About this transcript: This is a full AI-generated transcript of Data + AI Summit Keynote 2026 — Day 2 from Databricks, published October 28, 2026. The transcript contains 26,987 words with timestamps and was generated using Whisper AI.
"*music* *music* Welcome to the stage, Databricks co-founder and CEO, Ali Godzi. *music* All right, super excited! Wow, everybody made it, 8 a.m. It was hard for me, actually. All right, so today we have a super exciting program, but I want to recap a little bit what we talked about yesterday, and..."
[00:00:00] *music*
[00:00:30] *music*
[00:00:53] Welcome to the stage, Databricks co-founder and CEO, Ali Godzi.
[00:00:59] *music*
[00:01:08] All right, super excited!
[00:01:10] Wow, everybody made it, 8 a.m.
[00:01:12] It was hard for me, actually.
[00:01:13] All right, so today we have a super exciting program,
[00:01:17] but I want to recap a little bit what we talked about yesterday,
[00:01:20] and then I'm going to tell you sort of the talks that are coming up,
[00:01:23] and then we'll get into it.
[00:01:24] Okay, so as I mentioned yesterday,
[00:01:27] we think that the AI is plenty smart already.
[00:01:29] We don't need more intelligence for the AI, we need more context.
[00:01:32] If we can just infuse all the context that we have inside all of these organizations
[00:01:36] that you all represent to the AI, then we can actually do magic.
[00:01:40] And that's our job, essentially.
[00:01:43] So, how do we do that?
[00:01:45] And we said that there's really context itself is four things that we sort of have to nail.
[00:01:50] Number one, we need choice.
[00:01:51] So, openness is extremely important for us.
[00:01:54] The fact that we can choose any model, any infrastructure, any cloud, any file format.
[00:02:00] Avoiding lock-in, I think, is one of the most important things that Databricks can help with.
[00:02:04] That's a mission we've been on since we started the company.
[00:02:07] Since we, you know, our roots in this conference, the open source community.
[00:02:12] Because over the last 20, 30 years, the main thing that's slowing down organizations
[00:02:17] is the fact that they're getting locked in to various types of software that they keep adding to their complicated stacks.
[00:02:23] So, that's number one.
[00:02:25] Number two, we have to be able to control the AI.
[00:02:28] Make sure that we have security rules, policies, and auditability over the AI, the data, and the infrastructure.
[00:02:35] And then, cost controls.
[00:02:37] This is going to get extremely expensive.
[00:02:39] So, I think you all already know that AI can get very expensive quickly.
[00:02:43] But I think we're just scratching the surface.
[00:02:45] I think in the next 6 to 12 months, most organizations are going to see that the costs are going up so much
[00:02:51] that it's actually prohibitive to their companies.
[00:02:54] And then, finally, context.
[00:02:55] We talked about it yesterday.
[00:02:56] Genie ontology is how you can actually compute in the background, continuously,
[00:03:02] and learn from all the data that you have in the organization.
[00:03:06] Not just in Databricks and Lakehouse.
[00:03:07] And that really helps speed up the agents, and it actually lowers the cost, and actually increases the quality.
[00:03:13] So, it's kind of like an index, if you will.
[00:03:16] Kind of like a page rank that Google did, but for the enterprise.
[00:03:20] Okay.
[00:03:21] So, this is what the architecture we presented yesterday looked like.
[00:03:24] I'm not going to walk through all of it here, but I'm just going to highlight the talks that are coming up.
[00:03:29] Okay.
[00:03:30] So, one of the things I'm really excited about is that Casey and Ankit are going to get up here and talk about Agent Bricks.
[00:03:36] So, this is the developer side of agents.
[00:03:39] So, if you want to build your own agents, Agent Bricks gives you all the tools for that.
[00:03:43] Okay.
[00:03:44] It gives you sandbox, it gives you memory, and it also comes with open source Omnigent that we're going to talk about today.
[00:03:49] But when you combine Agent Bricks, Omnigent, and Unity AI Gateway, you can actually start doing really interesting things with cost control.
[00:03:58] So, we'll see a demo of that, and I'm very excited about that.
[00:04:01] I hope that most of you actually look into how to do this.
[00:04:04] So, the idea is you use this advisor pattern.
[00:04:08] And the advisor pattern, the way it works, is that you use a much cheaper, smaller model.
[00:04:13] It could be an open source model.
[00:04:15] But you tell it, hey, here's a tool you can use.
[00:04:17] This tool is a much smarter, bigger model that's expensive, but only use it when you get stuck.
[00:04:23] And if you do that, you can now use that with your coding agents.
[00:04:26] You can use that with, you know, any AI work that you're doing in the organization.
[00:04:30] And we've seen that really curve the costs, while at the same time allowing you to keep the intelligence that you have.
[00:04:36] So, that's a really exciting demo and an exciting talk that we're going to hear about this.
[00:04:41] We're going to go back a little bit to Genie code.
[00:04:44] I mentioned yesterday that Genie code is really good at writing data engineering pipelines.
[00:04:48] We saw with zero ops that it could automate the headaches of pipelines going down and troubleshooting them and fixing them.
[00:04:55] But Genie code is also really amazing data scientist and machine learning agent.
[00:05:01] So, we'll hear a whole talk and demo by how actually you can write whole pipelines and also troubleshoot the pipelines and fix them and sort of get into the automated loop of machine learning and democratize that inside the organization.
[00:05:14] So, that's really interesting talk.
[00:05:16] We're going to look at apps.
[00:05:19] This is very exciting.
[00:05:20] As I said, we think there's new agentic system of record or agent system of record that's appearing where you can actually start democratizing apps inside of your organizations to tens of thousands of employees, hundreds of thousands of employees.
[00:05:34] And the key thing there is, how do you actually authenticate and authorize and give access to the data?
[00:05:39] But we're going to go far beyond that.
[00:05:41] So, a very exciting talk on that.
[00:05:43] And then we're going to see LakeWatch, which is the security information event management system that we launched two months ago.
[00:05:51] We're going to see how we've extended that.
[00:05:53] We'll also hear a little bit from Panther, the company that we announced that Databricks has agreed to acquire.
[00:05:59] And then finally, customer lake.
[00:06:02] This is the CDP.
[00:06:03] On LakeWatch and CDP, we also made it a little bit more tutorial-like so that we can bring everybody along if you don't have a background in how sims work or how customer data platforms work.
[00:06:17] But next up, we're actually going to have Matei come on stage.
[00:06:21] And he's going to talk about the new exciting open source project called Omnigent that he actually open sourced over the weekend.
[00:06:27] And it's central to everything we do at Databricks.
[00:06:29] It's actually the underpinnings of Genie.
[00:06:31] We actually built that on that pattern ourselves.
[00:06:33] So that's super exciting.
[00:06:35] And I want a request here for the camera crew.
[00:06:38] Please zoom in on his shoes, okay?
[00:06:40] And make sure you capture that, because that's going to be a really important part of his talk.
[00:06:45] But before we hand it over to Matei, I wanted to show you a quick video from a very important deep partner of ours, which is Mukesh Ambani of Reliance Industries in India.
[00:06:57] So Mukesh and Reliance Industries, they're on this amazing mission that we've been partnering with them on, which is how do you democratize AI?
[00:07:07] How do you democratize data to everyone in the world?
[00:07:13] There's a lot of folks that cannot afford very expensive models.
[00:07:17] So how do we reduce our reliance on these very expensive models and have maybe open source models that are more available and where the costs are much more manageable and then actually get it out to everybody and actually help with sovereignty in the many different nations that need AI as well?
[00:07:33] So very excited about that. So with that, I want to play you this video. Thank you.
[00:07:47] Dear Ali and all the delegates at the Databricks AI Summit. Namaste. Greetings from India.
[00:07:59] Thank you for inviting me to address this prestigious gathering of the global data and AI community. Today, with Reliance Intelligence, our mission is to democratize intelligence for 1.5 billion Indians.
[00:08:15] Our motto is simple. AI everywhere for everyone. We want AI to create a more equal India, bring prosperity to every farmer, entrepreneur and neighborhood shopkeeper, transforming education, healthcare and public services for every Indian.
[00:08:37] Solutions that work for everyone. Solutions that work for India can serve other countries in the global south, where four-fifths of humanity resides.
[00:08:46] Partnering with the best AI companies in the world allows us to make progress in this mission.
[00:08:52] That is why we chose Databricks. What impressed us most was not only the technology, it was the alignment of vision.
[00:09:01] Ali and his team understood that our ambition was not simply to modernize infrastructure. It was to build the foundation for innovation, productivity, customer value and societal value.
[00:09:16] Together, we are creating a unified intelligence platform across Reliance. We have migrated petabytes of data across geo, retail, energy, materials, media and beyond to serve more than 800 million customer relationships.
[00:09:34] This is one of the most ambitious enterprise data transformations undertaken anywhere in the world.
[00:09:42] For many years, enterprises used data mainly to look back. The next generation will use data to make better decisions, act faster and continuously improve.
[00:09:55] This is the shift from dashboards to decisions. Our leaders are already using Databricks Gini to ask questions in natural language and receive useful intelligence in real time.
[00:10:09] We are now scaling this to thousands of decision makers across Reliance.
[00:10:14] Friends, India is perhaps the most demanding testbed for AI because of our scale of aspirations, diversity, language needs, affordability constraints and speed of adoption.
[00:10:28] If an AI solution works for India, it can serve the world. Together with Databricks, we hope to create a blueprint for how enterprises can harness data and AI responsibly, openly and at scale.
[00:10:43] Ali, thank you for your partnership, friendship and vision. The best of what Reliance and Databricks can build together is yet to come.
[00:10:53] My best wishes to all the participants of your summit. Thank you and have a good day.
[00:11:04] Welcome to the stage, Databricks co-founder Matej Zaharia.
[00:11:09] Hi.
[00:11:10] All right. Hi, everyone. Welcome to the second day of keynotes. So I'm excited to kick off the morning today and talk about a new open source project we launched over the weekend.
[00:11:31] We just released it Saturday. This is a new layer from our engineering team for working and building with agents.
[00:11:39] And it's based on patterns and challenges that we saw coming up everywhere where people are using agents.
[00:11:44] And we realized that we're going to need this new layer on top of the current agentic stack to make it really effective.
[00:11:52] So I'll talk about what that means, but we call it a meta harness, a harness of harnesses, if you know what an agent harness is.
[00:11:59] Okay. So as everyone probably knows, agents are really starting to work. Coding agents are the first type that were really successful.
[00:12:09] Pretty much every developer has probably used them at this point. But the same is true in other fields.
[00:12:14] And you saw with Genie and with other agents that you can make very effective agents. And especially because coding is a lot easier now, you might wonder, do we even need agent infrastructure at all?
[00:12:28] If I don't like the current agents, can I just vibe code my own? So it seems like, hey, maybe we're kind of done with at least simple software and we're set.
[00:12:37] And this is what the world should look like. You know, agents are everywhere. Things are working.
[00:12:41] But actually, when you use agents, it can be quite annoying. If you're a manager, it can be very scary because of security issues and stuff like that.
[00:12:51] So, you know, the world isn't really perfect yet. So there is a problem. And actually, I didn't realize until I was making the talk and looked very carefully at this image that we even see part of the problem in the image.
[00:13:03] It's the harness, the thing around our little agent there. Okay. So what's an agent harness? So the agent harness is basically the software around the language model that is the agent's interface to the world.
[00:13:19] So this is something like Claude Code as a coding agent. There's Claude the model. That's a thing that just predicts that outputs tokens.
[00:13:28] And then there's Claude Code the harness, which connects it to files and a user interface and security permissions and stuff like that.
[00:13:35] And then there's Codex. And then there's, you know, any custom agent you make, you have some kind of software around it called the harness.
[00:13:42] And because it's so easy to build harnesses and because it actually helps to co-design the harness with the language model, everyone is building them.
[00:13:51] And indeed, everyone has. So there are so many agent harnesses here. These ones here are just some of the popular ones for coding.
[00:13:58] And then if you make custom agents, there are lots of agent SDKs and stuff like that. It's just a ton of these out there.
[00:14:06] Actually, this weekend I was at the pet store and this reminded me a lot of it. Like pick the harness for your agent, any shape, any color, you can put it in there.
[00:14:15] So what's the problem with these? Well, the issue is that agents sometimes have to interoperate.
[00:14:21] And there are also a lot of common challenges that the harnesses have where it would help to have a common layer that lets us interface across them.
[00:14:30] And that requires people to sort of build the layer and agree.
[00:14:33] If everyone is just vibe coding their own thing, they will still be incoherent.
[00:14:37] So, you know, there's no amount of like faster coding will solve this problem.
[00:14:41] So what are some of the challenges with interoperability of these?
[00:14:45] So the first is composing agents.
[00:14:48] When I'm doing coding and when a lot of the engineers at Databricks do coding, they use many coding agents at the same time.
[00:14:55] If there's a bargain, you know, one of the agents is stuck, they ask another one.
[00:14:59] They give them different inputs. They ask them to work on different things in parallel.
[00:15:03] So how do you compose agents with different harnesses?
[00:15:06] It's easy to do within one of them, like cloud code or codex, but like what if you want to use both?
[00:15:11] And there's a lot of other techniques for using multiple models to get better costs, better quality, and so on.
[00:15:18] The second one that's interesting is collaborating across humans and agents.
[00:15:22] You know, if I'm using one coding agent and someone else is using another one, how do we collaborate?
[00:15:28] Because ultimately we have to get stuff done together.
[00:15:30] So it's actually a huge problem that, you know, that we think we can we can improve on.
[00:15:35] And then the final one is I've got all these agents, all these different pieces of software, the harnesses.
[00:15:41] How do I control all of them for cost and security?
[00:15:44] So we start these problems repeatedly in Databricks actually from two angles.
[00:15:49] First, the engineering side where we have our own info around coding agents.
[00:15:54] And, you know, all the thousands of engineers are using them.
[00:15:58] And also on the customer facing side where we build a GenTech products like Genie
[00:16:02] and we have to keep changing how they work and we build custom agents for our customers.
[00:16:07] So let me dive into each of them a little bit more.
[00:16:11] So composition, there's so many patterns emerging now for how to combine agents to get a better result.
[00:16:18] So if you've looked at coding, maybe you've heard of loops.
[00:16:21] This is how you have some agents maybe supervise each other or run in a loop to try to finish a task.
[00:16:26] And there's all these patterns for loops.
[00:16:29] Another pattern that's emerging that's really interesting is called advisors.
[00:16:33] So you can have a small, fast, cheap model that's doing a bunch of the work and then an expensive model.
[00:16:41] And maybe the expensive model is like driving the small one, getting it to do the busy work.
[00:16:46] Or maybe the small one is calling to the expensive one and it's stuck.
[00:16:49] So Harvey AI had this blog post on custom trained model with a frontier model as an advisor.
[00:16:55] Anthropic has a post about it.
[00:16:57] Cloud Code actually uses multiple sizes of models for different things.
[00:17:01] Actually, my research lab at UC Berkeley, we wrote a paper on advisor models too.
[00:17:06] And then there's many other patterns of composing them.
[00:17:10] We have a post about how in Genie, if you use multiple LLMs, you can get better quality and better cost.
[00:17:16] So that's the composition that does kind of continue to be an issue as agents and models evolve.
[00:17:22] Second one is collaboration.
[00:17:24] You know, whose desktop looks like this?
[00:17:26] This is what my life is like all the time.
[00:17:28] I've got all these agent windows open, plus things like Slack and Google Docs.
[00:17:33] And all I'm doing is copy pasting text across them.
[00:17:36] This is pretty ridiculous.
[00:17:38] There should be an easier way to collaborate, but each agent has its own harness, its own app, and it's stuck there.
[00:17:45] And again, if you're a company managing this, it's also a huge pain to manage everything.
[00:17:51] So just as a simple thing, what I am working with one of these coding agents, and I have to go away or whatever,
[00:17:58] why can't I just hand my session to another engineer and have them continue with the full context?
[00:18:03] That's a simple thing that any harness would benefit from.
[00:18:07] And then the final one is control.
[00:18:09] Agents can be hacked through prompt injection, so you know, you shouldn't really trust them.
[00:18:16] Agents will also make mistakes even without being hacked, like deleting databases and so on.
[00:18:21] With Lakebase, you could just revert the database, but in general, you know, it'll be an issue.
[00:18:26] And they can also cost a lot.
[00:18:29] And people might say, okay, well, agents are very dangerous.
[00:18:32] Let's ask users for permission.
[00:18:34] Here's the average permission prompt when Claude asked me something.
[00:18:37] It's like, do you want to run this 100-line Python script, yes or no?
[00:18:41] So, you know, it's a huge pain as a user.
[00:18:44] Just waiting, you know, just hoping that users will review every action is not enough either.
[00:18:49] So this control and security layer is another common problem across harnesses that we want to solve.
[00:18:55] So we think that we need a new layer above the harness level to manage and build with agents.
[00:19:02] And we also think the layer should be open so anyone can plug in their own harness, their own workflows, their own tools.
[00:19:08] And we call this layer the meta harness.
[00:19:11] So that's what we decided, you know, across teams and Databricks.
[00:19:14] And then we decided we have to make it open source because that's going to be much better if anyone can contribute and it works with the latest things.
[00:19:22] And that's what we just released with Omnigent.
[00:19:25] So how does Omnigent work?
[00:19:27] So at the bottom, you've got existing agents.
[00:19:30] It works with coding agents.
[00:19:32] Actually, just since Saturday, we added two more, Antigravity and Courser, that it supports because people were asking for them.
[00:19:39] It also works if you're building custom agents with something like OpenAI Agents SDK.
[00:19:44] You've got a lot of the same features on top.
[00:19:47] Now, all these harnesses are different pieces of software, sometimes tailored to that LLM.
[00:19:52] But what we do is on top of them, we give them all a common interface so we can do the collaboration and composition and sharing that Omnigens try to do.
[00:20:03] And the interface is basically, you know, you can send messages in, you can get a stream of events out, they can call tools, that's what it is.
[00:20:11] So Omnigens then has two pieces.
[00:20:13] First, there's a runner, which wraps around your agent and can enforce security and other types of controls by sandboxing it and just monitors it and make sure it's running.
[00:20:24] And then you can also optionally connect Omnigens to a server where you get collaboration and central policy features.
[00:20:33] So with your server, if you connect your agents to that, you get a central history for everything.
[00:20:39] You can get a catalog to share agents with people.
[00:20:42] You can set up MCPs and skills centrally.
[00:20:45] You can also set up security policies and share artifacts out.
[00:20:49] And then on top of this, no matter what kind of agent you have, you also get a uniform interface.
[00:20:54] So you still get the agent's terminal UI if it has that.
[00:20:58] But any agent is also accessible on the web, native app, mobile, and also REST API.
[00:21:04] And Omnigens is open source.
[00:21:07] It does not require Databricks in any way.
[00:21:10] We developed it so you can just run it anywhere.
[00:21:13] And we can use it both at small scale and large scale.
[00:21:16] So it just uses standard technologies like Docker and Postgres and OpenTelemetry to run the server.
[00:21:22] OK.
[00:21:23] So what does this give you now that you have this layer?
[00:21:26] So there's three different things.
[00:21:28] So first, for composition, you can now easily combine agents that use multiple harnesses and RLMs.
[00:21:34] You can actually just write a little YAML file to define an agent that's using multiple of these together.
[00:21:40] It's also super easy to switch agent harnesses while the agent is running and continue with a different one, which I'll show in a bit.
[00:21:49] Second, you get collaboration.
[00:21:52] So we support live collaboration.
[00:21:54] Read.
[00:21:55] You can even invite other people to write to your session and talk to the agents and comment on files and things in real time.
[00:22:02] So this makes it way easier to work with these.
[00:22:05] Once this is available, like all my work I try to do through this.
[00:22:09] And then finally, for control, we looked a lot at how can we make security around agents better.
[00:22:15] And we actually came up with what I think is a much nicer, more powerful model than what you have today.
[00:22:21] Today, in most agent tools, the controls are just for each type of tool call.
[00:22:26] Do I allow it or deny it or ask the user?
[00:22:30] They're very static.
[00:22:31] We support contextual policies where the permissions vary based on what the agent did before.
[00:22:37] So, for example, if the agent read a confidential document, maybe we don't let it send an email outside the company anymore.
[00:22:44] But if it only read, you know, public marketing docs, then it's okay to send an email.
[00:22:48] That's-- so tracking state makes it much more powerful.
[00:22:52] And we also have policies on cost, which none of the agent frameworks do.
[00:22:56] So you can tell her, you know, don't spend more than $5 on this, for example.
[00:22:59] We also built a very flexible OS sandbox to protect the agent from harming your machine.
[00:23:08] So that's three things that it does.
[00:23:10] To show you a little bit of this in action, I'd like to show a demo.
[00:23:15] Okay.
[00:23:23] All right.
[00:23:24] Cool.
[00:23:25] So I'm here with Omnigine setup.
[00:23:29] So this here is-- this is actually my dev box inside Databricks.
[00:23:34] It's like a machine that I develop on.
[00:23:35] And then there's my laptop.
[00:23:37] So the first thing I'll show you with it is how it just wraps existing agents.
[00:23:41] So I'm just going to start Claude, but with this Omni in front of it.
[00:23:46] And that's just going to start it.
[00:23:49] It's starting this runner.
[00:23:50] And then, as you'll see, we'll basically just have the standard Claude interface.
[00:23:54] So this is actually the instance of Claude in Databricks that's been configured, you know, with all our settings.
[00:24:02] It's got all these, like, MCPs and, you know, different things in there, specifically for Databricks.
[00:24:07] And I can ask it to do stuff.
[00:24:09] So currently, I'm in the Omnigine project.
[00:24:12] So what is this project?
[00:24:15] I can ask it.
[00:24:17] And you'll see it's doing something.
[00:24:19] And we even customized these things that Claude says, archiving.
[00:24:22] If you don't know what archive is, don't worry.
[00:24:25] So there it is.
[00:24:26] It means something in Databricks.
[00:24:28] Okay.
[00:24:29] So now I'm just starting Claude.
[00:24:30] So that doesn't seem that exciting.
[00:24:32] But what do I get through Omnigine even as an individual developer with an existing agent?
[00:24:37] One of the things I get is a web UI for my agent, too.
[00:24:42] So the same session I was doing, I have it on the web.
[00:24:46] I can see all the files and things in here if I want.
[00:24:49] And, you know, I can see it here in a nice chat form.
[00:24:55] I can even see the same terminal in here if I want.
[00:24:58] So it's the same thing.
[00:24:59] But let me ask the agent to do something.
[00:25:02] Like, actually, since I was doing this before, I'm going to check out the readme first.
[00:25:13] Oops.
[00:25:14] Sorry.
[00:25:16] Because I was trying this before.
[00:25:18] So let me do add a joke to --
[00:25:24] So I'll just tell it to add a joke to the top of the readme.
[00:25:27] Okay.
[00:25:28] So I sent a message from the web.
[00:25:30] Now, in the terminal, you still see it here.
[00:25:33] So, like, we do this mapping of user input and output across the two interfaces.
[00:25:38] And, okay, it's saying, hey, I'm adding -- I've added a joke in there.
[00:25:42] If I look here, it says that it added it.
[00:25:45] Now, the other nice thing about the web UI is, of course, it's much more visual.
[00:25:50] So if I look here, I can see there's a change file to readme.
[00:25:54] I can see there's the joke at the top.
[00:25:56] But this is just like a markdown file.
[00:25:58] So I can just view the whole thing.
[00:25:59] I'm going to close this to make it a little bit bigger here for you.
[00:26:03] Let's move this.
[00:26:04] Okay.
[00:26:05] All right.
[00:26:06] So this is it.
[00:26:07] I can view the diff.
[00:26:09] And I can even put in comments about it.
[00:26:14] So, for example -- oh, I guess there were a few because I was doing this before.
[00:26:19] But maybe I can tell it, like -- I'll just tell it to delete the jokes.
[00:26:26] Okay.
[00:26:28] And then, I don't know.
[00:26:29] I could try to change other things.
[00:26:30] But I can leave some comments in here and tell it to address all of the comments.
[00:26:34] And this is just sending it.
[00:26:36] So I get this nice visual interface, and it's doing its thing.
[00:26:39] The other thing we built for it that's, like, very, very helpful -- I'll just tell it -- I'll tell it to just add one.
[00:26:49] The other thing we built is actually a native app.
[00:26:52] So in my browser -- sorry -- yeah.
[00:26:57] If I go here -- in my browser, you know, I've got all these tabs.
[00:27:01] It's hidden all the time.
[00:27:02] I closed all my tabs this morning.
[00:27:04] But in the native app, it's, like, very easy to find it, and you get notifications and stuff from the system.
[00:27:09] So let's look at this again.
[00:27:11] Okay.
[00:27:12] We got a joke in there.
[00:27:13] Okay.
[00:27:14] So what else can I do?
[00:27:15] The other thing I can easily do is I can switch agents.
[00:27:18] So I can take this same session and say, hey, let's actually continue from here with Codex and just fork the session.
[00:27:26] And it's going to start Codex here.
[00:27:28] And you'll see that it has the same history.
[00:27:31] Okay.
[00:27:32] So now in the terminal, this is actually OpenAI Codex.
[00:27:36] And I can just tell it add a second joke too.
[00:27:43] And you'll see it adding that in a second.
[00:27:47] This is a little bit slower, but it's going to be in there.
[00:27:51] So this is pretty cool.
[00:27:52] If I'm stuck with one agent, I can switch to another one.
[00:27:55] And then there are two of them.
[00:27:57] But then the super interesting thing that really is enabled by Composition is I can also use a multi-agent coding.
[00:28:07] So we actually include an example here, Polly, which is a multi-agent coding supervisor that can use both Cloud and Codex to do subtests.
[00:28:16] And I'll start it in the same folder and tell it ask--
[00:28:28] Okay.
[00:28:29] So I'll tell it to ask these guys to both rate the jokes at the top of the readme.
[00:28:34] And you'll see when it starts that it's going to kick off a couple of them.
[00:28:39] It's checking to see if I have Cloud and Codex installed.
[00:28:43] It's launching some workers, okay?
[00:28:45] So in here, you'll see some workers will pop up.
[00:28:48] There's a Cloud worker.
[00:28:51] And then it's going to start a Codex worker too.
[00:28:54] And it's telling them to rate these things.
[00:28:56] So these are, again, these are just Cloud code and Codex.
[00:28:58] They do the same things you could do otherwise.
[00:29:00] I can click on these guys and even see, you know, what they're doing in here.
[00:29:05] There's the terminal.
[00:29:06] They're trying to rate things, et cetera.
[00:29:08] If I go back at the top, you'll see that the supervisor hears.
[00:29:13] And eventually, it's going to say, okay, so first one, pretty good.
[00:29:17] Second one, not quite as good.
[00:29:19] So that's what Polly thinks.
[00:29:21] Okay.
[00:29:22] So obviously, I could do more things.
[00:29:24] I use these to do all kinds of other stuff.
[00:29:26] Like, for example, the thing I was doing last night is I asked to find all the open GitHub issues
[00:29:32] about something and then investigate them in parallel and see if they've been fixed in the code.
[00:29:38] So there's a lot of cool stuff you can do with this that I don't have time to show.
[00:29:44] But the final thing is-- so that was like composition of agents.
[00:29:48] The other thing you can do that's really cool is collaboration.
[00:29:51] So let's say I think I have a great change to the project here.
[00:29:55] And I want to get an engineer to review and commit this.
[00:29:58] I'm just going to call one of our engineers.
[00:30:01] Oops, sorry.
[00:30:04] And this is Ankit.
[00:30:06] He's on the AI team.
[00:30:08] And I think he's a good reviewer for this readme.
[00:30:11] So I can just invite him to it.
[00:30:14] And let me just give him a call here.
[00:30:17] Let's see.
[00:30:24] Oh, man.
[00:30:25] Matei's calling me on vacation?
[00:30:30] Hey, Matei.
[00:30:31] Hey, man.
[00:30:32] You busy?
[00:30:33] Not really.
[00:30:35] Okay.
[00:30:36] I got this important change to the readme.
[00:30:37] Can you take a look and tell me which version you prefer?
[00:30:40] Thanks.
[00:30:41] Yeah.
[00:30:42] I shared my session with you.
[00:30:43] All right.
[00:30:44] Let me take a look.
[00:30:48] I know the agent said that joke one is better.
[00:30:51] But I kind of like joke two.
[00:31:00] So as you can see, thanks, Ankit.
[00:31:03] That was very easy.
[00:31:04] So I invited him.
[00:31:05] He can see the same session.
[00:31:07] He can leave comments and things.
[00:31:09] And he can send a message in there and tell my agent to do things.
[00:31:12] So I can hand off work to someone else.
[00:31:15] And obviously, all the history, all the sessions, any agent, whatever,
[00:31:19] are also stored on your server.
[00:31:21] So you can analyze them later and do things.
[00:31:24] Okay.
[00:31:25] So final thing I'll show is policies.
[00:31:27] Actually, I guess I can show it in here, too.
[00:31:29] So the stateful policies.
[00:31:30] So notice at the top here, we actually track the cost of each agent, how much it's spent.
[00:31:38] And we also allow you to set policies.
[00:31:40] And these are contextual.
[00:31:41] They can depend on the current state.
[00:31:43] So one of the policies we allow is set a budget on it as it's spending in this session.
[00:31:51] Like, you don't let it spend more than, say, a dollar without asking the user.
[00:31:55] So I'll just make something very silly here.
[00:31:58] It already spent 25 cents or whatever.
[00:32:02] So I'll say, ask me if you're trying to spend more than 20 cents.
[00:32:05] And then I'll tell it, you know, how, you know, I'll just tell it to do something that's going to involve, you know, reading a lot of files.
[00:32:20] And you'll see, as it's thinking, eventually when it begins to do stuff, it's going to ask me for permission because it's out of budget.
[00:32:27] So this is, like, pretty nice.
[00:32:30] You can also tell it, like, launch a sub-agent, but only let that agent spend a dollar.
[00:32:35] There's all kinds of things.
[00:32:36] And if you're managing this for a team, you can set policy centrally.
[00:32:40] And if I approve it here, it's going to continue.
[00:32:42] And eventually, when it gets to $5, it's just going to stop.
[00:32:45] So this is a really cool feature, again, on top of any agent, clod, codex, whatever it is.
[00:32:51] And it's just scratching the surface.
[00:32:52] You'll hear more about these soon.
[00:32:54] Cool.
[00:32:55] So that's the demo.
[00:32:56] So let me just wrap up with a couple more things.
[00:32:59] A couple more things in the slides.
[00:33:01] Thank you.
[00:33:05] Cool.
[00:33:06] So one thing you may be wondering is there's a lot of functionality in this.
[00:33:09] Why are we open sourcing it?
[00:33:11] So we really think working with agents has to go above the level of a single harness and SDK and so on.
[00:33:18] We think this meta harness layer is going to be needed.
[00:33:21] And it will benefit from being open the same way that data formats and sharing protocols and stuff like that benefit from being open.
[00:33:29] Because it's an ecosystem things where people sort of have to agree on a way to do it to make it work well.
[00:33:34] So that's why we're making it open source.
[00:33:37] We use it a lot internally.
[00:33:39] We hope you use it as well.
[00:33:40] And let's figure out together how to build things with this.
[00:33:44] We already got a lot of really great contributions since Saturday.
[00:33:47] The other cool thing about open sourcing this is everyone we saw building an agent wants an app like what I showed.
[00:33:53] Maybe not for coding, but with collaboration, sharing, security, all that stuff.
[00:33:59] So now you've got an open source app you can use to build any agent with a great interface.
[00:34:04] I didn't show it, but you can even get a mobile interface to your agent through this.
[00:34:08] So we have a lot more coming up, more ecosystem support, also auto-optimization of the agents,
[00:34:14] which you'll hear about hopefully in future events.
[00:34:17] So that's Omnigent.
[00:34:19] We really think that AI needs a layer above the harness.
[00:34:22] It makes it way easier to work.
[00:34:24] Certainly, our team has been working with this all the time.
[00:34:27] We're developing not just the engineers, but product managers and designers and so on.
[00:34:33] And we invite you to check it out.
[00:34:34] It's early on, but we think it will be an important layer.
[00:34:37] Thank you.
[00:34:44] Hundreds of thousands have come to Databricks to learn the skills that actually matter in data and AI through Databricks Free Edition.
[00:34:51] From building agents, to analyzing biomedical literature, to unifying fragmented environmental data sets, there are endless ways to apply Free Edition.
[00:35:00] Getting students access to the Free Edition before they enter the workforce helps bridge that gap between classroom learning and industry practice.
[00:35:09] It gives them access to a modern data and AI platform that is extremely relevant for employers.
[00:35:16] One of the biggest advantages of working with students with prior Databricks experience is that they can get started much more quickly.
[00:35:23] Rather than learning the platform, they can spend their first few weeks working to solve real business problems.
[00:35:29] Today, we're announcing the availability of Genie, GPUs, AgentBricks, LakeBase, and Lakeflow Designer on Free Edition, where you can learn on a production platform.
[00:35:50] This means you get virtually all of Databricks for free, forever.
[00:35:54] Hundreds of thousands of people have already built data and AI projects and learned on Databricks.
[00:35:59] Sign up and start building today.
[00:36:00] Welcome to the stage, Databricks Director of Product Management, Casey Uhlenhut.
[00:36:08] Hi, everyone.
[00:36:21] So, yesterday was about consuming AI.
[00:36:25] You saw how Genie allows you to get AI-powered answers to all of your knowledge workers.
[00:36:30] But today is about developers.
[00:36:33] Developers are the teams who are building agents that run your business and provide amazing customer experiences in your products.
[00:36:42] And AgentBricks was built for them.
[00:36:44] AgentBricks is our agent developer platform.
[00:36:47] We launched it a year ago, and we've seen amazing success so far.
[00:36:52] Over 100,000 custom agents have been built on AgentRanks since we launched.
[00:36:57] We've had over a quadrillion tokens go through our AI gateway.
[00:37:03] And our MLflow solution, which provides GenAI experimentation and tracing, has had over a billion downloads.
[00:37:11] And our customers are seeing a lot of value from this solution as well.
[00:37:15] Where 7-Eleven is able to repair maintenance fixes faster.
[00:37:19] MasterCard is able to detect fraud across millions and billions of transactions.
[00:37:26] And AstraZeneca is able to help their drug discovery team get commercial insights faster.
[00:37:34] This is all amazing, but it's only one piece of the picture, and we're only just beginning.
[00:37:42] When you see agent demos and agent pilots, you're usually just seeing the agent loop.
[00:37:49] But the agent loop is only 1% of the work that you have to do.
[00:37:54] It's the other 99% that's going to get you.
[00:37:59] Where you have to figure out authentication.
[00:38:01] You have to figure out how to connect to MCPs.
[00:38:03] How you're going to manage identity.
[00:38:05] How you're going to have this run for a long time without breaking or hitting time outs.
[00:38:12] And you have to do all this work for just one agent.
[00:38:16] But you're trying to build thousands of agents in your company.
[00:38:21] So it's no wonder that we find that developer teams end up stuck building infrastructure rather than building agents.
[00:38:31] And as we've talked to developer teams, there's really three problems that start to emerge for them.
[00:38:36] The first is that the frontier seems to change every week.
[00:38:41] A new model comes out.
[00:38:42] A new agent harness comes out like you just saw with Omnigent.
[00:38:45] And now the solution that you built six months ago is at least one generation behind, if not more.
[00:38:53] And if you want to be able to change to be on the cutting edge, you might have to actually change stacks.
[00:38:59] Which then means you have to replatform your entire solution.
[00:39:02] The next problem that you have is that your competitive advantage is in your data.
[00:39:08] But your data can be fragmented across many different sources.
[00:39:12] And your data can come in many different formats.
[00:39:15] All the way from PDFs to external APIs.
[00:39:19] And it's really hard to help your agent select the right context from all this data.
[00:39:25] And lastly, agents are the most privileged actor in your enterprise.
[00:39:31] Where they have access to sensitive customer data and they can take real actions.
[00:39:37] And that's why you're seeing the headlines emerge with tons of these different scary headlines.
[00:39:43] Where agents are deleting production databases.
[00:39:46] They're leaking sensitive data.
[00:39:48] And to make it even worse, these agents can spend a lot of money and not a lot of time.
[00:39:54] So as a developer, you could end up blowing through your budget before you even got to production.
[00:39:59] These are the challenges that we keep hearing over and over again.
[00:40:03] Which is why we believe an agent developer platform must provide choice, context, and control.
[00:40:10] Agent Bricks is our developer platform where we provide answers to all of these.
[00:40:16] So that developers can focus on building agents rather than on building infrastructure.
[00:40:22] So now we're going to go through how Agent Bricks provides each of these values.
[00:40:27] So we'll start with choice.
[00:40:30] Agent Bricks provides native model access and integration with all of the frontier models.
[00:40:36] So you get Opus.
[00:40:37] You get GPT.
[00:40:38] You get Gemini.
[00:40:39] You get Quinn.
[00:40:40] And even recently you get Kimmy.
[00:40:42] And today we're excited to announce in partnership with SpaceX that you now get access to the Grok model family as well.
[00:40:50] Thank you.
[00:40:55] But sometimes the frontier models aren't enough.
[00:40:58] You might have a very specific task that you need to accomplish.
[00:41:02] And it turns out you're going to need to build a custom reinforcement learning model.
[00:41:07] So we also provide an entire model platform for you to use reinforcement learning and build custom models that are trained for your specific task.
[00:41:15] In fact, our own Genie research team recently used RL and was able to actually outperform Opus on their tasks in there.
[00:41:25] And you're able to serve and deploy these models alongside all of the rest on the Agent Bricks platform.
[00:41:30] But choice isn't just about models.
[00:41:33] When you're building custom agents, you want to use your own framework or build your own.
[00:41:37] And so today we're excited to announce the Omnigent that you just saw with Matei.
[00:41:42] You can now build and deploy all of those agents inside of Databricks.
[00:41:46] And, of course, you can continue to work with your agent framework of choice, whether it's the OpenAI, Agent SDK, LandGraph, or many more.
[00:41:54] So we believe choice is about giving you the flexibility to stay at the cutting edge without having to replatform your solution every time.
[00:42:02] And we're able to do this by giving you a unified interface across all of your models so that you can easily switch amongst them.
[00:42:09] The best model in the world, though, is only as good as the information you give it, which takes us to context.
[00:42:17] Context is how your agent understands about your users and your business.
[00:42:22] And, as we mentioned before, context is just data.
[00:42:26] So what we do in Agent Bricks is we allow all of your context to be centrally managed inside Unity Catalog.
[00:42:32] So you can connect your external data there and all of your other data is accessible in there.
[00:42:37] And all of this is callable as tools by your agent.
[00:42:41] So your agent can discover all of them, including dashboards.
[00:42:46] But discovering context isn't enough.
[00:42:50] You also need to make sure their agent pulls the right context.
[00:42:54] And that's why the Genie Ontology announcement we made yesterday is such a big deal.
[00:42:59] Genie Ontology provides the metadata layer over all of your agentic assets so that your agents can effectively and efficiently retrieve and select context.
[00:43:09] So now that we have this foundation set up, the next hard part is sometimes your context comes in complex formats.
[00:43:19] You might have millions of PDF contracts and invoices that you wish you could feed to your agent, but you can't.
[00:43:26] And that's why we built document intelligence.
[00:43:28] Our document intelligence solution provides custom-built models for parsing, extraction, and classification.
[00:43:37] And our solution is the best on the quality cost curve out there.
[00:43:41] Because we know you have millions such of these documents and we don't want you to break the bank just so that your data or your agents can get access to that valuable data.
[00:43:49] And, of course, we know that not all of your data is inside Databricks, so we allow you to securely connect to all of your external APIs with our managed external MCPs.
[00:44:01] And we'll provide some of them out of the box for you so you don't have to worry about all of those authentication issues.
[00:44:08] The last part about context is sometimes when you're building these custom agents, you want to personalize them per user.
[00:44:15] But you don't want to build an agent per user.
[00:44:17] That's way too much.
[00:44:19] And so instead you want something that's like agent memory.
[00:44:22] So today we're excited to announce our agent memory services where you can connect your custom agents to our managed APIs.
[00:44:32] And we will store user preferences, conversation history, and sessions for you so that you can personalize your content to your users.
[00:44:50] Context is probably the most difficult problem that exists in the agent space today, which is why we have an entire research team dedicated to just solving the hardest problems here.
[00:45:01] So our research team is working to solve agent memory, agent retrieval, and agent search.
[00:45:06] And all of the solutions and techniques that they come up with are then baked into all of our Agent Brick services that we provide for developers, as well as they're integrated into Genie.
[00:45:18] So now that you have connected your agents to all of your knowledge, now you want to act on it, but safely.
[00:45:25] And so that takes us to control.
[00:45:28] There are two main products that we want to discuss in this control section.
[00:45:32] So as we mentioned earlier, agents can, you know, run wild out there and go rogue in your enterprise and even delete a production database.
[00:45:40] But you kind of want these agents to run code.
[00:45:43] They can do like amazingly creative things if you just kind of let them cook.
[00:45:48] But you don't want them to take down production.
[00:45:51] So you want to give them some kind of walled environment that they can operate in.
[00:45:55] And that's why today I'm excited to announce Databricks Sandbox.
[00:46:00] Databricks Sandbox provides VMs that you can have your agents securely run and execute code in.
[00:46:06] But this isn't just your generic VM solution.
[00:46:10] We are the only VM out here that provides you out of the box access to your data.
[00:46:16] So we will allow you to have downscoped access using on behalf of permissions to your data so that your agent can get creative with your data inside that VM that won't affect the rest of your system.
[00:46:33] Also, these sandboxes are incredibly fast.
[00:46:36] They take just milliseconds to launch.
[00:46:38] So you can create and destroy these sandboxes just as fast as your agent can work.
[00:46:43] Which makes this extremely exciting for being able to work and play with your context securely in your agents.
[00:46:51] And now for the piece de resistance, Unity AI Gateway.
[00:46:57] Unity AI Gateway is our centralized governance, security, and cost control solution that allows you to unify governance across models, agents, MCPs, and skills.
[00:47:10] And this is probably the most exciting product that we have that's really going to be a game changer that allows you to run your agents autonomously in your enterprise.
[00:47:20] There are five new capabilities that we've added to AI Gateway that we need to go through.
[00:47:27] So I need you all to buckle up because we're going to dive right in.
[00:47:31] So the first, remember those thousands of agents that your team is building?
[00:47:37] Well, they're all going to probably build them with different tools and platforms in there.
[00:47:42] And that's okay.
[00:47:43] But wouldn't it be great if you just had like one inventory of all the agents that were built in your enterprise?
[00:47:49] Well, now that's possible with our agent registry that's part of Unity AI Gateway.
[00:47:54] You can register all of your external agents, all your agents built on Databricks, all your MCPs, all your skills.
[00:48:00] So you have one system of record of all of the agentic assets in your organization.
[00:48:06] The next capability that we have is related to security.
[00:48:11] We provide contextual security policies.
[00:48:14] This one's best explained through an example.
[00:48:17] Imagine you have an agent that touches sensitive customer data with PII.
[00:48:23] You want to change the state of that agent to now say, hey, it's been tainted because now it has access to sensitive data.
[00:48:31] So when it takes its next action, you probably want to change what it can actually do.
[00:48:36] So you may allow it to still send an email to a coworker, but now you absolutely do not want to allow it to publicly post that data.
[00:48:46] However, if in a different agent session with that same agent, it only touched public information, then it's totally okay for it to then access and publish things to the public web.
[00:48:59] And this is the best illustration of what a dynamic and stateful policy can do that's contextual to your data.
[00:49:07] And this is only possible on Agent Bricks because our agent platform is built and integrated into our data platform.
[00:49:14] And so this is the kind of security policy that then is going to allow you to bound your agent to safely run autonomously.
[00:49:22] Because at every action and every step your agent takes, we can change how the policies are run.
[00:49:30] On the same hand with security, we have now partnered with many security organizations to allow you to have security guardrails that you can use to further control your agent behavior.
[00:49:43] Now that we've covered security, let's go into the thing that we're all thinking about, token maxing.
[00:49:49] Cost controls are a huge issue for many developers today and their organizations, where you can end up deleting your budget before you really even got that far ahead in your project.
[00:49:59] So now with AI Gateway, we support budgets, where you can set spend limits for different users and user groups in there.
[00:50:07] But what you really probably want is not just to know how much you've spent.
[00:50:13] You would actually rather there be like some intelligent system out there that kind of helps you control your spend automatically.
[00:50:20] So you don't have to really think about it.
[00:50:22] And that's where intelligent routing comes in.
[00:50:25] With intelligent routing, we can figure out what the best model is for the task based on the complexity.
[00:50:31] So we can route trivial questions to cheap, fast models, and we can route complex questions to expensive and powerful models.
[00:50:41] The other thing that we provide with smart routing is because we are multi-cloud and multi-AI, you can have the most highly available agent out there where you can set up fallbacks across your model providers as well as across your clouds.
[00:50:55] And the last major capability that we've added to Unity AI Gateway is agent tracing.
[00:51:03] Using MLflow tracing, we can log every input, every output, and everything that happened in the middle of your agents and centrally store them all in a Unity Catalog table.
[00:51:13] And you own that table and all of that data because it is super sensitive to your enterprise.
[00:51:19] Once all of those traces are in Unity Catalog, you can then use Genie to analyze them to debug what's going wrong with your agent.
[00:51:27] Or you can connect it to our SIM system with Lakewatch and get real-time incident alerts when data leakage happens.
[00:51:34] So this is what AI governance should look like in the enterprise, where all of your assets can be centrally managed.
[00:51:43] You can have contextual policies to enforce security.
[00:51:46] You have an open ecosystem where you can connect to partners.
[00:51:50] And you can have the cost controls that you need.
[00:51:53] All of this will allow you to run your agents autonomously.
[00:51:57] It also works really well for admins in your enterprise, where now you can actually safely and responsibly democratize AI across your organization.
[00:52:07] So on that note, I'm very excited to welcome to the stage my colleague Ankit Mathur, who is going to show a demo of how you can use Unity AI Gateway to democratize access to coding agents across your developer org.
[00:52:23] With that, please help me welcome to the stage Ankit.
[00:52:26] Hi, everyone.
[00:52:39] My name is Ankit, and I'm a software engineer at Databricks.
[00:52:42] And yes, I am back from vacation.
[00:52:44] Today, I'm going to show you how our Unity AI Gateway will help you roll out AI at your company while keeping costs in control.
[00:52:52] Now, as a software engineer, I myself use AI every single day.
[00:52:56] And what we're seeing is that leaning on AI is starting to become incredibly expensive as costs are growing exponentially.
[00:53:03] The top 1% of spenders are spending $90,000 per user per year.
[00:53:09] And that exponential isn't showing signs of slowing down.
[00:53:12] What started happening is that we don't just have engineers writing code anymore.
[00:53:16] Every single person at your company is using agents.
[00:53:20] They're using them in parallel.
[00:53:21] And the models are good enough now that they're going to keep working uninterrupted for as long as we let them.
[00:53:26] So it's no surprise that costs are growing out of control.
[00:53:30] But are we using AI responsibly?
[00:53:33] And what's going on inside of that bucket of spend?
[00:53:36] That's what the Unity AI Gateway can help you answer.
[00:53:39] And so in today's demo, I'm going to demo three of the six things that Casey talked about.
[00:53:44] The first will be budgets.
[00:53:46] The second will be smart routing.
[00:53:48] And the third will be agent tracing.
[00:53:50] So let's jump right in.
[00:53:52] And we'll start our day exactly where every other developer starts their day, which is inside of the terminal.
[00:53:57] And so with the Unity AI Gateway, you can easily point the coding tools that you already use towards the gateway.
[00:54:05] So all I need to do is something like this.
[00:54:12] And boom.
[00:54:13] Codex starts just as it normally would with a couple differences.
[00:54:17] It's now pointed at the Unity AI Gateway.
[00:54:19] So your inference and MCPs are tracked through a central choke point.
[00:54:24] And I can continue to see exactly the same Codex UI that I was using before.
[00:54:28] So I can ask a very simple question like, find me a coffee shop near Moscone Center.
[00:54:40] And it'll work just as normal.
[00:54:41] Except this time, your MCPs are getting routed through a central registry.
[00:54:44] So you only need to authenticate once.
[00:54:46] And your inference is being tracked.
[00:54:48] Now on the right here, you'll see something interesting.
[00:54:50] Which is that because all of our coding tools can point at the same gateway, we're able to specify and define a budget that spans all of the tools that we're using.
[00:54:58] So I, as an engineer, I have a budget of $50 a day.
[00:55:02] And the policies are a lot more subtle.
[00:55:04] So here you can see that Codex actually was able to warn me that I'm nearing a limit on my smart budget policy.
[00:55:10] So what does that look like?
[00:55:12] These are not as simple as hard caps.
[00:55:15] Instead, they're far more subtle.
[00:55:17] Since your developers are being extremely productive with AI, you probably don't want to stop them.
[00:55:21] And so when they start their day, you can use any coding tool and any model that's available to you.
[00:55:27] Because all of these are available through the gateway.
[00:55:29] But as you get closer to the top of your limit, we might start guiding you towards a more token efficient model like Databricks Kimmy, which comes out this week and is significantly cheaper than all the other models.
[00:55:41] So that's great.
[00:55:42] But then I can also go, when I get to 100%, I might actually start enforcing that that happens.
[00:55:48] Now, my session here continues just as normal because I don't want to get in the way of active work.
[00:55:54] But if I was to, for example, start something new, I can type the same thing, gateway codex.
[00:55:58] And this time, what would happen is that it's aware of our real time budgets because our budgets are burning down in real time.
[00:56:05] And so we can take a look at this.
[00:56:07] And it's automatically recommending that we switch the more token efficient model.
[00:56:11] So this is awesome.
[00:56:13] And it's a lot better than what we've already got.
[00:56:15] But can we do any better?
[00:56:17] And the answer is yes, because we have Omnigent.
[00:56:21] And when Omnigent can connect to the Unity AI gateway, we can do a lot more sophisticated stuff.
[00:56:27] Specifically, when you have a meta harness, you can make decisions at the higher level than an individual task, which allows you to make intelligent routing decisions.
[00:56:36] So here, we see the intelligent router when Omnigent connects through the gateway.
[00:56:40] And for that same query that I had before, which was relatively simple, what happens is that when I start running it,
[00:56:46] the intelligent model router makes a decision.
[00:56:48] And it's like, hey, this is actually a pretty simple local recommendation.
[00:56:51] I can use a cheap model for this.
[00:56:53] And it has this great advantage that it has really great latency when it answers my question.
[00:56:57] But if I was to do something slightly more complex, like in the Omnigent open source repo, find the top four bugs in the repo.
[00:57:05] Now, because Omnigent is a meta harness, it's able to take a look at that work and distribute it to other coding agents.
[00:57:12] And so for that distribution task, it actually picked a simpler, medium quality model.
[00:57:17] But for each of the individual things it's doing, it's able to make a more sophisticated decision.
[00:57:22] So for example, for a top level scan of the repository, which is very token expensive, it picked a cheaper model like Databricks Kimmy.
[00:57:28] For an edge case review, it picked something like Gemini 3.5 Flash, which is extremely intelligent but cheap.
[00:57:34] And for core logic reviews and deep audits, it picked medium and expensive models because their quality is extremely important.
[00:57:42] Now, all of these things get kicked off in the Databricks sandbox so they can work in parallel while still maintaining enterprise grade security.
[00:57:50] So I can click into any individual one of these agents, like for example the surface scan, and look at how fast Kimmy is running.
[00:57:56] Latency is another great advantage of being able to pick a cheaper model for the task.
[00:58:02] And so this kind of work can get done much faster, which frees you up to continue your work and work on something else.
[00:58:09] All right, so that's cool and I'm glad that we were able to do this.
[00:58:13] It's important to have a system like this in place so that you're having those intelligent routing decisions being made.
[00:58:19] But how do you know if that system is working?
[00:58:23] Well, this is where agent tracing is incredibly impactful.
[00:58:26] Because without data, we can't be data driven.
[00:58:29] And today, coding agent sessions aren't centralized to a single place.
[00:58:33] And so that's where, again, the Unity AI Gateway, when you connect a coding tool to it, can automatically route those traces to Unity Catalog.
[00:58:40] Now, this is some of the most sensitive data in your organization.
[00:58:44] Because it is extremely sensitive, your CEO might have access to context that you don't have access to, so you don't want just anyone to be looking at it.
[00:58:52] That's why when it's in Unity Catalog, you can use the same tools to govern that data as you already do for your most sensitive data in Databricks.
[00:59:00] Things like ABAC policies and govern tags mean that only the right folks have access to it.
[00:59:05] All of this goes into a central trace table that's available in the Unity AI Gateway.
[00:59:10] But another advantage of data being in a central place is that you can use Genie 1 to analyze it.
[00:59:17] So before we rolled out the router, for example, within our own engineering team, I was able to take a look at a sample of our coding traces and assess that, actually, because our default model was the most expensive one, like Opus, 90% of our sessions were overkill.
[00:59:31] That's already way more data than I would have had otherwise.
[00:59:35] Otherwise, I would have just been guessing.
[00:59:37] And the nice thing about that data being in one place is I can also then look at what kinds of things are going on in there.
[00:59:43] Why was it overkill?
[00:59:45] Well, you know, for example, was it just people saying hi or hello?
[00:59:48] Was it paved path tasks that are actually not so complex?
[00:59:52] This can help you guide what skills you need to build as well.
[00:59:56] Those skills can also be put in Unity Catalog's AI Gateway.
[01:00:00] And that will allow you to be more token efficient with the things that your developers and engineers are doing on a day-to-day basis.
[01:00:07] And it also gives you that full level of observability.
[01:00:12] So where does this leave us?
[01:00:15] We hope that with the Unity AI Gateways, budget policies, intelligent model router, and tracing, we can start to bend that exponential curve
[01:00:23] and still be aggressive about deploying AI within our organizations, but be more mindful about exactly which models we're using.
[01:00:29] Because intelligence is one of the most precious resources that we have in the ecosystem.
[01:00:34] And so hopefully we can take the world from the old world of token maxing to something a little bit more like value maxing with the Unity AI Gateway.
[01:00:42] That's all I have. Thank you very much.
[01:00:44] And back to Casey to wrap up Agent Bricks.
[01:00:46] Thank you Ankit for that amazing demo.
[01:00:58] So as you can see, Agent Bricks is our complete end-to-end agent developer platform.
[01:01:05] We believe it's important to provide choice so you have flexibility to always remain at the cutting edge without having to replatform your solutions.
[01:01:13] We provide you context so that your agents have access to the right metadata to grab the right data for all of your agent tasks.
[01:01:22] And as you saw with Unity AI Gateway, we provide controls so that you can run these agents autonomously in your enterprise.
[01:01:30] And also that Data Bricks sandbox so that you can have it securely execute on downscoped data.
[01:01:38] While I did show you a lot of features today, we didn't go through everything that we've launched in the last year.
[01:01:44] And so here's a list of all the many other features that we've implemented on Agent Bricks today.
[01:01:55] Again, Agent Bricks is our agent developer platform.
[01:01:57] It is uniquely integrated into our data platform, which allows us to provide capabilities that no one else can.
[01:02:04] You saw some of that with our intelligent routing with Unity AI Gateway.
[01:02:08] You'll see it with Data Bricks sandbox when you get to play with that.
[01:02:11] And you'll see it with our contextual security policies as well.
[01:02:15] Thank you for all the things that you have built on Agent Bricks so far and all the feedback that you've given us.
[01:02:20] It's really helped us build an amazing solution here.
[01:02:23] And we can't wait for what you build next.
[01:02:26] Now I'm excited to introduce one of our longest running partners where we're going to hear from Satya Nadella from Microsoft.
[01:02:33] Thank you.
[01:02:38] Satya Nadella: Satya, welcome back.
[01:02:41] I love this.
[01:02:42] We do this every year.
[01:02:43] Our partnership is stronger than ever.
[01:02:45] You know, things are moving so fast since just last year we talked.
[01:02:49] You know, such a big opportunity ahead.
[01:02:51] I'm curious, what have you seen in the market?
[01:02:53] Where are things headed?
[01:02:54] Satya Nadella: First of all, Ali, it's fantastic to be back with you.
[01:02:58] And obviously, thank you for the partnership.
[01:03:00] The two companies have worked together now close to 10 years.
[01:03:03] And here we are both innovating and really helping our mutual customers.
[01:03:09] The thing that I would say, even since we last talked to now, a lot has changed.
[01:03:16] But I would say the core structure of how customers are evaluating what AI can do for them is changing.
[01:03:24] So what I mean by that is, instead of just talking about here is the latest model and its capabilities, we are now going from frontier model worship to actually building a frontier ecosystem, building one company at a time.
[01:03:39] Whether it's an AI-native startup or whether it is an enterprise, that's where it is at.
[01:03:44] And I think that's where even us partnering and enabling every enterprise to be able to fully participate at that frontier with their own IP, I think is going to be key.
[01:03:54] Yeah, that makes a lot of sense.
[01:03:55] I saw your build keynote.
[01:03:57] And in a sense, it's like the future of the firm.
[01:04:00] And, you know, what will it look like?
[01:04:02] What kind of IP will you have?
[01:04:03] And it seems that data is more important than ever.
[01:04:05] So, you know, really protecting that data that each firm has and building IP that you yourself own and protecting that.
[01:04:13] That's how you're going to compete in the future.
[01:04:15] Is that sort of how you're thinking about it?
[01:04:16] That's exactly right.
[01:04:17] In fact, there are two parts about data, right?
[01:04:19] One is the context, which both of us, you know, deeply care about in terms of building the systems, which are the context systems.
[01:04:28] In the past, we said, hey, here is the system of record or system of engagement.
[01:04:32] But now we are creating those systems of context, which, by the way, will make, by definition, even your use of a frontier model, token efficient, right?
[01:04:41] The one way to waste a lot of tokens is to not have the best context.
[01:04:45] And so the work we are doing is all about helping companies at that layer.
[01:04:50] Then the next piece is, of course, you have this massive data asset.
[01:04:54] And it's not a static asset.
[01:04:56] It's an ever changing continuous asset, even including the agent traces.
[01:05:01] So if you can take that data and tune, in fact, like, in fact, one of the things that I love, you were at our CEO Summit and you saw this, is a good 5B model from MAI is able to beat frontier class evals.
[01:05:16] Because why?
[01:05:17] Not because of some magic of ours.
[01:05:19] One, it's a good lineage model that was, you know, good learning model.
[01:05:23] But the magic is the enterprise data.
[01:05:25] When you put it into an RLE at the enterprise level and feed it its traces, it beats, you know, a much more, you know, expensive, big, large frontier model because the value is coming from your data.
[01:05:38] So right now, a lot of people are token maxing, using frontier models with their context and, in fact, leaking the traces.
[01:05:45] I think that this is going to reverse in a hurry.
[01:05:47] Yeah, it makes a lot of sense.
[01:05:49] It's like this for loop.
[01:05:50] It just does this random walk going over your data again and again and again, processing the same thing, spending tokens, taking a lot of time when, you know, that context could have been gathered.
[01:06:00] So, this is, we're super excited.
[01:06:02] So, you know, we're really integrated now with the whole Azure platform.
[01:06:06] So, we have Databricks Genie.
[01:06:08] We integrate it with M365 co-pilot.
[01:06:11] IQ, I think, is really amazing what Microsoft is doing with that.
[01:06:14] Entra, Agent 365, Teams.
[01:06:17] I'm wondering, these deep integrations that we're putting together, how does it change things for customers?
[01:06:23] No, it's fantastic to see all this, right?
[01:06:25] Genie inside a Teams, inside a co-pilot.
[01:06:28] That means I can get to all of that context that you have been able to bring together and right where I'm using it along with, you know, these surfaces.
[01:06:38] The other thing that I love about it is you also expose the MCP server of Genie in co-pilot studio.
[01:06:44] So, it's available for somebody to build their agent more efficiently with your MCP.
[01:06:49] And also, of course, the work that we are doing and even the read/write through one lake and what you are having, that's fantastic.
[01:06:56] And so, this ability to bring our innovation together to essentially power this agentic future where the enterprise feels that they can get the best economics of tokens,
[01:07:06] but more importantly, long-term, they can build their own token capital.
[01:07:10] And this context here, I think, is just fantastic to see.
[01:07:14] So, yeah, zooming out, you know, it's been almost a decade together.
[01:07:18] You know, what will the future look like between our firms and what's companies going to do out there with all this AI?
[01:07:24] What do you foresee in the future, next few years, maybe 10 years as hard to work out?
[01:07:28] Yeah, it's fascinating, right?
[01:07:29] I mean, obviously, we got together in what was, at least at that time, you know, the opportunity for you and for us,
[01:07:36] to sort of rebuild what is cloud-native data tier look like and reimagine that and bring, in some sense, great innovation and great value to customers.
[01:07:46] And now, look, I think the next 10 years for us will be a lot of what we talked about.
[01:07:50] But fundamentally, my thesis, Ali, is that, you know, if a firm is about the tacit knowledge, right, and learning.
[01:08:00] In fact, I always say this, which is you can always buy a tool.
[01:08:03] You can always even, you know, you can even sometimes outsource a job or a task, but you can never, as a firm, outsource your learning.
[01:08:12] So, in this new world, what does it mean to build that tacit knowledge that is compounding both human capital and token capital?
[01:08:23] And so, I think what the two companies are super aligned on in terms of sense of purpose and mission, as well as the technology innovation,
[01:08:31] is to help every firm accrue that value for themselves.
[01:08:35] And today, they have unbelievable human capital that uses digital systems to build that tacit knowledge.
[01:08:42] Tomorrow, they'll have human capital and token capital that is building that tacit knowledge that really drives enterprise value for them.
[01:08:49] And that's what I hope to achieve partnering deeply with you.
[01:08:53] Yeah, that's so well put.
[01:08:54] Very excited about that future, helping these firms actually figure out how they can, the core DNA, what will it look like in the AI era,
[01:09:01] and how can they keep that and actually compete with that in the market.
[01:09:05] So, very excited about this.
[01:09:06] Thank you so much for the partnership, Satya.
[01:09:08] Looking forward to another decade together.
[01:09:10] Absolutely.
[01:09:11] Thank you so much, Ali.
[01:09:14] Welcome to the stage, Databricks Director of Product Management, Justin de Brabent.
[01:09:30] It's a longer walk than it looks.
[01:09:33] Hey everybody, I'm Justin.
[01:09:34] Let's talk about apps.
[01:09:36] So, we launched apps last year with the mission to democratize data and AI.
[01:09:43] And as you've seen over the course of the talks yesterday and today, and certainly in your own experiences, we're no longer constrained by the cost of writing code.
[01:09:52] Coding agents have proliferated.
[01:09:54] Code is cheap.
[01:09:56] But the bottleneck is enterprise data.
[01:10:00] Enterprise apps are not just code.
[01:10:02] They need access to the data.
[01:10:04] And the real challenge in going from prototype to production is governed access to that data.
[01:10:11] And that's why we built the first enterprise app platform with data security and governance at the foundation.
[01:10:19] And that's what Databricks apps is.
[01:10:22] And over the last year, you all have been busy building.
[01:10:29] It's incredible.
[01:10:30] We have over 5,000 customers using apps.
[01:10:33] 150,000 apps have been built.
[01:10:35] That's 6x growth since last year.
[01:10:37] And it's not just the apps are being built.
[01:10:39] They're being used by hundreds of thousands of users on a weekly basis.
[01:10:42] Many of them that weren't using data and AI in their day to day before.
[01:10:47] So the favorite part about my job is that I get to talk to you all.
[01:10:53] And every PM will say they love talking to customers.
[01:10:56] And that's usually true.
[01:10:57] Obviously, there's some conversations that are harder than others.
[01:11:00] Feature requests, support issues, et cetera.
[01:11:03] But with apps, it's a little bit different.
[01:11:05] Because when I go into these customer conversations, I get to hear firsthand what you're building.
[01:11:10] I get to see the excitement of, hey, I built a thing that I didn't know was possible and I want to show it to you.
[01:11:16] And it's really cool.
[01:11:17] So I figured I'd start this talk today by showing you a few examples of some of the awesome things that our customers are building.
[01:11:23] So first off, Thales.
[01:11:25] Thales is a multinational aerospace company building software and avionics for global airlines.
[01:11:32] And Thales reinvented their in-flight intelligence platform, bringing together system health, content engagement, passenger satisfaction across the entire global fleet.
[01:11:44] And with this app, the Thales Ops team is able to anticipate cabin issues before passengers notice, ultimately improving the passenger experience.
[01:11:53] And even though this is delivered through an app, it's powered by the best of the Databricks platform.
[01:12:00] We have LakeBase for real-time data.
[01:12:02] We have analytics in the lake house.
[01:12:04] We have Genie.
[01:12:05] We have machine learning models powering the forecast.
[01:12:08] It's everything brought together in one cohesive app.
[01:12:11] Or finally, Joby.
[01:12:13] Joby is a manufacturer for all electric air vehicles.
[01:12:17] And behind every single air vehicle that they build is an enormous material planning complexity challenge.
[01:12:23] They built an app that provides a single unified view for demand, inventory, parts, and orders across every manufacturing program in their fleet.
[01:12:34] So planners can spot shortages before they become manufacturing problems and eventually delays.
[01:12:41] There's a theme of airlines, aircraft, air industry.
[01:12:45] I want to call out that what I think is cool here is these are all very, very different applications.
[01:12:51] Right?
[01:12:52] Different teams, different applications.
[01:12:53] And this is what you see across all of our customers is everybody is really innovating and doing different things across all their different industries.
[01:13:03] So how does the platform deliver that?
[01:13:07] We have three unique buckets of capabilities.
[01:13:09] First, we help you govern.
[01:13:11] We provide governed access to enterprise data.
[01:13:14] Host, we provide infrastructure for every application across the enterprise.
[01:13:19] And finally, build.
[01:13:21] We provide you tools and integrations to do AI first development and build production ready apps.
[01:13:27] Let's double click into each.
[01:13:29] So first, govern.
[01:13:31] At the core of Databricks governance for apps is our managed auth offering.
[01:13:35] We have out of the box identity through Databricks SSO.
[01:13:38] You don't have to have a separate IDP.
[01:13:40] You don't have to have a separate login for each individual app.
[01:13:43] It's all managed through Databricks.
[01:13:45] Governance comes out of the box with Unity Catalog.
[01:13:49] You can use Unity Catalog to manage resources and data that the app has access to either at the app level or the individual app user level.
[01:13:58] So there's a lot of power and configurability.
[01:14:03] And finally, we now support real-time observability out of the box.
[01:14:07] So as a platform team, you get real-time insights not only into who's using the app, but what they're doing, what services they're accessing.
[01:14:16] And this is all ingested into ZeroBus in real-time and made available to Unity managed tables.
[01:14:27] Host.
[01:14:30] Databricks provides managed containers and infra, secure cloud-agnostic deployments across our stack.
[01:14:37] When we launched last year, we only had a single always-on instance type for apps.
[01:14:41] There's one instance.
[01:14:43] Now, we have configurable instances and horizontal auto-scaling.
[01:14:47] We also do load balancing and sticky sessions across these multi-instance apps.
[01:14:52] So no app is too big.
[01:14:55] And finally, we help you iterate safely.
[01:14:58] We now have version builds and zero downtime deployments to ensure risk-free releases.
[01:15:05] And finally, build.
[01:15:07] Apps is framework agnostic.
[01:15:09] You can use Python or JavaScript across a range of the most popular app frameworks.
[01:15:15] You can build production apps faster with AppKit.
[01:15:18] This is our TypeScript apps SDK.
[01:15:20] It helps with pre-built plugins for building apps using Lakebase, Genie, Agents, Jobs, Lakehouse, etc.
[01:15:26] It really makes the hard parts of app building super easy.
[01:15:31] And finally, it works with coding agents.
[01:15:33] We've invested a ton with our partners to build custom skills, MCPs, and plugins to make it easier for coding agents to build and deploy Databricks apps.
[01:15:41] And of course, we're not doing this alone.
[01:15:44] We're powered by a very strong partner ecosystem, the best AI coding partners out there.
[01:15:48] So it's exciting, but this is still only the beginning.
[01:15:54] One billion.
[01:15:55] That's how many new custom enterprise apps are forecasted by 2028.
[01:16:03] One billion.
[01:16:04] Process that.
[01:16:05] It's huge.
[01:16:06] And the really interesting aspect about this is that most of them are projected to be built by citizen developers.
[01:16:14] So many of the apps that I just described and the challenges that we addressed were for central teams building these applications.
[01:16:20] Teams with access to Databricks resources and the right access to data.
[01:16:25] It's a completely different set of challenges as we try to democratize innovation of who can build these apps to these citizen developers.
[01:16:32] So to highlight that, I want to share a quick story.
[01:16:35] We'll call this the prototype to production gap.
[01:16:39] This is a real story of a sales ops manager here at Databricks.
[01:16:43] This person wanted to build an app to streamline their team's QBR process.
[01:16:48] So as part of that app, they wanted to pull in customer consumption, recent conversations from Salesforce, forecast in our model, and bring it all into one cohesive app.
[01:17:03] And she built a prototype.
[01:17:04] She used Replit to Vibecode up a prototype.
[01:17:06] Was super excited about it.
[01:17:07] Showed it with her leadership.
[01:17:09] Thought all the hard work was done.
[01:17:11] It was not.
[01:17:12] She went to then to our platform engineering team.
[01:17:15] And immediately our platform engineering team said, okay, what data do you need?
[01:17:20] What's the sharing policies, dependencies, code, cost?
[01:17:24] And it ended up being a lot of back and forth to get the app actually deployed to production.
[01:17:30] Because the sales ops manager obviously didn't have access to our production environments with all of our customer information.
[01:17:35] And this isn't about just happening once.
[01:17:38] This is happening thousands of times across your organization.
[01:17:41] And I guarantee some of you on the platform team have experienced this.
[01:17:44] Vibecoding is taking over.
[01:17:46] And unless we have a governed solution to it, it's going to overrun us.
[01:17:50] So we spent the last couple quarters going out and talking to you all and asking this question.
[01:17:58] How do we unlock the next wave of builders without sacrificing security, governance, or budgets?
[01:18:05] And this is what you told us.
[01:18:08] First off, we need governance and ops that scale to thousands of apps.
[01:18:14] The governance model of apps today is largely a single app basis.
[01:18:17] How do we scale to many thousands of apps built by hundreds of developers?
[01:18:22] And today we're excited to announce a new governance construct called app spaces.
[01:18:27] App space has helped you govern and operate thousands of apps through multi-app policies.
[01:18:32] You can apply governance and security policies at all apps that are deployed in that space.
[01:18:36] As an admin, you can restrict who the apps in that space can be shared with.
[01:18:42] So in my sales ops example, maybe those apps can only be shared with the sales team.
[01:18:46] And finally, it provides scalable operations.
[01:18:50] How do you track app usage, cost, and other operational metrics for all apps that are deployed in that space?
[01:18:59] The next challenge that we heard is we need more cost-effective infra for lightweight apps.
[01:19:04] This idea of apps infrastructure being always on and now auto-scaling is great for big mission-critical apps, but it's not great when you have these lightweight or low-use apps.
[01:19:16] You know, maybe it's an ops dashboard that is used a couple times a month.
[01:19:20] So we need more cost-effective solutions there.
[01:19:22] And we're excited to announce a completely new apps architecture called serverless micro-apps.
[01:19:27] Yeah, cost.
[01:19:32] I love it.
[01:19:34] So serverless micro-apps helps you run lightweight apps effectively.
[01:19:40] It does it through fractional CPU and memory with dynamic scale to zero, which many of you have been asking for.
[01:19:49] It has fast start-ups, so it starts back up when the app is used on the order of seconds.
[01:19:54] And even though this is shared infra, it's still with full isolation.
[01:19:58] So we're providing kernel-level isolation using micro-VMs.
[01:20:06] And finally, the last thing that we heard, the last major challenge, is that we need Vibe-coding tools that understand your business.
[01:20:14] Not just your data, but the whole context of where you're deploying these apps.
[01:20:18] The resources, the environment, the sharing policies, et cetera.
[01:20:22] And that's why we are super excited to announce the release of Genie App Builder.
[01:20:32] Genie App Builder is Vibe-coding that understands your business.
[01:20:36] It's context-aware.
[01:20:38] It leverages Genie Ontology to Vibe-code with a deep semantic understanding of your business and your data.
[01:20:43] It's embedded.
[01:20:44] It's 100% Databricks managed.
[01:20:47] So it provides the strongest security and governance guarantees.
[01:20:51] And finally, it's production quality by default.
[01:20:55] It uses the AppKit SDK and custom skills to ensure that high-quality apps are out of the box.
[01:21:01] So all the best practices are baked into all the apps that Genie App Builder creates.
[01:21:07] OK.
[01:21:08] Let's see all this in a quick demo.
[01:21:12] So for this demo, I'm going to be a Supply Ops Manager for the Brickster Cup.
[01:21:20] And what I've done is I've created an app that's going to be my inventory command center.
[01:21:26] So it's going to help me track inventory levels across various sites, demand, forecast, how that's changing as certain teams advance or not.
[01:21:36] And then it's going to let me actually take action directly from the app to adjust my inventory levels, trigger new production runs, etc.
[01:21:42] And what I'm going to do is I'm actually going to start from a prototype that I Vibecoded using one of our Vibecoding partners.
[01:21:50] And I'm going to use Genie App Builder to integrate this into my production environment, connect it to live data, and ultimately deploy it.
[01:21:55] OK.
[01:21:56] So let's get going.
[01:22:01] OK.
[01:22:01] So here we are in the Genie App Builder homepage.
[01:22:04] You can see the kind of prompt-based text up here.
[01:22:07] Here we have a set of templates.
[01:22:08] So these are per app space.
[01:22:10] You can configure custom app templates.
[01:22:13] These are configured by your admins.
[01:22:16] If we go up here and look into the app space, we can see the list of app spaces that I have available to me.
[01:22:21] The amount of apps that are deployed in each.
[01:22:23] I can go into my Supply Chain Ops app space and see the resources that are allocated to this space.
[01:22:28] So what data, what compute, what OBO credentials and scopes I can use, how these apps can be shared, any sort of external connectors.
[01:22:39] So it's all the governance settings for this app space.
[01:22:42] I'm going to go back out and we're going to build an app.
[01:22:46] And we're going to build it in this Supply Chain Ops space.
[01:22:49] So now as an app builder, I can select the space I'm trying to build in.
[01:22:53] And I have all the resources available to me in that space.
[01:22:56] Now, like I said before, we're actually going to start from an existing prototype that I vibe coded.
[01:23:01] This is available in this GitHub repo here.
[01:23:03] So I'm just going to import this.
[01:23:05] It will attach this as context into Genie App Builder.
[01:23:08] I could also add additional context.
[01:23:12] So down here, if I had documents or screenshots of prototypes, I could also attach those.
[01:23:17] For simplicity, I'm just going to use this GitHub prototype and say let's build an app based on this.
[01:23:23] OK?
[01:23:24] So now we're going to kick off the Builder UI.
[01:23:29] So as this populates, you can see the prototype app that I have over here on the right.
[01:23:36] There's no live data attached to it.
[01:23:38] It's just kind of the layout.
[01:23:40] It has my inventory, recommended actions, demand across my different sites.
[01:23:46] And I'm going to go ahead and say let's connect this to my Databricks environment.
[01:23:49] So now what Genie App Builder is doing is it's going through all the resources that I have available to me in the app space.
[01:23:55] And it's attaching those to the various components of the app.
[01:23:59] So I found the SQL warehouse, the serving endpoint that I need for my inventory forecast, lake-based instance for the real-time inventory data.
[01:24:06] It's also looking at the data that I need.
[01:24:09] So I now have my warehouse inventory.
[01:24:11] And you can see it's actually, over here on the right, it's starting to propagate this data in real-time.
[01:24:16] So now the app preview is being propagated with the real data.
[01:24:23] Over here I have a Genie agent, and now it's gone into a set of default questions for this Genie agent.
[01:24:30] And over here you can see I have a slight issue.
[01:24:33] So this is my production manufacturing data.
[01:24:36] It looks like I don't have access to this in this space.
[01:24:39] So I can go over here and actually request access to the admin.
[01:24:42] Now this isn't going to happen immediately.
[01:24:44] The admin needs to approve this.
[01:24:46] And you can see that I need this data for this chart here.
[01:24:49] So what I'm going to do is I'm going to remove this manufacturing data.
[01:24:54] And what it should do-- there we go.
[01:24:56] What it should do is now it's removed the graph and the manufacturing data from this.
[01:25:01] And I can go up here, and we also have this preview capability.
[01:25:04] So I can actually see the current and the proposed preview.
[01:25:09] And what's actually happening under the scenes is we're actually using these serverless micro VMs to spin up multiple instances of the app and visually preview them side by side.
[01:25:17] I'm going to go ahead.
[01:25:18] This looks fine.
[01:25:19] I'm going to approve the changes.
[01:25:20] I don't need this data for the V1 of the app.
[01:25:22] I can also-- it's not just about the integrating the data.
[01:25:26] I can also go up here and make simple visual changes.
[01:25:30] So here, let's just say-- let's turn this graph green.
[01:25:36] There we go.
[01:25:37] Should update.
[01:25:38] All right.
[01:25:39] So now the graph is green.
[01:25:41] And then what I can do is let's test the boundaries of the app space.
[01:25:46] So what I'll do is I'm going to ask Genie App Builder to send me a Slack summary based on the recommendations from this agent.
[01:25:55] So let's see.
[01:25:57] Custom agent.
[01:25:58] And I essentially want a daily summary here.
[01:26:01] Now, in order to do this, I need the Slack connector configured at the space.
[01:26:06] And unfortunately, it's not available.
[01:26:08] So this is Genie App Builder respecting the governance boundaries of the app space that I've set up.
[01:26:14] Same thing.
[01:26:15] Let's check the sharing policies.
[01:26:17] So let's say we can share this app with everybody at my company.
[01:26:24] Probably not the best thing to do.
[01:26:26] And yes, this app space only allows sharing apps with the supply ops group.
[01:26:31] So Genie App Builder is enforcing the governance boundaries of the space that it's being deployed in.
[01:26:36] I can also go up here and see the inventory that is attached to this app.
[01:26:40] So what the app is actually using.
[01:26:42] And also what else is available in this app space to me that the app is not using.
[01:26:46] And you can swap out these as necessary.
[01:26:49] And then I can quickly go up here and press Deploy.
[01:26:53] And now you can see this app is deploying.
[01:26:56] And I now have this app deployed in production.
[01:27:00] And it spun it fast on the order of seconds.
[01:27:02] So that's it.
[01:27:04] That's how we go from -- thank you.
[01:27:10] That's how we go from prototype to production using Genie App Builder.
[01:27:18] All of what I just showed is available today.
[01:27:21] So talk to your account teams for how to get access.
[01:27:27] The other thing that if you want to get started to check out is we have a new developer hub.
[01:27:32] Developers.databricks.com.
[01:27:34] Check it out.
[01:27:35] There's prototypes, prompts to get going.
[01:27:37] We have a bunch of sample apps, how-to guides.
[01:27:40] It's really going to help you kind of get going and get the wheels turning with apps.
[01:27:43] So hopefully I inspired you, showed you some customers, building some really cool things.
[01:27:48] On the note of customers building cool things, let's take a look at a video from one of our
[01:27:53] customers at the Virtue Foundation and what they're building with Genie.
[01:28:03] The Virtue Foundation is an NGO.
[01:28:05] The mission is to deliver clinical care to those in need in low and middle income countries.
[01:28:12] Genie has allowed us to build a custom agent that's specific to us.
[01:28:17] The agent has the skills and the tools that utilizes your own proprietary data.
[01:28:22] And fundamentally, you begin to see insights that you couldn't see before.
[01:28:27] Built off my clinical experience.
[01:28:29] All members of our team have traveled and worked for years.
[01:28:33] And this becomes data we can all action upon because it becomes a source of truth.
[01:28:40] What I am most excited about is the maturation of this agentic capability.
[01:28:45] Ask a question, give me an answer.
[01:28:46] How many hospitals are in Ghana?
[01:28:48] There's my answer.
[01:28:49] How many hospitals do cardiology in Ghana?
[01:28:51] There's my answer.
[01:28:52] Using Genie has allowed us to take what would have taken a research assistant months.
[01:28:59] We are now able to get answers to questions in seconds.
[01:29:03] The most important part of this is that this is connected back to real human lives.
[01:29:09] The fact that we're solving real problems for real people's lives is just a humbling experience.
[01:29:18] Welcome to the stage, Databricks director of product management, Mike Del Balso.
[01:29:34] All right.
[01:29:35] Justin was right.
[01:29:36] This is a deceptively long walk.
[01:29:39] Hey, everyone.
[01:29:40] So my whole career I've actually been working in production machine learning.
[01:29:44] I started out at Google working on the teams, the ads machine learning teams that actually kind of kicked off the whole MLOps space.
[01:29:53] Then I helped start the machine learning team at Uber and helped build and lead the ML platform there called Michelangelo.
[01:29:59] And for the past few years, I've been running my own production machine learning startup, which recently joined Databricks.
[01:30:06] And we've been working on those same problems here at Databricks.
[01:30:09] And the whole time, my whole career, the whole machine learning industry has had one goal, the single dream, which is to make machine learning as easy and automated as possible.
[01:30:21] So we can bring intelligent decisions to all of the important places, all the important decisions across our business.
[01:30:27] And many companies, many have tried and many have failed.
[01:30:32] And I'm really excited to talk to you today because that's finally changing.
[01:30:36] So this diagram here, many of you may recognize this.
[01:30:40] This is from a paper that my team at Google published something like 12 years ago.
[01:30:46] And the point of this paper is to really explain how hard it is to do production machine learning.
[01:30:50] You know, this diagram shows, if you look at the red, that's the machine learning code.
[01:30:54] And everything around it is the point of the diagram.
[01:30:56] All these other things around it are all these other systems you practically need to build and run and maintain to run true ML systems in production.
[01:31:05] It's a really hard problem.
[01:31:06] And as a hint, this is all a data problem under the hood.
[01:31:12] And why are we talking about this right now?
[01:31:13] Well, we're building more software than ever.
[01:31:17] And you may have thought that Gen AI would reduce the demand in machine learning, but it's actually only accelerated it.
[01:31:24] With all of the new software and all the new product experiences we're building, we have more products to recommend, we have more prices to offer, we have more fraud to detect, more loans applications to decide on.
[01:31:39] We have a lot more decisions to make in the company.
[01:31:42] And the problem is that it's actually really hard to do this.
[01:31:46] And so we only use, we ration machine learning, we only use it in a handful of places, right?
[01:31:50] Think in your company, how many models you have in production compared to all of the other decisions that you make in the company that could be automated.
[01:31:58] And because we're not using intelligent machine learning there, we're leaving money on the table, right?
[01:32:04] We're using, we're basically making dumber decisions.
[01:32:06] We're using heuristics and rules of thumb and statically coded logic.
[01:32:10] And what we really want to do is automate machine learning so we can make it really easy to scale to all the important decisions we have in the company and keep up with all the scale of the new software, all the new decisions we're going to add as we deploy all this new software.
[01:32:26] And we just, just think back to that diagram I showed you, it's a really hard problem technically, but also, you know, to automate this, you know, there's people in the loop in every single stage.
[01:32:38] There's a person making, there's an expert, there's a data scientist, someone with a PhD making a decision about which feature to build, which model looks good, what are we really optimizing for?
[01:32:48] Which model should we promote to production?
[01:32:50] So if we're going to automate machine learning, we need to do a lot to accelerate these folks and enable the automation to make decisions in the same way they do.
[01:33:00] And there's actually only one technology that can do this, and that's agents.
[01:33:05] Now, agents have done an immense amount to completely revolutionize how we build software, all of software engineering.
[01:33:15] But they haven't had that same impact with machine learning yet.
[01:33:18] We're still kind of building and operating machine learning in the same way.
[01:33:22] Now, why is that?
[01:33:24] Well, there's two reasons.
[01:33:26] One, is only a handful of companies even have, like, the right infrastructure.
[01:33:31] It was a really hard problem.
[01:33:32] You saw the diagram of all the stuff you've got to build to run this complete machine learning loop in an automated way.
[01:33:40] The second problem is that if we want agents or software to automate and make the same decisions that our machine learning engineers are making,
[01:33:51] they need to have all the same type of information so they can make the same quality of decisions.
[01:33:56] We can really automate that.
[01:33:57] And that context is just not available today to our agents.
[01:34:00] And I'm really excited because Databricks is the first company that can solve this.
[01:34:04] And we're doing this in two ways.
[01:34:06] The first is with agent-ready machine learning infrastructure.
[01:34:10] So getting that infrastructure into a place that your agents can actually drive throughout that whole loop.
[01:34:16] Second, of course, we need the agents that can drive the loop with your context to do it in the same way that your teams would do.
[01:34:25] So let's look at both of these, right?
[01:34:28] Agent-ready machine learning infrastructure.
[01:34:30] If you're here, you're probably building your ML models on Databricks.
[01:34:35] Databricks has been at this for more than a decade now.
[01:34:38] More than 500 million models have been trained on Databricks in just the past year alone.
[01:34:43] And so Databricks already has a complete leading platform for machine learning.
[01:34:52] And that's why Databricks is recognized as the leader in data science and ML.
[01:34:57] Databricks is the dot and the highest up, most top, and most to the right.
[01:35:03] And so that's why you're probably building your ML models on Databricks today.
[01:35:08] So what's the problem?
[01:35:10] Well, the problem is that we built all of this and we solved all of these problems for humans to build an ML.
[01:35:19] But now we're going to have agents do it for us.
[01:35:22] And that's what we've spent the past 18 months optimizing for at Databricks.
[01:35:28] Databricks, we made the stack agent-ready.
[01:35:32] So what do I mean by that?
[01:35:34] First thing, we made everything serverless.
[01:35:36] We don't want agents to have to configure clusters and maintain cluster state and stuff like that.
[01:35:41] Everything is serverless out of the box.
[01:35:43] Easy for an agent to spin up and spin down and scale to whatever scale they need.
[01:35:50] Secondly, this whole data loop, this whole machine learning lifecycle,
[01:35:54] is now instrumented with a single data model.
[01:35:59] So your training data, your features, your serving data, your monitoring data, your inference,
[01:36:05] they all land in the same governed tables.
[01:36:08] That allows your agents to immediately see the impact of any changes they make
[01:36:14] and have a single source of truth for the entire machine learning lifecycle,
[01:36:19] making it much easier for them to optimize and drive this loop.
[01:36:23] And thirdly, everything is governed by default.
[01:36:27] So you probably already know that we govern all the machine learning assets.
[01:36:31] So your models, your experiments, your serving endpoints,
[01:36:35] all of that's already governed in Unity Catalog today.
[01:36:38] But we now have governance on the actions that your agents can take so that you can trust any changes
[01:36:48] that your agents are making to automate this machine learning loop.
[01:36:53] And I'm really excited because we're extending this further today.
[01:36:57] We are launching a new compute type in Databricks that makes it really easy to build deep learning models and LLMs.
[01:37:08] And that is called AI Runtime.
[01:37:11] This is serverless GPUs available on demand at whatever scale you need for no upfront commit.
[01:37:18] You can use this to build deep learning models and LLMs.
[01:37:24] And, you know, we've actually had this in preview for the past two months and we've seen huge adoption from it.
[01:37:29] There's hundreds of companies that have already moved their deep learning workloads to this platform
[01:37:34] and are finding huge success.
[01:37:36] Yesterday, Ali mentioned Merck's latest drug discovery models
[01:37:42] and we've been really fortunate to partner with them and help them build on this platform.
[01:37:46] So I want to announce one more thing, which is we're expanding it today to support multi-node training.
[01:37:52] So now you can build these models at any scale you need on any size of data.
[01:37:57] We're really excited about this because this brings a whole new domain of machine learning within the Databricks platform.
[01:38:03] And there's no reason that you shouldn't be bringing this to your deep learning teams today.
[01:38:11] Thank you.
[01:38:14] Okay, so that was the infrastructure.
[01:38:16] We've been talking about infrastructure for a long time, right?
[01:38:19] Let's talk about the agents to actually drive this loop, right?
[01:38:21] This is getting to a more interesting thing here.
[01:38:24] Agents that drive the loop, it's important for them to do it with your context in the same way that your team would actually build machine learning, right?
[01:38:31] So let's talk about the first part here.
[01:38:33] It's building.
[01:38:35] So this is building the models, evaluating them, etc.
[01:38:39] Extremely excited also to announce Genie code for machine learning.
[01:38:42] This is an enhanced set of capabilities and intelligence for real-world machine learning tasks built into Genie code, and it's available today.
[01:38:52] Nobody has seen more machine learning in the real world than Databricks.
[01:39:00] So we've taken all of those lessons, what works, what doesn't work, what good looks like in production, and we've built that into Genie code.
[01:39:08] Genie code is also natively integrated with the complete machine learning stack.
[01:39:14] So it knows how to use all of those components, all that infrastructure I just showed, better than any other coding agent.
[01:39:22] And what I'm most excited about is we've upgraded the context in Genie code.
[01:39:27] So it now knows how your machine learning teams and how your data science teams actually work, so it can make recommendations and do work alongside your teams.
[01:39:37] And I'll show you exactly what I mean by that.
[01:39:41] So you remember this diagram.
[01:39:43] This diagram is a little bit deceptive because, as I mentioned, this is from like a decade ago, and that was the problem we were thinking about at that time.
[01:39:49] That was solving the human problem.
[01:39:51] But when people are making decisions and driving that whole loop, they're actually drawing on many other sources of information to make these decisions, right?
[01:40:01] What experiments do we run already?
[01:40:04] What does this business metric mean?
[01:40:07] What worked?
[01:40:08] What didn't work?
[01:40:09] Where do we keep that evaluation script?
[01:40:11] And all this context is scattered across our whole platform.
[01:40:16] And if you don't have access to this data, if your agent doesn't have access to this data, it's going to use the wrong data.
[01:40:24] It's going to use the wrong evaluation.
[01:40:26] And it's going to tell you something that kind of looks right, but it won't be right.
[01:40:31] And you won't be able to trust it.
[01:40:34] So what we've done, yesterday Ken announced Genie Ontology.
[01:40:39] We've built all of this into, integrated it with Genie Ontology and empowered Genie Code with Ontology.
[01:40:46] So all of that context is built into Genie Code and this allows Genie Code to know how your machine learning teams work and how they build models, evaluate models.
[01:40:56] Now let me show you specifically what I mean, right?
[01:41:00] This allows Genie Code to do things like building features in the same way that your team builds features today, using the exact same patterns.
[01:41:11] It allows Genie Code to reuse the training scripts and understand what are our training workflows?
[01:41:16] How do we actually build models?
[01:41:18] What's our existing pipeline?
[01:41:20] And reuse that so it actually fits in with the workflows that your team already has.
[01:41:26] But it's not just about building.
[01:41:28] It's also about bringing these models to production and figuring out what are the better models.
[01:41:33] And so Genie Code also respects the evals you have.
[01:41:37] It finds the eval scripts that you have and it reuses those.
[01:41:40] So when Genie Code says a model is good, you can actually trust it because you know it's holding a model to the same bar that your team already trusts.
[01:41:51] And that allows Genie Code to really act like a senior machine learning engineer on your team.
[01:41:58] We found that Genie Code performs way better in real world data science and machine learning tasks than other leading coding agents now.
[01:42:08] And this allows companies like Danfoss, a multi-national engineering company, to go from no machine learning to a complete production machine learning system in just 90 minutes.
[01:42:23] And this is a type of project that would take weeks, months, and I know for a lot of companies, a lot of customers in this room, sometimes it takes quarters, right?
[01:42:33] It's really hard to do this.
[01:42:35] This is a gigantic speed up.
[01:42:38] And so we're really excited about that.
[01:42:39] And that's just the build phase here.
[01:42:43] But the bad news is that the build phase, if you talk to a real world ML team, building is not the majority of their job.
[01:42:53] They're actually spending 60, 70, 80% of their time on maintaining their machine learning applications.
[01:43:00] It's just part of the problem.
[01:43:02] And if we're building models 100 times faster, we're going to be building 100 times more models.
[01:43:08] And so we can't take on 100 times the maintenance burden.
[01:43:12] No team can scale to that.
[01:43:14] And so that's why we need to think about the operating side of this loop as well, right?
[01:43:19] After a model's in production, how do we keep it healthy?
[01:43:23] And you know, anything can go wrong with machine learning models.
[01:43:27] Data can drift.
[01:43:28] Data can drift.
[01:43:30] The model server, the inference server, can have latency spikes.
[01:43:33] There can be any number of reasons for errors.
[01:43:36] Upstream data pipelines can break.
[01:43:39] There's a long tail of a thousand issues.
[01:43:42] And to solve this and automate this, something I'm really excited about, and I think it's the biggest launch in machine learning in the past 10 years.
[01:43:52] I'm very excited to introduce Genie's zero ops for machine learning.
[01:43:55] This is an agent that runs your machine learning systems for you.
[01:43:59] So there's a really big paradigm shift with this.
[01:44:02] Instead of typical monitoring systems that, you know, are looking at your metrics, and when they find an issue, they bring you a problem.
[01:44:10] They say, hey, this thing looks wrong.
[01:44:13] Zero ops for ML doesn't stop there.
[01:44:16] It finds the problems, and then it takes it upon itself to discover the root cause, come up with a fix autonomously, and bring you the solution.
[01:44:25] So instead of getting problems given to you all the time, you're getting solutions brought to you.
[01:44:30] And so it monitors every model continuously.
[01:44:33] It investigates these issues and resolves them autonomously, but it still leaves your team in the loop to approve before anything deploys.
[01:44:41] And the flip here is that for the whole history of machine learning, we've always had a team member on-call for our models.
[01:44:49] And zero ops for ML inverts that and allows an agent to be on-call, and your team is just in a position to approve the recommendations from that agent.
[01:44:59] And that allows Genie's zero ops for ML to solve all of these kinds of problems we just talked about, while still leaving your team in the loop just for the final approval.
[01:45:09] And so rather than me showing you a bunch of more slides about this, I'd like to invite Amber on stage to actually show us how it works live.
[01:45:17] Amber, thank you.
[01:45:29] Thanks, Mike, for setting the stage.
[01:45:32] And hello, everyone.
[01:45:33] My name's Amber, and I'm here to show you how Genie Zero Ops not only finds issues with your ML systems, but fixes them automatically.
[01:45:41] At Bricksports Global, my team and I have a model in production that forecasts ticket sales for every game of the Data and AI World Cup.
[01:45:49] And this model is mission critical because that forecast affects tens of millions of dollars in event staffing and operation spend.
[01:45:57] And we don't want to end up with a lot of empty seats because we overpriced tickets.
[01:46:01] With this model, I need to be alerted as soon as any issues happen.
[01:46:05] So I'm going to quickly use Genie Code to help create some alerts.
[01:46:11] I'm going to create an alert.
[01:46:15] And I'm going to ask Genie Code to create an alert that will trigger if observed ticket sales are not within plus or minus 5% of the forecasted sales for any game.
[01:46:24] Here we can see how Genie Code builds and registers the agent automatically.
[01:46:30] No extra setup needed.
[01:46:32] All right.
[01:46:33] I'll allow this, yeah, alert creation here.
[01:46:41] And the alert's created.
[01:46:42] Now let's fast forward to the end of the week.
[01:46:45] It's Friday at 5:30 p.m.
[01:46:47] I'm coming out of back-to-back meetings.
[01:46:50] I open up my email and crap.
[01:46:53] That alert was fired 30 minutes ago and the ticket forecast has suddenly become less accurate.
[01:46:58] This is a critical issue where every minute could cost my company millions of dollars.
[01:47:03] So I need to just drop everything and debug what's going on.
[01:47:06] But wait.
[01:47:07] There's this new button.
[01:47:08] View investigation in Genie Zero Ops?
[01:47:11] That sounds promising.
[01:47:12] Let's see what it does.
[01:47:14] When I tap it, it opens up the Genie Zero Ops inbox where we can see an open thread with the highest severity.
[01:47:21] So I'm just going to go and click into that in order to see the investigation that Genie Code started automatically about 30 minutes ago.
[01:47:29] So instead of waiting for me to investigate this issue manually, Genie Code got started immediately and has already explored and tested dozens of hypotheses.
[01:47:39] That's amazing for me because these issues are costly and every second counts.
[01:47:43] Let's see what it did.
[01:47:46] First, Genie Code tries to understand what triggered the alert.
[01:47:49] And this is where Genie Code's context and ontology shines.
[01:47:52] Most agents would provide a surface level generic suggestion that still requires me to do a majority of the investigation myself.
[01:48:00] But Genie Code goes deeper.
[01:48:02] It's powered by the ontology of my full ML project, so it analyzes notebooks that built the model.
[01:48:09] It looks at Unity catalog tables.
[01:48:11] It runs SQL queries.
[01:48:13] It's checking for changes in the feature values.
[01:48:17] It's running ML flow experiments and more.
[01:48:20] Overall, it really approached the problem like a real ML expert, continuously iterating until it found the root cause.
[01:48:26] And my favorite thing is all this is done in the background.
[01:48:30] Work that would have taken me hours to do manually, GD code root caused it before I even realized there was an alert.
[01:48:36] And it looks like it figured out the underlying issue.
[01:48:39] Our model was missing a feature that conveys the stage of the tournament.
[01:48:43] This means that the model was unable to account for the increased demand and fanned attention that the knockout rounds receive.
[01:48:49] So as soon as the tournament entered the knockout stage, our forecast became less accurate, as you can see here.
[01:48:56] And then it didn't stop there.
[01:49:00] After figuring out this issue, GD code went even further.
[01:49:04] It started building and testing for the solution.
[01:49:07] It figured out that we had existing data for both the game's tournament stage and matchup.
[01:49:13] And then it created features based on that underutilized data.
[01:49:16] GD code then decided to retrain a model with these features, did hyperparameter tuning, registered the model in MLflow, all while following best practices for ML on Databricks.
[01:49:29] That's awesome.
[01:49:30] And it essentially solved the problem for me automatically.
[01:49:33] And since GD code has full lineage across our ML workflow, it reused the same evaluation criteria from the previous model to confirm the new model's accuracy.
[01:49:43] And it gets our forecast back to where it needs to be.
[01:49:46] Now it's time to roll out this new model to a small amount of traffic.
[01:49:49] And I'm feeling really good because I can trust what GD code did because I saw the evaluations it used.
[01:49:54] Now I can deploy this model in one click.
[01:49:57] Let's do it.
[01:49:58] All right.
[01:50:04] See version two.
[01:50:07] And I'm done.
[01:50:15] Now, normally a stale model like this is a major team-wide incident that massively affects our business for hours, if not days.
[01:50:22] Potentially costing tens of millions of dollars.
[01:50:25] But GD code zero ops resolved it quickly and with minimal disruption to my team.
[01:50:30] And as Mike mentioned earlier, we're entering the 100X era of ML.
[01:50:34] But with 100 times more models comes 100 times the complexity.
[01:50:38] And that's really what we have to protect teams from.
[01:50:41] And that's really the opportunity here.
[01:50:44] Give ML tool -- give ML teams the tools that they need to stop spending all this time fighting operational toil and start building breakthrough AI.
[01:50:53] And that's Genie zero ops for ML.
[01:50:55] Thank you.
[01:50:56] Could you be my one?
[01:50:58] Could you be my one?
[01:50:59] Could you be my one?
[01:51:01] Okay.
[01:51:02] Pretty cool, right?
[01:51:03] I'm going to have to jog here.
[01:51:09] Okay.
[01:51:10] So we saw three things today.
[01:51:11] The first is AI runtime and public preview.
[01:51:14] Serverless GPUs.
[01:51:15] You can build your deep learning models and your LLMs with it.
[01:51:18] Go try it now.
[01:51:20] Second, Genie code for machine learning is generally available.
[01:51:25] It's built into the Genie code you have right now on your laptop, so you can start using it today.
[01:51:30] And third, Genie zero ops for machine learning.
[01:51:33] It's coming this summer, so we're really excited to work with you with that at that time.
[01:51:38] So think back to the dream I started this talk with.
[01:51:42] The machine learning industry has always been about automating machine learning.
[01:51:47] And we're finally at the point where we have the infrastructure and we have the agents to drive the loop.
[01:51:53] And that allows us to build machine learning 100 times faster, 100 times more models.
[01:52:00] And we're entering this era now where we can scale ML to all of the decisions across our business
[01:52:06] so we can have truly personalized customer experiences at every part of our product.
[01:52:12] And we can really optimize every business decision.
[01:52:15] So we're really excited to see what you build.
[01:52:18] Let your agents do the work now so your team can focus on what's next.
[01:52:24] So thank you very much.
[01:52:31] So I'd like to introduce, speaking of what's next, the next speaker.
[01:52:35] He's the GM for a customer lake.
[01:52:38] And he's here to tell us about the future of customer data and marketing.
[01:52:42] Tassos Argyros.
[01:52:48] Good morning.
[01:52:51] How's everyone doing?
[01:52:53] Whoo!
[01:52:56] Okay.
[01:52:57] It's a real pleasure to be here with you today.
[01:53:01] This is actually very, very special for me.
[01:53:04] I'll tell you why.
[01:53:07] I spent the last 20 years building two companies.
[01:53:10] I spent 10 years building a database company.
[01:53:13] But then I really wanted to do something that would bring the power of data to the business.
[01:53:18] So I spent another decade building a marketing technology company.
[01:53:23] One that would combine data but also combine a UI for the business.
[01:53:28] That was my passion.
[01:53:30] And it's really special to be here because in some ways I'm back at my roots.
[01:53:35] But what I'm about to show you is something that truly brings the power of data and agents to the business.
[01:53:42] But before we talk about this, I want to talk to you about customer experience.
[01:53:47] Why do we talk about it?
[01:53:50] The best companies differentiate on customer experience.
[01:53:53] 80% say that they compete primarily on the basis of the experiences they create for their customers.
[01:54:00] And we live in a very interesting time.
[01:54:05] Because AI is dramatically changing how people buy and as a result how those experiences need to be designed.
[01:54:14] Why is that happening?
[01:54:16] First of all, humans do not buy on their own anymore.
[01:54:20] How many of you do you consult with an LLM before you buy anything?
[01:54:27] Be honest.
[01:54:28] Okay.
[01:54:29] I think this is like 80%.
[01:54:32] Very soon, it's not just going to be advice from a chatbot.
[01:54:37] It's going to be delegating agents to do all the research, all the price discovery, all the buying, figure out shipping.
[01:54:44] Everything will happen through agents.
[01:54:46] This means that buying decisions are now happening in seconds or minutes.
[01:54:52] Not days or weeks, which was a buyer journey until now.
[01:54:57] And finally, because commerce is going headless and agents are really good with APIs, agents can be on multiple channels, buying, researching, all at the same time.
[01:55:10] So this creates new challenges, right, to how things were done before.
[01:55:17] Because the way things were done before is very simple.
[01:55:20] Someone in a company, usually marketing, would start with the universe of customers, create segments, we call them audiences, that would have millions of people, and then activate those audiences through campaigns.
[01:55:35] And the way these campaigns would be designed to begin with was through this big waterfall model.
[01:55:42] Right?
[01:55:43] So marketing would do the strategy, data, engineering would put the customer 360 data together, marketing
[01:55:49] ops would create the audiences, creative would do content, marketers would activate measure,
[01:55:54] right, and monitor the campaign as they were running.
[01:55:58] This waterfall worked, but it had some issues, right?
[01:56:02] First of all, it takes a really long time, and it's very rigid.
[01:56:05] It's hard to change dynamically.
[01:56:07] So for all these limitations, though, these waterfall campaigns worked fine up to now.
[01:56:14] But do they work for the era of agentic buying?
[01:56:21] And the answer is no, right, of course not.
[01:56:24] When agents buy on multiple channels at the same time, in seconds, you can't spend weeks designing a single campaign.
[01:56:33] But what if, what if we could personalize every interaction for every customer in real time all the time?
[01:56:43] For their customers and their agents.
[01:56:45] That would be a game changer.
[01:56:48] But to do that, we need to think differently.
[01:56:51] We need to invent a new type of campaign.
[01:56:54] One that doesn't rely on this waterfall model.
[01:56:57] And one that can personalize audiences of one instead of audiences of millions.
[01:57:04] What if we could do this with agents, right?
[01:57:07] So agents create the new challenge on the buyer side.
[01:57:10] What if agents are also the solutions?
[01:57:12] What if we could have an agent for every customer?
[01:57:15] And then what this agent would do is to process the latest signals from every single customer.
[01:57:21] Determine what's the next best action and create a highly personalized content just for that individual customer.
[01:57:28] And from there on, that would go back to the customer.
[01:57:33] The customer would process, transact, and then create more signals that would then feed back to the agent in this infinity-like loop.
[01:57:41] We think this is the future.
[01:57:44] And this is such a new approach to how marketing can do campaigns that it deserved a new name.
[01:57:50] We call it infinity campaigns.
[01:57:51] And we call it infinity, and that's because the symbol looks like an infinity symbol.
[01:58:02] But importantly, these are campaigns that are evergreen.
[01:58:06] They don't need to have a start and an end.
[01:58:09] Because we have agents, the agents continuously adapt.
[01:58:12] So the same campaign can adapt to new customer behaviors, new business goals, and it can keep running.
[01:58:20] Also, infinity campaigns can provide content that's highly personalized.
[01:58:26] It can pick from a multitude, infinitely large number of variations of content for what each individual customer needs at that particular moment.
[01:58:36] And finally, if you employ agents, they can respond very, very fast.
[01:58:41] You don't have any limits on how fast these agents can respond.
[01:58:46] Now, it's important to note that human remains in control.
[01:58:51] The marketer is the chief strategist, the chief orchestrator, and the chief approver of what the agents are doing.
[01:58:59] But now they have an army of agents that can do all the one-to-one personalization work that humans were just too slow to do before.
[01:59:10] Okay.
[01:59:11] So hopefully, I'm convincing you that this new concept of infinity campaign is very cool.
[01:59:18] But the question is, can we make it work?
[01:59:20] How do we make this happen?
[01:59:22] And the way this would have happened in the past was through CDPs.
[01:59:28] And let me talk a little bit about what CDPs are.
[01:59:31] For those of you that don't know, CDPs stands for Customer Data Platform.
[01:59:35] And it's essentially a business application that sits between the data and all the marketing tools,
[01:59:43] the customer tools that actually send the communications to the customers.
[01:59:47] So what do CDPs do?
[01:59:48] First, they can turn raw data into Customer 360, organized data ready for business use.
[01:59:54] Then they have a business UI where marketers can come in and build audiences and journeys.
[02:00:00] And finally, they send data to marketing tools that serve content to the end customer like email tools or advertising platforms.
[02:00:09] And that process is called reverse CTL because the data goes from the CDP into the content or the advertising platform.
[02:00:19] The challenge here is that CDPs were middleware.
[02:00:22] So they were detached from the data.
[02:00:24] And it's not just about data anymore.
[02:00:26] Now, in order to make agents work, you need data.
[02:00:29] But you also need context.
[02:00:31] And you need data and context to be where the agents live.
[02:00:34] So we think you need a new generation of CDPs to make infinity campaigns work.
[02:00:41] And we call this new generation the agentic CDP.
[02:00:45] The agentic CDP is not just some general concept.
[02:00:49] There's three very specific things that a CDP needs to do in order to be agentic.
[02:00:54] First, it has to be embedded in the lake house because that's where the data, the context of the agents increasingly live.
[02:01:03] Second, it has to be built with LLMs at the core of its architecture.
[02:01:08] This is not just about having some co-pilot to help you do some operations better.
[02:01:14] As we discussed before, the agent is the core of how this new type of CDP works.
[02:01:19] And finally, it has to power infinity campaigns.
[02:01:22] It needs to have the capability to have evergreen, real-time, continuously adapting campaigns
[02:01:28] that personalize every individual customer with individual agents.
[02:01:32] So agentic CDP is a new category.
[02:01:36] We think there's going to be several new players there.
[02:01:40] However, we are one of the very first to launch an agentic CDP.
[02:01:46] And we call this Customer Lake.
[02:01:50] Thank you.
[02:01:53] Thank you.
[02:01:55] Customer Lake is an agentic CDP built specifically for the Databricks platform.
[02:02:02] It's embedded, meaning data never has to leave the lake house.
[02:02:07] And all the governance for data and agents that you've done with Unity Catalog will be transferred immediately into Customer Lake.
[02:02:16] It's democratized.
[02:02:17] It's built for the business users.
[02:02:20] So it has a business-first UI.
[02:02:22] And it is autonomous.
[02:02:24] It can empower one-to-one personalization with agents to provide the perfect customer experience a billion times a day.
[02:02:30] Customer Lake comes with two modules.
[02:02:36] The first is Profile Agents, which automates the process of going from raw data to Customer 360.
[02:02:42] The second is Campaign Agents.
[02:02:46] That's the business-first product that allows marketers to power Infinity campaigns with agents.
[02:02:53] Let's double-click and talk about it a little bit more.
[02:02:56] There's a lot of capabilities that we're introducing here.
[02:02:59] And in the remaining of this talk, we're going to talk a little bit more about these two products.
[02:03:04] And also show you how it works.
[02:03:07] Okay.
[02:03:08] Let's start with Profile Agents.
[02:03:10] Profile Agents are agents that automate the process of creating Customer 360 data that is business-ready.
[02:03:16] So with the prompt that defines a use case, Profile Agents get to work.
[02:03:21] They ingest, transform, enrich data, resolve customer identities,
[02:03:26] and evaluate profiles to generate this agentic loop that keeps improving quality.
[02:03:31] The first step is to prepare customer data for use.
[02:03:35] With Lakeflow Connect, we can ingest third-party data from applications or obviously use Lakehouse data that's already there.
[02:03:44] Our agents will semantically tag data and also improve quality such as fix emails, addresses, or phone numbers.
[02:03:52] And we can also bring third-party data in, which we call data hydration, through the Databricks marketplace.
[02:03:59] The second step is one of the key core features of Profile Agents, which is identity resolution.
[02:04:08] What is identity resolution?
[02:04:10] Identity resolution for first-party data, meaning data you already own.
[02:04:14] This is data you have collected from your own customers.
[02:04:18] It's the process of identifying the same customer across different data sets that do not share the same customer key.
[02:04:26] So you have information about the customers.
[02:04:29] You may have names, emails, addresses that don't quite match, but probabilistically you can tell it's the same customer.
[02:04:36] This is very important because if you don't do this, you may end up thinking the same customer as different people,
[02:04:43] and that leads to massive problems for marketing and customer experience.
[02:04:46] So we introduce here a new approach to identity resolution that we call the agentic identity resolution.
[02:04:54] We have agents that come up with rules on how to resolve these identities.
[02:05:00] If these rules are inconclusive, LLMs take over and try to make a determination,
[02:05:05] and if needed, it falls back to human feedback.
[02:05:08] And as more and more identities are being identified and resolved, the agents learn,
[02:05:14] and they enable this agentic loop that make the rules better, the LLM learning better,
[02:05:20] and the whole process continuously improves.
[02:05:23] Historically, this has happened through static SQL rules or machine learning.
[02:05:28] Both approaches have significant limitations,
[02:05:31] and we think the agentic identity resolution wins on all fronts over these traditional approaches.
[02:05:37] And when it comes to third-party data, meaning data that comes from third-party partners,
[02:05:49] we are introducing today a way to resolve identities through those partners
[02:05:55] without your data ever leaving the lake house.
[02:05:59] So this is a privacy-first approach to third-party identity resolution
[02:06:04] that we enable through the Databricks clean rooms.
[02:06:09] And the last step in profile agents, after having ingested, transformed, quality,
[02:06:15] checked the data, and identified the identities,
[02:06:19] is to output the right data products for the right use cases.
[02:06:23] So that involves schema mapping, survivorship, and also data evaluation
[02:06:28] so that the agents can self-improve the quality of the final output as needed.
[02:06:33] So that's profile agents, helping you go from raw data
[02:06:37] to business-ready Customer 360 faster than ever before.
[02:06:42] So let's talk about our second product here, campaign agents.
[02:06:47] Campaign agents are agents that automate campaign work.
[02:06:51] So given a prompt, the agents can create the right campaign brief
[02:06:56] from business context and business goals,
[02:06:59] the marketers review it and approve it,
[02:07:01] and then we deploy agents to do personalization.
[02:07:06] What's cool also is that campaign agents
[02:07:09] also do traditional CTP functionality like audiencing.
[02:07:13] But of course, we do it in a way that's Genie-powered,
[02:07:17] so everything can happen through prompts,
[02:07:19] but we also have a business UI that marketing and marketing operations
[02:07:23] can very quickly review the work of the agents, approve it,
[02:07:27] and with one click you can take audiences and turn them into campaigns.
[02:07:31] Infinity campaigns is the heart of what campaign agents do.
[02:07:36] The beauty of Infinity campaigns, because it's built around LLMs,
[02:07:40] is that it can start personalizing immediately.
[02:07:43] You don't have to spend months building ML models.
[02:07:46] You also don't need to use ML-based decisioning.
[02:07:49] That takes a really long time to start to work.
[02:07:52] It provides exceptional campaign performance out of the box,
[02:07:55] and, of course, batteries are included.
[02:07:59] We have guardrails.
[02:08:01] You can do simulation runs and measurement.
[02:08:03] Everything is built in.
[02:08:05] And for the first time, we're introducing a feature
[02:08:10] that so many of you have been asking.
[02:08:12] Reverse ETL.
[02:08:14] We can now have native managed connectors to multiple products
[02:08:20] in the advertising, marketing, and customer experience ecosystem
[02:08:23] that, with one click, you can make it work.
[02:08:26] We're launching with approximately 20 connectors
[02:08:31] to a rich ecosystem of partners.
[02:08:33] We have a dozen more connectors coming by the end of the year,
[02:08:36] and many more beyond that.
[02:08:40] Okay.
[02:08:41] So, it's time to see Customer Lake in action.
[02:08:45] So, I want to invite Justin to give a demo of Customer Lake.
[02:08:50] Welcome, Justin.
[02:08:58] Thanks, Sasso.
[02:08:59] Encore demo.
[02:09:00] I'm flattered.
[02:09:02] So, in this demo, we're going to give a quick overview
[02:09:06] of campaign agents.
[02:09:08] The assumption here is that you've already used profile agents
[02:09:11] to ingest, model, and resolve your customer data.
[02:09:14] And what I'm going to walk through is using campaign agents
[02:09:17] to build audiences using Genie,
[02:09:19] and then infinity campaigns to decision on a one-to-one basis.
[02:09:22] Okay?
[02:09:23] So, let's jump into it.
[02:09:25] Now, I'm a marketing manager for Bricksport,
[02:09:28] and here I can see my campaigns dashboard.
[02:09:30] This is my overview of my active campaigns,
[02:09:33] some key audiences that I'm tracking,
[02:09:35] my most active reverse ETL syncs,
[02:09:39] kind of my overview dashboard.
[02:09:43] Let's jump over here, though, into the audiences.
[02:09:46] Here's a list of audiences that I've created.
[02:09:48] I can see the size of these audiences, the percent change over time.
[02:09:52] What we're going to do is we're just going to create a new audience.
[02:09:55] And as I do this, I could create this audience using rules and filters,
[02:10:00] but we're going to use Genie.
[02:10:02] And I'm going to copy and paste a prompt in here,
[02:10:05] include our most valuable Team USA fans who have spent more than 500 on merch this season,
[02:10:10] and exclude anyone who already bought the USA fan kit.
[02:10:13] Okay?
[02:10:14] So, let's see what Genie comes up with.
[02:10:16] Genie is going to go and actually generate this audience.
[02:10:18] So, it's going to look at the attributes and the audiences that I already have created.
[02:10:21] And there we go.
[02:10:23] Genie generated this audience.
[02:10:24] So, now I have an audience of 800,000 customers.
[02:10:27] I can see the qualification criteria here.
[02:10:29] If I wanted to, I could go in here and tweak this.
[02:10:32] Right?
[02:10:33] So, this is editable.
[02:10:35] I'm going to trust Genie and go ahead and accept this.
[02:10:38] I come back out here, and now I have some quick insights that Genie generated.
[02:10:43] And we'll come back to that in a second.
[02:10:45] I actually want to show this.
[02:10:47] You can view the sequel that is back in this audience.
[02:10:50] So, this is starting to look probably more familiar to many of you in the audience.
[02:10:54] If you've ever supported marketing by pulling lists and writing custom campaigns,
[02:10:59] you can get really complex SQL to power some of these campaigns.
[02:11:03] And we got this from one prompt with Genie.
[02:11:08] Let's go ahead and save this.
[02:11:09] And I'll get brought out to the main insights dashboard here.
[02:11:13] And I can see kind of insights on this audience.
[02:11:16] These are automatically generated by Genie.
[02:11:18] So, dynamically, every audience is getting its own unique dashboard.
[02:11:21] And then I can immediately use this in a new campaign or even just a new sync.
[02:11:26] So, the point is all these audiences are actionable.
[02:11:28] This is not just insights.
[02:11:29] You can immediately run this in a campaign.
[02:11:31] Now, for the sake of time, I'm going to go over here.
[02:11:34] And I'm going to find a live infinity campaign.
[02:11:38] In this example, it's my in-venue fan experience.
[02:11:42] So, let's quickly look into this running campaign here and see how it's configured.
[02:11:48] All right.
[02:11:49] So, I can see the audiences to include, the goals, conversion metrics,
[02:11:54] any sort of guardrails that I wanted to apply.
[02:11:56] So, this is kind of the constraints that you're putting on the agent.
[02:11:59] I can see the messages that I made available as part of this infinity campaign.
[02:12:03] So, here I have a match day guide and a match day offer.
[02:12:06] And we'll come back to those in a second.
[02:12:08] And the really interesting part is looking at this agents tab.
[02:12:12] So, this is where the agent is actually looking on a one-to-one basis at each individual customer.
[02:12:18] So, you can see a list of sample profiles here.
[02:12:20] Here we have Marcus.
[02:12:22] This agent decided to defer.
[02:12:25] So, not send anything to Marcus right now.
[02:12:27] Let's take a look at why.
[02:12:28] We can actually see the reasoning behind this decision from the agent.
[02:12:33] What was evaluated?
[02:12:34] What was the timing that was considered?
[02:12:36] What sort of data signals were considered as part of this decision?
[02:12:40] Any sort of guardrails that were applied?
[02:12:42] Like I said, frequency caps, et cetera.
[02:12:45] If we go into somebody who got the offer here, we see this person got the match day offer.
[02:12:53] And same thing.
[02:12:55] Jeannie decided to send this personalized offer.
[02:12:59] And this is a preview of what they actually got.
[02:13:01] So, this is a 15% offer off a seed upgrade.
[02:13:06] Let's find somebody who actually got just the match day guide.
[02:13:09] So, here we go.
[02:13:10] If we preview this, we can see this customer got just a summary of the match day guide.
[02:13:14] A summary of the match day kickoff, parking information, et cetera.
[02:13:18] So, Campaign Agents is personalizing every single one of these messages for each individual customer in this campaign continuously.
[02:13:25] I can go out and see an overview of the insights on this campaign.
[02:13:29] I can look at performance by individual messages, opens, clicks, conversions, et cetera.
[02:13:35] To get kind of an aggregate view of how this campaign is doing.
[02:13:38] And with that, I'm back at the homepage.
[02:13:41] That's a quick run through of Customer Lake Campaign Agents.
[02:13:45] The world's first Agentec CDP.
[02:13:47] Thank you.
[02:13:48] Back to you, Tassa.
[02:13:50] Okay.
[02:13:51] Thank you, Justin.
[02:13:52] So, the area of true one-to-one personalization is here.
[02:14:05] It's very, very, very exciting.
[02:14:07] So, in developing Customer Lake, we were very fortunate and honored to work with some of the top innovators globally.
[02:14:19] Some of which were kind enough to let us put their logos up on the screen.
[02:14:24] These are world-class data and marketing teams that are pushing the boundaries of customer experience with data, agents, and marketing strategy.
[02:14:34] Thank you to all our early Customer Lake partners.
[02:14:37] And, similarly, we have been building Customer Lake surrounded by an elite ecosystem of partners that jointly serve our customers in terms of data, content, advertising, activation, and services.
[02:14:55] Thank you to all of our partners that are joining us today for this launch.
[02:14:59] So, with that, I'm excited to say that Customer Lake is in private preview today.
[02:15:05] If you're interested, please talk to your account team.
[02:15:09] Thank you.
[02:15:10] Please talk to your account team if you want to learn more.
[02:15:14] We also have a brand new CDP track in the breakouts.
[02:15:19] And we also have a marketing forum today at 1pm.
[02:15:24] You can find both Search for Customer Lake in your Data and AI Summit app.
[02:15:31] And you will see both the marketing forum and the breakout session.
[02:15:34] That's the place to learn more.
[02:15:36] Some of the breakouts is from the teams, the engineers, the product folks that actually built Customer Lake.
[02:15:41] And we also have some of our early customers talking about their first experiences with Customer Lake as well.
[02:15:48] Make sure to check it out.
[02:15:50] So, with that, thank you for your time.
[02:15:54] It's a very special moment for me to be able to bring together the worlds of business, customer experience, and data, all in one place.
[02:16:03] To build something that provides the perfect customer experience a billion times a day.
[02:16:09] Thank you.
[02:16:11] And now, Andrew Tu, GM of Lakewatch, to talk about Lakewatch.
[02:16:20] Thank you.
[02:16:20] Andrew Tu: Hello, everyone.
[02:16:24] I'm Andrew.
[02:16:25] I'm excited to talk to you today about Lakewatch, which is our new security Lakehouse product.
[02:16:42] We announced this just about two months ago at the RSA conference.
[02:16:46] And this is an area that's really strategic for us.
[02:16:50] You heard Ali share just yesterday the acquisition of Panther Labs, which is joining us to bolster the Lakewatch effort.
[02:16:59] And you'll hear right after me conversation with Panthers CEO Jack, together with our co-founder Reynolds.
[02:17:05] So, you'll learn a lot more about that soon.
[02:17:07] But I want to start with why.
[02:17:09] We've probably all read about the latest LLM models being able to find previously unknown vulnerabilities in just about any piece of software out there today.
[02:17:21] Whether it's open source or closed source, what used to take nation state level capabilities.
[02:17:28] You had to have teams of reverse engineers and hackers working for months or years to find these vulnerabilities.
[02:17:36] Now, anyone with access to the right LLM can find these things instantly.
[02:17:42] And what's more is AI is actually automating the entire attack process, too.
[02:17:49] We're seeing this happening in reality today.
[02:17:53] Anthropic reported recently that hackers were using Claude to attack the Mexican government.
[02:17:59] And you can actually read the entire chat log.
[02:18:01] It's really fascinating.
[02:18:02] You can hear -- you can see the hackers ask the agent to go and search for vulnerabilities in a range of servers.
[02:18:10] It finds some.
[02:18:11] They ask it to attack one of the servers.
[02:18:14] It's able to break in.
[02:18:16] Then they ask the agent to escalate privileges to root access.
[02:18:21] And the agent finds a vulnerability in a cron tab entry and gains root access.
[02:18:27] They ask it to go and search for sensitive data.
[02:18:30] And in the end, the hackers walk away with 200 million personnel records and gigabytes of sensitive data.
[02:18:36] So now, just about anyone with access to an LLM can put on an attack at a scale and speed that was previously unimaginable.
[02:18:48] That's the world we live in.
[02:18:50] And the only way that we're going to be successful in defending against this is if security and data teams work together.
[02:18:58] Let me explain why.
[02:18:59] To start, most companies today have a security operations team that's tasked with defending them.
[02:19:08] And this team does three things.
[02:19:11] First, they ingest a bunch of data from a wide range of data sources.
[02:19:18] Think employee laptops, servers, identity and access systems.
[02:19:25] All of that telemetry data, log data, comes into a central server.
[02:19:30] Then they write detection rules on top of that data.
[02:19:34] So these are queries for potentially malicious behavior.
[02:19:39] So for example, if I were to log in from San Francisco today, and then a couple minutes later, log in halfway around the world,
[02:19:46] that might be a sign that my machine was compromised.
[02:19:49] Or maybe it's just a sign I was debugging.
[02:19:51] So finally, all of those alerts go to an incident response team, which is a team of analysts that are looking through every one of those
[02:20:01] and investigating whether this is truly a threat or just a false positive.
[02:20:06] If it's a real threat, they can take some remediating action like blocking my IP address or forcing me to reset my password.
[02:20:14] So that's what these teams are doing.
[02:20:16] The challenge is the tools they're using are outdated.
[02:20:19] They typically use a system called a SIM, a security information event management tool.
[02:20:25] And these systems, they're not designed for the scale of threat that we're facing today.
[02:20:32] The first problem is that you can't even get all your data in.
[02:20:37] This is a mix of technical limitation and pricing problem.
[02:20:42] So what security teams are doing is they do a number of backflips to reduce the amount of data coming into these SIM tools.
[02:20:50] For example, they'll throw away data that's older than 30 days because they can't afford to store years worth of data.
[02:20:57] Second, they'll run data filtering tools.
[02:21:01] This filters down the incoming data, taking away things that are maybe less valuable.
[02:21:06] Hopefully, you're not missing anything by doing that.
[02:21:09] And then there's an entire class of data that today's solutions can't even ingest and can't consider.
[02:21:16] These are things like LLM logs, code, JIRA, Slack messages.
[02:21:21] This is really valuable context that would help an analyst determine whether Andrew was using a VPN and maybe debugging a service or whether indeed my machine was compromised.
[02:21:35] The second issue is that the process today is incredibly manual.
[02:21:39] If you look at these incident responders, the analysts who have to determine if a particular ticket is malicious or not, they were dealing with over 7,000 alerts a week in 2020.
[02:21:51] That number has gone up to over 30,000 alerts a week for a typical large organization.
[02:21:57] And it might take them 30 minutes to investigate one of those to see, you know, what was Andrew doing before this happened?
[02:22:04] Has he ever logged in from this location before?
[02:22:07] What did he do after?
[02:22:09] And so you multiply that out and already you would need a team of over 400 people to meet the number of alerts you're getting today.
[02:22:19] And this is only going to get harder and harder as we see more agentic attacks happening at larger scale and at faster speeds.
[02:22:31] So the reality is that with the tools we have today, we don't stand a chance in meeting these agentic threats.
[02:22:40] We're dealing with incomplete data because we can't afford and we can't scale to bring it in.
[02:22:45] We're dealing with humans fighting agent attackers.
[02:22:50] And we're dealing with proprietary data silos where you can't analyze all the data because it's locked up in different locations.
[02:22:58] The solution here is a big data platform.
[02:23:02] And that's why we built Lakewatch, the agentic security lake house.
[02:23:08] This is built on top of the Databricks platform.
[02:23:11] So all of your data is stored in open formats.
[02:23:14] We make it very cheap and easy to bring in data in lots of different ways.
[02:23:17] On top of that, you have unity catalog.
[02:23:20] So you can govern access to the data, ensure compliance.
[02:23:23] And then we build in agent bricks and genie on top so you can automate all of these manual processes and workflows.
[02:23:30] And finally, we give you this in a nice UI that security teams are familiar with and can use easily.
[02:23:38] And so the result is that you can leverage all of your data, whether it's structured, unstructured, semi-structured data.
[02:23:45] Anything you need can come into your security lake house.
[02:23:48] And then we can apply agents at every step of the process.
[02:23:52] On the ingest path, we can have the agents automatically apply ETL on that data, do data normalization to make it easier to work with.
[02:24:00] If a schema changes, the agents can automatically correct the ingestion pipeline to ensure you don't have any interruption in your data that's coming in.
[02:24:09] On the detection path, the agents can help author the queries and detection rules for you.
[02:24:16] And on the investigation path, the agents can automatically investigate these incidents so that the analysts are presented with a full case.
[02:24:26] And they can just verify that the agent did the right task and recommended the right action.
[02:24:31] And the result of all of this is that we can meet the scale and speed of threats that are coming in today.
[02:24:42] I'm going to hand it to our tech lead for Lakewatch, Michael Anderson, to give you a demo of the product.
[02:24:58] Good morning, folks.
[02:25:01] So I'm here inside Lakewatch.
[02:25:03] When you have it enabled, speak to your account team, you'll see it show up here in the app switcher.
[02:25:08] And Lakewatch is our agentic security lake house, and it's designed for the entire Sim journey.
[02:25:14] And that begins with data.
[02:25:16] So over here on the data sources side, Databricks is already great at bringing data into your lake house.
[02:25:23] But Lakewatch adds dozens of security-specific connectors that can bring all of your security data, including multimodal data, into the lake house for analysis.
[02:25:33] Once that data lands inside the lake house, we also have tools to normalize that data into the open cybersecurity schema framework, or OCSF.
[02:25:42] In fact, this is a new feature that we've added since our announcement at RSA.
[02:25:46] So today, there's now an entirely automated genie-powered normalization pipeline that will analyze your data and construct this entire normalization pipeline for you.
[02:25:55] And the benefit of having normalized data landing on your gold tables is that you can write your detections once, and do your investigation in a single uniform way, even if you have products from multiple clouds and multiple different vendors that have data that is very heterogeneous.
[02:26:09] Of course, once that data has landed in your lake house, Lakewatch is also designed for scalable detection.
[02:26:18] We support tens of thousands of detection rules running against that data, with features for managing those rules at scale, such as miter tagging and miter coverage analysis, so you can really work out exactly where you need to improve your detection coverage.
[02:26:32] Of course, because I have brought a bunch of data into my lake house, including information about my business and my code and so on, I can actually use Genie to automate the process of creating a detection rule too.
[02:26:44] This page here, when you come here to create a detection rule, this is something that would typically have taken a stock analyst days, sometimes weeks to do.
[02:26:51] You have to go out and research a threat, you have to craft a detection, you have to validate it.
[02:26:55] It's an incredibly time-consuming process.
[02:26:58] But now, I can actually go in here and I can ask Genie, please research and write me -- ooh, this is how you know it's life -- write me a detection for Storm2460.
[02:27:14] Storm2460 is a malware, it's a fairly complex malware that has multiple steps in its attack chain.
[02:27:20] Genie is going to go out and it's going to kick off a background research agent.
[02:27:25] That's going to go and search the web, look for articles and CVEs and analysis about how this particular attack path works.
[02:27:32] And in parallel, it's going to go and look at my data.
[02:27:35] Now, it understands my entire data pipeline.
[02:27:37] It knows I have CrowdStrike connected.
[02:27:39] It knows that's flowing through my normalization pipeline into my process activity OCSF table.
[02:27:44] It's going to read that table, look at the schema, look at the data that I have in there so it knows what it has available to construct this detection.
[02:27:52] And then it's going to go and combine that.
[02:27:55] Now it's busy sampling it to make sure that all of the data really actually is there.
[02:27:59] And the web research has come back.
[02:28:01] So it's going to combine that with the web research and start to form some hypotheses for how this can be detected.
[02:28:07] So building out a detection SQL.
[02:28:09] It's busy checking that I do indeed have the indicators in my data.
[02:28:14] It looks like I have lots of search utility activity.
[02:28:17] Fortunately, it's all legitimate search utility activity.
[02:28:20] So I haven't been breached yet.
[02:28:22] But that's the data that I need to detect this malware.
[02:28:25] And now it's going to write the SQL.
[02:28:27] It's tested it over the past 24 hours.
[02:28:30] No indication of attack.
[02:28:32] Now it's testing it over the last 30 days to make sure that it isn't getting any false positives.
[02:28:36] Because often you may get false positives and it would need to iterate on this.
[02:28:40] Refine the detection SQL to make sure that I have a good trustworthy invalid detection.
[02:28:45] So now it's come back.
[02:28:46] The detection is validated.
[02:28:47] Zero false positives.
[02:28:48] All of the data that this relies upon is present in my environment.
[02:28:52] And it's now going to take that detection SQL and turn that into a rule.
[02:28:57] Which means I've got the metadata that I need to manage this alongside thousands of other rules.
[02:29:02] And it's going to give me a little bit of a report here.
[02:29:04] This is how I'm detecting it.
[02:29:06] It's targeting all seven phases of this malware.
[02:29:09] You can see it's pre-populated my detection SQL for me here on the left.
[02:29:13] And then it's also giving me an overview of the MITRE metadata that it has tagged this rule with.
[02:29:20] So that I can sort of manage this in my corpus of detections.
[02:29:23] So this is a massive acceleration for your SOC team.
[02:29:28] We've taken something that would have been a week and it's now executed in a couple of minutes.
[02:29:32] But the security lake house in Genie doesn't just change how I author my detections.
[02:29:37] It also changes how I investigate and respond to what those detections produce.
[02:29:42] If I go over here into my security cases, you will see that this star here shows that many of these have actually been investigated by Genie in the background.
[02:29:52] So if I look here, this is an alert from my WAF.
[02:29:56] There's been a rate-limiting surge.
[02:29:58] But Genie has filled out an entire report here.
[02:30:00] It's gone and found related cases, both a similar pattern attacking the WAF, but also similar sort of activities done by the same IP address.
[02:30:11] It's gone and looked for these entities in the threat intelligence feed to see if this is a known malicious IP address.
[02:30:18] It's actually gone and looked inside my perimeter as well to see is there any indication that this IP has done something past the WAF?
[02:30:26] Like did it breach in any way?
[02:30:28] And it's given me this whole report so that by the time an analyst opens a case, they don't have to repeat the same runbooks and the same sort of repetitive tasks.
[02:30:37] All of that has been done for them and in fact Genie will learn from how analysts have investigated prior cases so that the next time a similar case comes along, it does the updated analysis on it.
[02:30:48] All an analyst needs to do is focus on the novel aspects of this case and focus on mitigating it.
[02:30:54] I can actually show you this in process.
[02:30:57] If I go back to my cases and I look at informational, you see there's one here that Genie has not investigated yet.
[02:31:02] This is a suspicious API request.
[02:31:05] This is a very low fidelity signal.
[02:31:07] The reality is when a SOC team is getting tens of thousands of alerts a week, they don't have time to investigate something like this where it's just a statistical differentiation from an established baseline.
[02:31:20] But with Genie, we now can.
[02:31:22] I can go in here.
[02:31:23] I can ask Genie to investigate this case.
[02:31:26] It's going to come in and pull up the event here and look through the event details and start to understand what's going on.
[02:31:32] Looks like it's a post request.
[02:31:36] The reason it's suspicious is this read context parameter deviates from the established baseline.
[02:31:42] And it's now actually going to find the service that provides this endpoint and look through my source code and try and understand what is the significance of this new pattern of request.
[02:31:54] In parallel, it's also going to check my thread feed to see this IP address.
[02:31:58] Is this known to be malicious or part of a kind of a known botnet or something like that.
[02:32:01] It doesn't find anything on that side.
[02:32:03] So it's going to keep looking through the code and understand what the significance of this is.
[02:32:09] Looks like it's found something here in the code.
[02:32:13] This read context parameter has a very specific pattern.
[02:32:18] And when that pattern is present, there's actually a hard coded back door in the code.
[02:32:22] It skips all of the access control checks and directly returns the decryption key.
[02:32:26] So this is, you know, this is an amazing finding.
[02:32:29] It found this very rapidly.
[02:32:30] It's actually going to extend its investigation now and try and work out who introduced this into the code.
[02:32:35] When was it introduced?
[02:32:36] Does it exist in more places?
[02:32:38] Or has this employee introduced other things into the code?
[02:32:41] And it finds it was committed a particular day, the same day that the suspicious request was made.
[02:32:47] It exists in two places in the code.
[02:32:49] It's now trying to work out who this person's manager is.
[02:32:54] And if there's any Jira tickets that explains maybe why they did this.
[02:32:57] What do they normally work on?
[02:33:00] And it's looking at the Jira tickets.
[02:33:01] And there's no Jira tickets related to this change.
[02:33:04] So there's no good explanation for this.
[02:33:06] So it's busy drafting up a report now.
[02:33:09] It tells me actually from the enrichment data, I can see this IP address comes from Google Cloud.
[02:33:14] Here's the exploit.
[02:33:16] Here's the context.
[02:33:17] I recommend escalating this to high severity and treating this as a true positive.
[02:33:24] So this is really a pivotal moment in security.
[02:33:28] This is the first time in history that my capacity to investigate cases and my capacity to be thorough in investigating every single thing in my infrastructure is no longer bounded or limited by the size of my SOC team.
[02:33:40] With the power of Genie and the data that Lakewatch lets you bring into your security lake house, we can investigate anything at arbitrary scale without having to have everything being driven by humans sitting in front of their machine.
[02:33:52] Thank you very much.
[02:33:53] I'll hand it back to Andrew.
[02:33:55] So what Michael just showed you is how with Lakewatch, you can now analyze all of your data.
[02:34:13] You're not limited by scale or cost concerns there.
[02:34:17] You can fight agents on the attacking side with agents on the defense.
[02:34:23] And you can break out of proprietary data silos because all of your data is stored in open formats, in an open ecosystem, and you can load data from anywhere you like.
[02:34:32] We launched Lakewatch just a couple months ago, and we already have a number of marquee customers using it.
[02:34:38] Just yesterday we had the cyber forum and we heard from Atlassian, Adobe, and NAB.
[02:34:43] But to give you one example, Adobe wanted to analyze petabytes of data, which was too much for their existing legacy solution.
[02:34:51] They came to Databricks, loaded all that data, and they're now able to apply AI and ML techniques to find previously impossible defined threats.
[02:35:06] To sum it up, the only way that we can meet today's security threats is by having security and data teams work together.
[02:35:14] So, if you are a data and AI expert, I want to encourage you to reach out to your security colleagues and see how you can help.
[02:35:23] And if you're on the security side, you should move to using an open security lake house so that you can analyze all of the data and you can apply agents to find the latest threats.
[02:35:35] Thank you, and I'm going to hand it to Reynold to tell you more about the Panther acquisition.
[02:35:44] All right, yesterday, Ali talked about the Databricks' degree to acquire Panther, a cybersecurity company providing a leading AI stock platform.
[02:35:59] And today we have Jack on stage, founder and CEO of Panther with us.
[02:36:05] So, Jack, I think many of the audience here are coming more from a data background, so they probably haven't heard that much about you and Panther.
[02:36:11] Can you tell us a little bit about your background and what Panther does?
[02:36:14] Yeah, of course.
[02:36:15] So, I'm Jack.
[02:36:16] I'm the founder and CEO of Panther.
[02:36:18] We're a cybersecurity company with the mission of making security teams smarter and faster than attackers.
[02:36:23] And my background has been in security.
[02:36:26] I've been working in cyber operations for about eight years.
[02:36:29] And prior to Panther, I was leading detection and response engineering at Airbnb.
[02:36:34] And during that time, we found that the typical off-the-shelf sims just didn't really work for our use cases.
[02:36:41] They were really slow, really expensive, very operationally burdensome.
[02:36:44] So, we did what I would not recommend doing, which is building your own solution.
[02:36:49] But we ended up open sourcing that, and that became the inspiration for Panther.
[02:36:54] I see.
[02:36:55] And who uses Panther?
[02:36:58] Oh, well, Panther is used by some of the best technology companies in the world: Coinbase, Docker, Block, many others.
[02:37:07] As we're sort of getting to know each other, I realized you did something fairly unconventional from a founder's perspective.
[02:37:16] So, you were the CEO.
[02:37:17] At some point a few years ago, you told the board, "Hey, I want to step back from my operational role and focus on actually building."
[02:37:24] And spent 100% of my day hacking.
[02:37:26] And then later on, stepped back as a CEO.
[02:37:28] Can you tell us a little bit about that story and why did you go back to the hands-on typing code?
[02:37:34] Yeah.
[02:37:35] So, when AI was really coming up with LLMs, like as LLMs were becoming much more advanced, we wanted to ask ourselves,
[02:37:44] "What does this look like to change the paradigm where agents really operate in the center of the SOC?"
[02:37:49] And we couldn't just think about agents as attaching to the old manual workflows.
[02:37:54] We really wanted to fundamentally rethink things.
[02:37:56] So, where that started is simple code gen, like can I use an LLM to create a detection rule for me,
[02:38:03] like the ones that we just saw in the demo?
[02:38:05] And then that progressed into, "Can AI actually triage this alert fully end-to-end?"
[02:38:10] And that was just a massive breakthrough, just because the limiting factors for so many years have just been around,
[02:38:17] "Can we actually process the amount of alerts that we are generating every single day?"
[02:38:21] So, agents really allowed us to break that fundamental scaling problem in the SOC.
[02:38:25] And I spent all of my time just trying to figure out, from first principles, how do we really rebuild this with agents in mind?
[02:38:32] I heard the resulting product was pretty amazing.
[02:38:35] It's the core reason that people buy Panther.
[02:38:39] The other theme of this conference is, actually for the data audience here, we've been talking about,
[02:38:44] "Hey, we think security teams in the future will operate a little bit more like data teams."
[02:38:48] They'll start to get to understand more about how to actually operate data and use data to do defense.
[02:38:52] What's your perspective on that?
[02:38:54] Yeah.
[02:38:55] So, my perspective here is, security teams should always be focused on security.
[02:39:00] That's their core specialty.
[02:39:02] They are operations people.
[02:39:03] They have a really great understanding of how attacks work.
[02:39:06] And in the event of an incident, they really just have to focus on that end-to-end.
[02:39:11] So, what we've always thought of is, like, how can we give them the benefits of a really great data platform,
[02:39:18] huge amounts of security data, without the burden?
[02:39:21] So, that's really how we've focused on it.
[02:39:23] And I think that they've gotten a lot of the benefits, especially as we've gone more into AI and agents.
[02:39:27] So, we want them to actually benefit from the data infrastructure without having to get into the nitty-gritty details of how it works.
[02:39:33] Exactly.
[02:39:34] And maybe looking forward with the acquisition, what are you most excited about?
[02:39:39] Yeah.
[02:39:40] So, I think that Databricks and Panther are quite complementary.
[02:39:43] I think that where we focus is really on the application side of things.
[02:39:46] Like, how can we use all of this data that's collected across the enterprise to adequately detect and respond to threats that could cause harm to the business?
[02:39:55] And also do compliance, right?
[02:39:57] Like, these are the things that drive the business forward, keep it safe, and just really support the growth of the underlying company.
[02:40:03] So, I think that there's a lot of really interesting use cases where the security agents can have access to the broader data sets,
[02:40:10] and then they can better enrich and triage and allow security teams to just have more accuracy in their response.
[02:40:16] Okay.
[02:40:17] Well, I don't want to keep you here for too long.
[02:40:19] Thank you very much, Jack.
[02:40:20] And we're now going to pass back to Ali to just give us some closing thoughts.
[02:40:24] Thank you.
[02:40:25] Yeah.
[02:40:26] Super excited about Panther.
[02:40:27] It's going to be awesome.
[02:40:28] Okay.
[02:40:29] So, I just want to recap the last two days.
[02:40:44] So, this is the end of our keynote.
[02:40:46] We started the keynote by saying that we have all this AI that's amazing, but we have to give it all this context that we have at the bottom.
[02:40:52] And I hope that we've convinced you in the last two days that Databricks helps you actually get that context, that data control into the AI.
[02:41:02] So, what did we go through?
[02:41:03] So, we started with talking about Lakeflow.
[02:41:05] And Lakeflow is really now the one place where you can get all your data from the different sources.
[02:41:10] And we've expanded that with hundreds of connectors into your Lakehouse, into any open format that you like.
[02:41:15] So, that's awesome.
[02:41:16] Then we talked about Lakehouse RT.
[02:41:18] That's the Raiden engine, Reynolds' dream engine.
[02:41:21] And it's blazing fast.
[02:41:23] So, you've got to try it out.
[02:41:25] You can now get the dashboards to render extremely fast.
[02:41:28] The agents can move much faster on your data.
[02:41:31] And then we talked about Lakebase.
[02:41:33] So, Lakebase is serverless Postgres, lowers your TCO significantly on any apps you want to build.
[02:41:39] But the really cool thing about Lakebase was the fact that you can now actually combine Lakehouse and Lakebase in what we call LakeTap.
[02:41:46] So, this is one of the main reasons you should use Lakebase.
[02:41:48] Okay.
[02:41:49] So, that was the agentic data.
[02:41:50] Now we have our data ready for AI.
[02:41:53] It's fast, accessible.
[02:41:54] We then talked about governance.
[02:41:56] So, there's Unity Catalog and Unity AI Gateway.
[02:42:00] My favorite demo was Ankit, after his vacation, showed that you can actually use Unity AI Gateway to use this advisor pattern.
[02:42:10] The advisor pattern will use a small open source model for most of your work.
[02:42:14] And then it uses a tool and uses an advisor that's much smarter for expensive tasks.
[02:42:20] That way you can get your costs down significantly.
[02:42:22] So, that's unified governance.
[02:42:25] Then on top of that, the most important thing we demoed yesterday was Genie Ontology.
[02:42:30] Okay.
[02:42:31] So, it's a graph that we construct of all the enterprise knowledge that you have.
[02:42:35] And we do that continuously in the background and feed that to the agents.
[02:42:39] The primary agents, we introduced one Genie 1.
[02:42:42] Which is just a single pane of glass for all of Databricks.
[02:42:45] No more workspaces.
[02:42:46] No more different data warehouses or anything like that.
[02:42:49] You just have one place.
[02:42:51] Any business user in your organization can go in and ask questions from that data.
[02:42:56] So, please help your organization get Genie 1 in everyone's hands.
[02:43:01] And you do that by turning on what's called automatic identity management, AIM, in Databricks.
[02:43:06] Then everyone can log in to Databricks.
[02:43:08] And then you give them a consumer role so that they can actually access Genie 1 without having to see all of the rest of Databricks.
[02:43:15] Which can be daunting for someone who's not from a data and AI background.
[02:43:19] Genie Agents is your intelligent co-worker that actually automates things, can do things on schedule.
[02:43:24] Genie Code, we saw today.
[02:43:27] Mike Del Basso showed that you can actually build really advanced AI models, machine learning models that do predictions.
[02:43:34] And you can automatically troubleshoot them and continuously monitor them.
[02:43:39] We also saw Zero Ops.
[02:43:41] That's my favorite Genie because it just maintains data engineering pipelines.
[02:43:45] If they go down, it can fix them automatically for you.
[02:43:48] So, this, I think, is a huge, big advantage for any of us that manage data pipelines.
[02:43:54] Finally, we saw Justin demo the app builder.
[02:43:59] So, now you can Vibecode your own applications on top of Databricks.
[02:44:02] And it's part of Databricks apps.
[02:44:04] Okay, and then this morning we started with Matei on the developer side.
[02:44:07] So, developers who want to write their own code or they want to do coding.
[02:44:11] How they can use this meta harness, as we called it, Omnigent, as part of Agent Bricks.
[02:44:16] And we showed you all the different parts of Agent Bricks that we've launched.
[02:44:19] Like memory or storage and so on.
[02:44:22] Okay, so that's agentic development.
[02:44:24] Apps.
[02:44:26] That's a really great way to democratize access to everything in the organization.
[02:44:30] So, it's not just Genie App Builder you can use with apps.
[02:44:33] But we also showed how you can now get these micro apps that are low cost.
[02:44:37] And they can scale up and down quickly.
[02:44:39] Lake Watch.
[02:44:40] This is the sim.
[02:44:42] We just had on stage here Panther.
[02:44:45] And we talked about how you, with agents, can actually automate a lot of the toil that the security teams are doing.
[02:44:50] Triaging detections or alerts that come up.
[02:44:54] Or doing threat hunting automatically.
[02:44:56] And then finally, Customer Lake.
[02:44:58] This is a customer data platform that lets you do Customer 360 and infinity campaigns.
[02:45:04] Okay.
[02:45:05] So, let's zoom out.
[02:45:06] What does this look like?
[02:45:07] Well, it basically is the four things that we mentioned yesterday that we think are essential for getting the context from the data and the processes into your AI.
[02:45:16] So, it's choice.
[02:45:17] Open source remains extremely important to us.
[02:45:20] It's cost controls and governance controls.
[02:45:23] And then that context layer with Genie ontology on top.
[02:45:27] So, in short, we want you to have any data, any model, and any harness, no cloud lock-in for your whole infrastructure.
[02:45:36] We want to have one place to govern all your data and AI.
[02:45:39] And one place where you can do all of your cost controls for all the agents and the AIs.
[02:45:44] And on top of that, we have enterprise context for all of your AI and all of your infrastructure.
[02:45:49] So, that concludes our talk today.
[02:45:51] Thank you for coming the last two days.
[02:45:53] We have great sessions the rest of the day.
[02:45:55] So, please go to the sessions.
[02:45:56] Go to the Expo Hall.
[02:45:58] Work with all of our partners.
[02:46:00] Look at the sponsors.
[02:46:01] And then finally, we have an awesome party tonight at Oracle Hall with Chainsmokers.
[02:46:04] So, see you all there.
[02:46:05] Thank you.
[02:46:19] Thank you.