested in um in coding using data bricks uh genie
[29:38] code. So what do we do? Um we provide a unified analytic a unified intelligence layer across the entire
[29:50] org. So this is unifying what data teams do and unstructured data of how people interact and putting it together in one place with that AI interface that makes it easy to access and use. That
[30:02] self-service component is very very important. It's not just bringing it together but making it accessible. It is fundamentally the holy grail of our data teams is making the data useful. Also
[30:14] not having to answer every single question from everybody who always wants a dashboard all the time, right? Having people be able to do that self-service saves us all a ton of thrash and lets us focus in on the types of data engineering or data work we need in order to build foundational metrics and
[30:28] things like that in the org. And then the discoverability so we don't do rework. We know it's already been built and we know that it's uh we know that it's already there so we can reuse it or update it if it's not sufficient. These
[30:40] are all super important. There we go. So how does this pan out in the real world? Well, we have a number of folks who've deployed glean and data
[30:52] bricks together. One of which you would call a top well let's say three auto automobile manufacturer in the world. Um they are their design and research engineering teams are using glean to
[31:04] query data bricks genie in order to get the data back in order to do the design and engineering work that they need to do. The savings that they're seeing there are absolutely remarkable. um they they were really doing siloed AI
[31:17] experiments across a bunch of different things. They've unified it all into a single context layer now and that gives them the sort of scale and corporatewide context that they need to get the answers that they're looking for. So that's SharePoint and Outlook and Slack
[31:30] and all the rest of the components of what they're doing in one place easily accessible. So yeah, they're planning, this particular automaker is planning on rolling out just under 150,000 users. Um
[31:44] uh and uh right now there's the data team is uh is uh freeing up, you know, in excess of an hour a day, maybe more than that for everybody in the data team because they're able to do this as well.
[31:56] Um so they're they're thinking in the first uh in the first year they're going to save around 300,000 employee hours as a result, which is a significant win for them. Um, we've got a real estate
[32:09] company, a real estate SAS company that's using us with data bricks. Um, so this is like combining documentation and employee insights and people analytics out of data bricks and being able to put it all in one place, saving every
[32:22] employee, you know, an hour or two a week, but that adds up across, sorry, yikes, across 7,000 employees. And um,
[32:34] but they're going that next step. They're actually building agents. They've built more than 3,000 agents today. And those agents are taking the work and automating it, going beyond sort of the delivery of insights, which would be our traditional thing, beyond
[32:46] the what, beyond the why, but to the what next. And so that is a really key and powerful thing that they're able to do. And then last but not least, we have uh an EV company which uh you I think
[32:59] you might have seen down on the show earlier. They're doing the same thing. They're they're looking at uh productivity gains in the couple hundred thousand uh uh hours a year type of productivity gains which is nothing to
[33:11] shake a stick at and that is unifying together all the things from their expanding workforce into one place um onboarding teams now better structured insights across the whole organization.
[33:25] All right. So, with that, um, I would encourage you to come talk with us and see this in action downstairs, uh, in, um, in the north hall where we have a booth, but I'd be happy to answer any
[33:37] questions you have or talk through, uh, any challenges you might have around context, data engineering, and AI going forward. So, uh, we have a couple of mics, and if I can't answer them, I've got some colleagues in the back who can
[33:49] help as well. All right. Thank you very much. Thank you.

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.