Lakeflow Jobs: Modern Orchestration for the Data Platform

Summary

This video introduces Lakeflow Jobs as a full orchestration platform—not just a scheduler—that runs at 1.7 billion tasks per month at a 99.5% SLA and charges nothing for the orchestration layer itself, only for the compute it runs.
Lakeflow Jobs supports data-aware orchestration with event-based triggers including file arrival, table updates, and data condition events, enabling workflows that respond to data changes rather than relying on fragile time-based cron schedules.
Genie Zero Ops surfaces root-cause analysis directly inside job monitoring, and Genie Code can generate new aggregation tasks on demand, accelerating how teams build, troubleshoot, and improve production orchestration workflows.

Lakeflow Jobs: Modern Orchestration for the Data Platform

Modern data platforms need more than a scheduler. They need orchestration that understands data, integrates with the platform, and scales with both engineering and AI workloads. Join this video to see how Lakeflow Jobs helps teams orchestrate pipelines, notebooks, dbt projects, SQL, and external components, natively on the Databricks Platform. We will show how Lakeflow Jobs goes beyond basic scheduling with data-aware orchestration that understands the assets and systems it is coordinating. You will see how teams can trigger workflows based on data changes, build modular workflows in Python, orchestrate external systems alongside Databricks workloads, and use Declarative Automation Bundles (DABs) for a more reliable CI/CD path. We will also cover how Serverless Jobs and observability features such as lineage and system tables help teams reduce operational overhead and better understand what is happening across their workflows. Finally, we will preview how agentic authoring and diagnostics can speed up building, troubleshooting, and improving production jobs, and how you can embed agents directly into your workflows. Whether you are modernizing from legacy schedulers or building new workloads on the lakehouse, this video will share practical patterns, product direction, and live demos for modern orchestration on Databricks.
Talk By: Gabor Ratky, Engineering Manager, Databricks ; Saad Ansari, Product Manager, Databricks ;

Chapters

FAQs

What is Lakeflow Jobs and how does it differ from a traditional scheduler?

Lakeflow Jobs is a full orchestration platform that goes beyond cron-based scheduling to support data-aware triggers, multi-task workflows with conditional logic, and observability features like lineage and system tables. It handles notebooks, pipelines, dashboards, SQL files, and external components natively on the Databricks Data and AI platform.

What are data-aware triggers in Lakeflow Jobs?

Data-aware triggers allow workflows to start automatically in response to data events such as a file arriving in a location, a table being updated, or a data condition being met. This eliminates the brittle timing dependencies common in traditional scheduled pipelines, where a downstream job fails because an upstream job ran longer than expected.

How does Genie Zero Ops integrate with Lakeflow Jobs monitoring?

Genie Zero Ops surfaces root-cause analysis directly inside the job monitoring interface, so engineers can understand why a job failed without switching to a separate observability tool. This agentic observability capability is part of Databricks' broader effort to automate post-production operations described in this video.

What is the pricing model for Lakeflow Jobs?

Lakeflow Jobs does not charge for the orchestration layer itself—teams only pay for the underlying compute that their tasks run on. This makes it cost-efficient compared to third-party orchestrators that add separate per-task or per-workflow fees on top of compute costs.

Full transcript

[00:09] So, uh, welcome everybody. If you were here for the chain smokers, I hate to disappoint you. Uh, it's just the Lake Flow boys. Um, uh, thank you for being here. It's kind of fancy this place. Uh, let's see if, uh, everything is working.
[00:21] All right. So, uh, these, uh, very dense paragraphs. You'll see this, uh, throughout the day, uh, or throughout the week. Basically, we're very excited about where we are going to show today and we'll do our best to make sure that
[00:34] it's a reality by the time you leave. Uh but uh you know, we're among friends here. Um I'm Sad. I'm a product manager uh at uh data bricks. I work on LakeFlow and I have uh basically worked as a
[00:48] farmer or partially. I worked on Fitbit for cows and I used to be a developer before and I am happy to be in the Bay Area. I studied here as well. And uh here's Gabbor. >> Hey folks, I'm Gabbor. I'm an
[01:00] engineering manager on the lakeflow team specifically focused on orchestration and been with data bricks for four and a half years which is quite a lot in in data bricks years. Uh before that as
[01:12] essentially for the last 15 years been building data pipelines and data platforms. So I got the job to make things better. >> All right. So just to make sure we are all in the same place where we're
[01:24] supposed to be, we're going to talk about LakeFlow jobs today. And u so jobs is not just a scheduler, it's basically a way to build, trigger and run uh
[01:36] reliably your data pipelines. You can construct your flow. Uh you define your unit of work. Let's say these are notebooks, pipelines, uh dashboards and to stitch these together with control flow primitives like loops and
[01:49] conditionals and you get this uh sophistication uh of uh this control flow without that complexity. And then you can also trigger all of this with uh not just cron but using event-based
[02:01] triggers uh file arrivals, table updates, data conditions and lastly you delegate the infrastructure to us. So we carry the pager and uh you get all of this uh for free uh including
[02:13] observability. Um, so just to kind of level set, you've probably seen some of these numbers before. Uh, it's a production grade uh, orchestrator that we run. Uh, it's 1.7
[02:25] billion a week uh, or sorry, uh, a month a week very soon uh, at that um, and um, 395 SLA. As I mentioned, we do carry the pager for you on your
[02:38] behalf to run the orchestrator. And it comes at a nice uh price point of uh $0, which means uh we don't charge you for the orchestration. You pay for the compute, but the orchestration comes for
[02:50] free. Um and so, um let's see, can I move this along? There we go. Uh so, job's grown a lot over the last few years, and I do mean a lot. You have a ton of these tasks that
[03:05] you can run. Um you have uh various these orchestration primitives that you can use and uh kick off these jobs using a dozen different mechanisms. You get observability, you get compute and you
[03:17] can govern it all with dabs and with uh you can bring it in with Unity catalog. But the problem actually sorry I should have clicked off the animation uh while I uh spoke about all of these. Um the
[03:31] problem that sort of is inherent with this is it's a lot and it can get very challenging to discover to debug to author using this tool and that's really uh one of the first sort of challenges
[03:44] that we want to try to address and the way it um sorry this thing is working. There we go. Um so yeah so the first one you know like often times um uh customers say that you know I I I want
[03:57] to use this feature and we're like hey it's already there you need to just you know click here and click there and click there and there's a lot of click ops often times that's needed which makes it complex to to find what you need to troubleshoot to uh basically get
[04:10] your job done and we recognize that. Um we also ran uh some analysis on our jobs and we noticed that a lot of these are not optimized or not running at the right time. So hope uh we were talking
[04:23] to a customer yesterday and they were talking about like we hope the data is arriving at 8 a.m. and we hope so hope is not really a great strategy for data pipelines. You want the pipelines to run when the data is available rather than
[04:36] on a clock. And then finally, um this comes up a lot where um customers often need to orchestrate stuff that's outside of data bricks and either you're rolling your own um you
[04:49] know uh connectors or integrations or you're using external orchestrators like airflow ADF and um that's uh that's a big problem. Uh you know we we've recognized that and we want to start
[05:01] working on this. So um just quickly one two three for the the complexity piece that we talked about where there's a lot of this out there we have Genieode and we'll go through Gabbor is going to take
[05:13] us through some of the uh improvements and some of the innovation that we're driving over there. You've probably also seen some of it in some of the keynotes uh from yesterday. Um we're also bringing uh data awareness and context awareness into theuler sorry into the
[05:27] orchestrator. It is not just theuler. Um, and so we can give you best advice on when to run your workloads. Um, and then finally, uh, we're going to help you orchestrate
[05:40] everything, not just within data bricks. This should be the only orchestrator that you should ever need. And we'll go through the framework that we're going to be, uh, announcing today, which, uh, lets you orchestrate, uh, primitives outside of, uh, just the workspace. And
[05:54] so so jobs is you know oftentimes people think of like the jobs canvas as the entry point and it is uh not just a web page. Uh there's various uh personas that interact with jobs. Um obviously
[06:07] you can get the schedule button if you're familiar with dashboards or pipelines or notebooks. There's a little icon there up on the right that you can use. Uh we're actually launching uh scheduling from Excel and from workbooks. So you can actually
[06:20] orchestrate data bricks from outside. Um then obviously there's this uh jobs canvas which is very helpful especially when you're building more complex uh workflows. And then finally for the engineers in the house uh we have uh
[06:33] dabs. I'm a big fan. I used to work on dabs uh before. Um and uh but what we've noticed is that these different personas uh have also proliferated. Oh clicked too fast. Uh the big reveal is genie
[06:46] code. So uh basically because it's the underlying uh tech is everything is a job and you can source control it, you can put parameters in um we're exposing all of those interactions via Genie
[06:58] code. So you can give your outcome and we figure out the rest. So if you want to build something, if you want to debug or troubleshoot something, if you want to productionize it, you want to source control, Genie has all those skills to
[07:10] be able to do that. And then I think with that we'll just uh jump straight in and stop talking and start showing. >> So we brought you a live demo and s talked about all the different features that uh Lakeflow Jobs has that we've
[07:23] built over the years. But I want to start in a simple and familiar place. I'm in the workspace. Uh I'm in a workspace folder and I have various assets that I've built interactively. a
[07:35] notebook, uh, a SQL file I wrote, and then dashboards and it maybe alerts that build on top of those. Um, and if you're familiar with LakeFlow and you're familiar with LakeFlow jobs, uh, and
[07:47] data bricks as a platform, you'll see that we have a schedule button very often. And well, a schedule button just does that. I'm ready with my transform transactions notebook reading some raw transactions data. I know this data or
[08:01] as s said I hope this data shows up uh every day by 2 am in the morning. So let me just schedule that to run 2 a.m. in the morning.
[08:13] And with that, this notebook now has a schedule, the 395 reliability. We guarantee that this will run at 2 a.m. in the morning. But what really happens when I go back to that workspace folder
[08:25] is we now have a job. And that job is very simple. But you may notice that it's just another thing on that list. And why is that great? This is something new that we finally uh made simple is
[08:37] that when I share this directory with SAD and I collaborate with him, uh we now have a single unified permission model across the assets that you work on interactively and the job
[08:51] that orchestrates them. We call these assets in the workspace. They apply to both jobs and pipelines. and we're launching this uh to a workspace uh near you. So with that, I now have a job that
[09:03] runs this notebook. Let's see how this job looks like uh when I click on it. Well, it's a pretty simple one. It has a single task, but you can have actually multiple tasks and some of the
[09:16] capabilities um really shine in LakeFlow jobs when you orchestrate multiple tasks and you introduce dependencies, conditional execution. As a matter of fact, we have a lot of tasks you can
[09:28] pick from, right? All the different assets that are available
[09:42] constructs that you might be familiar with if health conditions the for each task that allows you dynamic u metadata driven programming um and more advanced uh construct. You
[09:54] can refresh dashboards um and you can even run AI and ML uh workloads from within Lakeflow jobs. But that might just be a lot. What if I would have a
[10:06] simple way to author and extend my jobs? Well, I already have the transform transaction notebook scheduled. So, let me ask Genie code to extend my job by orchestrating an additional SQL file
[10:20] called aggregate daily sales. Now, let me ask Genieode to do that. And while Genieode is working on this,
[10:32] code here already has all the context about this job, what um what uh tables it's reading from, what tables it's producing. Um and it understands that
[10:46] that SQL file I need to orchestrate with a SQL task. In order to execute a SQL task, I need a SQL warehouse. Because this job does not use a SQL warehouse yet, it goes out and lists the available
[10:59] SQL warehouses in this workspace. and it will uh infer what might be the best SQL warehouse um to run this workspace run this um uh SQL file in. Um I think if
[11:14] you've listened to Ali's keynote, he talked quite a bit about context. Uh what makes Genie code special is that whenever you pull it out, whether it's in LakeFlow jobs or other parts of the platform, it has all the context about
[11:26] your work. I can talk about jobs specifically that if you have a job that ran before it will have access to lineage. Um if that job uh processes data it can actually look at the shape
[11:39] of the data and with that uh Genie code is now thinking that okay I need to find a warehouse and we should be good to go
[11:51] very soon. Yes. All right. This is perfect. They find a warehouse called LakeFlow jobs demo. Surprise, it picked that one. Um,
[12:04] now it is making changes to my jobs. And while this example is simple, this could be a complex change, a change that's unsafe. It's changing the code that's running. Uh, so it shows a diff u uh of
[12:18] the changes. when I approve this change now I have the job extended with that additional uh task and while the example is simple you can imagine that for complex changes this works uh very
[12:32] similar I'm ready to run my pipeline interactively um let me close Genie for a second and as I run my pipeline and if you're not familiar this is our um uh view uh
[12:47] some sort of entry view for a single run. Um I can observe here what's going on. See how the notebook uh is processing and my second task which we've added with GI code will run right
[13:01] after. So let's just give it a second uh for this to complete and let's see if my pipeline works as expected. We'll just to make things a bit faster,
[13:13] we'll refresh this page. All right, I guess there's a lot of Hey, of course the demo gods are not with me. I have a failed task. What do I do?
[13:27] Let's see what happened. All right, my SQL query has an error. And I picked a simple one so that you can uh we can all follow together. Um, but I don't
[13:39] actually have to go deep and start analyzing the problem myself. I can ask Genie code to do just that for me. Um, so let me let's ask Genie code uh to fix the job and once again we will look at
[13:54] the error. We will look at what happened with the whole job possibly look at the data that it processed root causes the problem and proposes a fix and it now applied that fix. Um you can
[14:09] uh you can also review the diff uh of what changed. Uh but in this case I'm going to just simply accept all and use a feature called repair to repair this job run. It only repairs the task that
[14:23] have failed which should now be um which should now be uh working. And let's see if I was right. This is my second attempt at aggregating the daily sales.
[14:46] And well, it's doing something. So that's definitely a good sign. All right, let me go back to that refresh. My
[14:58] connection is acting up a little bit. And now we have a succeeded job. So, we've built quickly a pipeline with the help of Genieode and and troubleshooted. Um, so when I go back to the job,
[15:10] this job is still scheduled to run at 2 a.m. But we know exactly what it does, right? It takes the raw transactions and it processes and aggregates uh to uh daily sales and we might put a dashboard
[15:22] on top of it. But I think we can do better than this. S talked about data and context awareness. So, let me just get rid of this 2 am. And now I have Genie code look at my
[15:34] job, infer what it does and suggest what I really should be triggering on. Well, this job reads new rows from transactions row. So it really should run when that table updates, right? So
[15:46] let me just accept what Genie code suggests. And now this job will only run when that table updates. Um, but maybe data trickles into that table and I
[15:59] don't want to always spin up a new job run, spin up a cluster to to to do just that. So I can do something more advanced than that. Maybe I can ask Genie
[16:11] to trigger this job only when there are at least 1,000 unprocessed rows in the source table.
[16:23] Now, I did not mention any features that LakeFlow Jobs has. I really talked about my intent. What am I looking for? I want this job to run when there are at least a thousand unprocessed rows that it can
[16:36] process. So, it's worth my time and my DBUS. So, it looks at that it triggers on table updates and it knows that there's a feature in jobs that we're launching soon called SQL conditions.
[16:50] It looks at the queries and it looks at the actual data to understand what it means to be unprocessed. In this case, the table has a processed uh column and proposes a query to me. Let me run this
[17:04] query every time and only if this query um succeeds, only this query returns a row uh will I run your job. So I'm going to allow Genie code to create this query
[17:17] and I will let it configure my job to run only when there are at least a thousand unprocessed rows. And this is kind of a madeup example, right? But you can imagine that as we build features
[17:30] and as we teach Genie code through skills, we can really apply best practices here. Genie code can really incorporate everything we've learned about data pipelines and help you just build u uh following best practices. So
[17:44] here's my suggested change. Um and when I look at this uh job, I think I can close this now. It now triggers when the table updates
[17:56] and will use an additional SQL condition to determine whether you should really run that uh job. And I think that concludes my demo. Everything worked. Yay.
[18:08] >> Awesome. Thank you, Gabbor. >> So, I had mentioned earlier that okay, so we're bringing all of this context awareness within uh jobs and you have the ability to orchestrate everything
[18:20] within data bricks. But often times what comes up is how do we uh orchestrate things outside of data bricks. Am I at that point yet or >> oh yes you are. So I'm I'm kind of
[18:32] jumping ahead a little bit. So we talked about solving the discoverability problem, the dis the debugging problem. We solve uh we're solving the context awareness problem and then finally we need to solve the universal
[18:44] orchestration. How do you orchestrate everything within data bricks? So I'm going to give it a shot now with the demo. >> So let's see. Um I'm going to do it by hand. I'm a little old school that way. Um, and then I'm going to add uh task
[18:58] type. And oh, there's a snowflake query task type here. Okay, let's see. Um, um, actually I know I'm it's kind of tongue and cheek, but there's legitimate cases where customers have said, "Hey,
[19:10] we need we have multiple warehouses. We need to be able to orchestrate across them." So, let's I'm just going to do something really silly. Uh, 42, sorry, the meaning of life. I incremented it. Um, and then let me just see what my
[19:23] warehouse name is over here. And okay, and then what I want to do is I want to take the results from this task and then
[19:36] I want to send a Slack message. So let's create that other task here. And then I need uh let's say send results. I've got a Slack connection here. And
[19:48] then let me actually copy the name of the channel.
[20:00] And then I'm going to cheat. Oh, where's my
[20:12] How do I get the sticky notes? Oh, there we go. Okay. So what I did here is I created a snowflake task. I'm running a query on the warehouse. Uh made sure that the warehouse is running and then we're
[20:24] going to interpolate some of the parameters that are going to be passed from the snowflake task onto the uh Slack uh task. So let's take a look at this. It's running uh at the moment. I
[20:36] hope I turn on the warehouse um so we'll run out of time soon. And um let's take a look here. So actually Gabbor is going to uh tell us a little bit more about how this is built. Uh but what is
[20:50] happening is we are building these integration framework uh within jobs. So these obviously we've authored ourselves and there will be a bunch of these that show up in the task pallet. But the idea is that you can now build your own. So
[21:02] if you have any bespoke uh setup that you need uh special configuration, you can actually do that yourself. So uh normally it takes a few minutes. Okay. So it it was running the the libraries. Oh, sorry. It's installing that. Now
[21:14] it's actually running the code which connects to this warehouse. We have um secrets uh that have or actually there's a connection uh that we have. So this it's uh ran this uh query there. And let's just look at the warehouse just to
[21:26] make sure. I'm going to refresh this guy. Showing this. Yeah. Uh come on refresh.
[21:39] Okay. Hopefully it refreshes soon. But let me go back here and show you uh what happened with Slack. And then um it sent this message to Slack. And then here you can see voila. All right. So uh that was
[21:54] uh data bricks uh orchestrating snowflake. So that's kind of fun. Um Gabbor, you want to take us into the sausage making here, please? >> Sure. >> I think I mentioned I heard I joined
[22:06] data bricks over four years ago. I think the one of the main feedbacks I've heard that great that you have an orchestrator but I want to orchestrate more. My company my system is not just data bricks yet. Uh and I'm actually
[22:20] extremely excited to see uh how you can now run snowflake queries and uh u slack send slack messages. And we're building this very flexible very simple and
[22:34] extensible. So let me talk a little bit about that. So we're introducing integrations of which you've seen two. Um and these are simple Python functions or classes.
[22:47] These Python functions you can author yourself or you can deploy existing ones uh whether those are built uh by and maintained by data bricks or by a
[22:59] community. uh we've given a lot of thought how to make this secure. So authentication which is a really important part of this uh is done through Unity catalog connections. This
[23:11] means that secrets are never
[23:23] functions. These are really just functions that execute once. They have no state. You can think of the Slack message you've just seen calling a REST API, the web activity if you're familiar with ADF activities uh and similar
[23:37] examples. Now the second flavor might be familiar to all of you. We're bringing sensors to data bricks within your jobs. Sensors can pull an arbitrary condition may that be a message cue um you know
[23:51] some presence of data through quering SQL or waiting for an external file. The third most complete or most um yeah most complete uh flavor of integrations are
[24:04] operators and operators cover the full life cycle. They let you execute some external workload pull for its completion and we make sure that you
[24:16] can gracefully handle if you decide to cancel that uh job run for any reason. So if you're executing an external query, let's say in BigQuery, and you happen to cancel that run, you can actually make sure that that BigQuery uh
[24:29] gets canceled as well. So there are no zombie uh queries uh eating up uh your cost. With that, let me run through a few properties of these light govern um
[24:43] integrations that we've designed and are building uh for reuse. They run on what we call lightwe lightweight ripple VMs. Really the the the important word to
[24:55] hear out is light. They're really light. They're not a full-blown Spark cluster. We don't think you need a Spark cluster to make a function call. Um so these are the things the smallest unit that we can
[25:07] spin up for you. And as a matter of fact, we can spin this up really fast. So it's already in the order of seconds and we're hoping let's say tens of seconds today. And we know we can cut this back even further. So we want to
[25:20] make this really fast and really cheap. When sensors or operators are waiting, they don't actually use that compute. So we can defer that compute. We can release that compute. So if you're
[25:33] falling infrequently, there's really just no compute and no cost associated with it. Um I've already mentioned cancellation that we uh have first class support and unity catalog connections
[25:46] allow you to configure your connection to external systems in unity catalog which means full governance uh who has access and whatnot. Um but also it means that various authentication protocols
[25:58] like OOTH or bearer tokens uh including support for refresh tokens are completely handled by UC and as you'll see from an example in a second uh you really don't have to uh uh deal with it.
[26:11] Um it's really important that what you build internally or or what you deploy to your workspace is discoverable. Discoverability is important. So these integrations show up automatically in the task ballet. You may have notebooks
[26:24] today that you use to call some external. You can't really make that a native part of lakeflow jobs and the authoring experience. Uh we want to make these discoverable not just in the task
[26:36] bullet, not just discoverable for users but also for genode. Last but not least, I think when we talk about building integrations plural many, and as you've heard from the keynote, we're planning
[26:48] to release uh uh 50 of these integrations uh very soon. Um we want to have a great distribution story and we think that dabs data bricks automation bundles are a great way um to deploy
[27:02] these and you know expect this integration in a GitHub repository uh near you soon. With that um I want to give you just a very quick sneak peek of how uh some of
[27:16] these integrations look like without going into too much technical detail to give you an uh idea. Let me go to a workspace where I have
[27:30] this data data bricks automation bundle open um that has a number of examples and I'll show you how a Slack example might look like. It's not long. It's
[27:42] about 20 lines of code making a request through Unity catalog um to a service uh to a Slack service uh and the and the method. And this line is
[27:56] quite interesting. Unity catalog can inject the necessary tokens to make that call. So I'm actually not doing anything about authentication. I let this thing called the workspace client uh my
[28:10] workspace and unity catalog within that workspace to do all authentication for me and this is just one pattern of doing it. There are other ways but this makes it very simple to make requests through unity catalog. Um similarly
[28:25] a sensor will have a simple poll method that can decide whether it's complete or whether uh it needs uh to continue polling. and an example operator. In
[28:37] this case, we built our run job operator um which is similar to the run job task. We have native Linda system really implementing three methods um to handle the full life cycle of the uh of the uh
[28:52] operator and the workload you're orchestrating. And I'll give it back to Sad because we took these examples and we took these uh projects and we gave it
[29:06] to our fields engineers. >> Yeah. Because the the first question you're going to be asking is of course uh thank you data bricks you gave us the tools but how about you ship some of these yourselves. So we basically
[29:18] enlisted our field uh engineering team. This is the honor role here. We just wanted to take a moment to acknowledge their contribution. Uh so they participated in a hackathon and um we have a ton of these that are going to be
[29:32] coming to a uh GitHub repo near you and uh we will start actually graduating these into the task pallet as well over time but uh yeah I can see the cameras are coming out uh so you're going to
[29:44] hold me to all of these uh I'm sure um all right so we >> you want to maybe call out a few questions >> actually there's uh the thing I'm most excited about is human in the loops So we actually talked to a customer
[29:56] yesterday where they had um uh approval chains that they have for uh promoting their pipelines. And so you want to be able to uh have a mechanism where you're
[30:08] not constantly running compute as Gabbor uh mentioned earlier with the uh sensor pattern. You only run compute when you're polling. So you could be you have this durability and deferability built
[30:20] right into the into the uh sensor as well. Uh obviously the AI ML ones are also quite uh common. Uh we are seeing a lot of these actually each one of these is backed by some sort of a customer
[30:33] need that uh is it's solving. So I'm really really excited super excited uh about this. Uh all right so obviously when can we use this? Please bring your cameras out
[30:46] again. Uh scan the QR code. There's a bunch of features that we just could not get through. Uh even though we have some time left over, Gabbor, the demo gods were with us today. Um but there's a lot
[30:59] of features that we actually didn't talk about. Um uh there's DAB's uh authoring visually. There's a lot of integration with um with Genieode and Genie Zero Ops, which I'm sure a number of you are
[31:12] excited about. Um one important call out is that it's using the same Genie framework. So you're not going to look at two different recommendations. So there is this inherent um let's say uh determinism in how Genie interacts with
[31:24] you. So any of the RCA uh that Gabbor was showing is actually built straight into zero ops as well.

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.