Beyond Data: AI-First Collaboration with Apps, Genie Sharing, and MCP in Marketplace

Summary

Databricks Marketplace is expanding beyond raw datasets to include AI-first collaboration assets: Genie Spaces, third-party applications, and MCP servers, all governed and running in the consumer's own environment.
Genie Sharing lets organizations hand partners a curated natural-language interface to their data instead of raw tables, with provider IP—instructions, benchmarks, and knowledge stores—remaining hidden while output is governed by aggregation controls and query quotas.
Third-party apps and MCP servers in Marketplace enable plug-and-play analytics applications and agentic tools that run entirely within the consumer's workspace, eliminating data exfiltration risk and enabling fast time-to-value.

Beyond Data: AI-First Collaboration with Apps, Genie Sharing, and MCP in Marketplace

Data sharing got us this far. The next leap is sharing the intelligence on top of that data: the agents, applications, and AI experiences your partners and customers actually want. With Databricks Marketplace, the ecosystem is expanding from raw datasets to a full set of AI-first collaboration primitives: Genie Spaces, third-party applications, and MCP servers, all governed, and all running in the consumer's own environment. In this video, we'll demonstrate three new ways to collaborate across organizational boundaries: Genie Sharing: a first-of-its-kind privacy-safe agentic collaboration tool. Hand a partner a curated natural-language interface to your data instead of raw tables. Provider IP (instructions, benchmarks, knowledge store) stays hidden, while output is governed by aggregation-only columns, row limits, and query quotas. Recipients can even blend in their own data without exposing it back to the provider. 3P Apps in Marketplace: plug-and-play applications running entirely in the consumer's workspace, from financial forecasting to manufacturing analytics. No data leaves your boundary. MCP servers in Marketplace: turn Marketplace assets (tables, notebooks, volumes, Apps) into agentic tools that Genie, Agent Bricks, or any external LLM can call as governed UC objects. Key Highlights AI-First Ecosystem: A curated library of Genie Spaces, Apps, and MCP tools, natively integrated with your governed data. Privacy-Safe by Default: Genie Sharing gives external recipients an agentic experience without exposing raw rows or proprietary logic; Apps run in-place to eliminate exfiltration risk. Speed to Value: Move from discovery to production in days, for data, agents, and applications alike.
Talk By: Akram Chetibi, Director, Product Management, Databricks ; Tia Chang, Product Manager, Databricks ;

Chapters

FAQs

What is Genie Sharing in Databricks Marketplace?

Genie Sharing is a privacy-safe agentic collaboration tool that lets data providers share a curated natural-language interface to their data with external partners instead of exposing raw tables. The provider's IP—including instructions, benchmarks, and knowledge stores—remains hidden, while output is governed by aggregation-only columns, row limits, and query quotas.

How do third-party apps in Databricks Marketplace work?

Third-party applications in Databricks Marketplace are plug-and-play solutions that run entirely within the consumer's workspace, meaning no data needs to leave the recipient's environment. This architecture eliminates exfiltration risk while enabling partners to deliver applications ranging from financial forecasting to manufacturing analytics.

What are MCP servers in Databricks Marketplace and how are they used?

MCP servers in Databricks Marketplace turn existing marketplace assets—such as tables, notebooks, volumes, and apps—into agentic tools that Genie, Agent Bricks, or any external LLM can call as governed Unity Catalog objects. This enables organizations to expose their data and AI assets as callable tools within broader agentic workflows.

How has Databricks Marketplace evolved from data sharing to AI-first collaboration?

Databricks Marketplace began with Delta Sharing in 2021 as an open protocol for zero-copy data sharing, initially focused on Delta-format tables. It has since expanded to cover notebooks, models, volumes, AI apps, Genie Spaces, and MCP servers, evolving from a data catalog into a full AI-first collaboration ecosystem.

Full transcript

[00:08] All right, let's get started. Hi everybody, my name is Akram. I am a director of product here at data bricks. I look after our ecosystem integrations uh sharing and collaboration. And this is Tia. >> And my name is Tia. I'm a product manager also working in the ecosystem
[00:21] space particularly um on marketplace as well as open sharing. >> Awesome. So, welcome to this session where we're going to talk about uh sharing and collaboration beyond data uh
[00:33] in the age of agents and AI. Super excited to tell you a little bit about some of the innovation we're doing and what we're hearing from customers and partners in in this space. So, this is probably not news to anybody
[00:47] here. Uh but collaborating beyond your four walls uh beyond your own company with your customers, with your suppliers, uh with your supply chain stakeholders
[01:00] is absolutely critical to kind of integrate your data and AI platforms with your ecosystem. Whether that's kind of peer-to-peer point-to-point integrations or it's licensing thirdparty solutions from partners or
[01:14] monetizing your own assets or even internally sometimes we see customers doing uh collaborations across business units in different countries or if there's M&A uh or uh SAS applications
[01:26] basically if you have your own data or AI assets sitting in different systems of records or software solutions you want to bridge that gap and basically break down those sites. and collaborate. And this is something that we hear time
[01:38] and again from our customers that being able to connect your data and AI platform to your ecosystem is a critical way to show value beyond the technical and performance and cost and like the
[01:50] technical value of the platform is the ability to kind of access assets that otherwise wouldn't be accessible within your four walls. So many of you I would assume anybody here familiar with delta sharing?
[02:02] >> Yes, most of you. So we have been at work for this area for a long time and the foundation has been delta sharing that we launched uh in 2021. So four or five years ago that really pioneered
[02:15] this idea of open collaboration and open sharing and it started with data specifically data that was in delta format initially and then it expanded based on that open crossplatform crosscloud cross region collaboration to
[02:29] expand to more assets and more formats. So you as you heard this morning in the keynote now Delta Sharing is graduating and becoming open sharing its own open source pro protocol that we're donating
[02:41] to the Linux Foundation and it's expanding from data to AI assets which is what we're going to talk about today and also expanding in terms of recipients with adding even more connectors. There is now hundreds of
[02:54] partners and thousands of customers and many many platform almost every single platform you can think of every major platform now supports uh open sharing formerly delta sharing and we're adding
[03:06] also support for on premises so you can see this ecosystem is already expanding uh pretty drastically and pretty fast and there is hundreds of partners I have a slide here with all the logos every industry you can think of if you're in
[03:18] healthcare life sciences you can see the IQVAS of the world if uh financial services, factset, S&P, London Stock Exchange Group on the horizontal side like SAP, advertising,
[03:30] the trade desk, I mean you name it, every industry you can see the major players are using this kind of open sharing uh ecosystem. Uh there is going to be a number of sessions throughout the week about open sharing. Today what
[03:42] we really want to focus on is uh beyond data. So sharing data, we've been at hard work at this for the last four or five years and we're now thinking about the next wave of collaboration and
[03:54] sharing beyond the data. Meaning how do you collaborate on the intelligence that sits on top of the data and not just the data itself and that's really the focus of the session today. So you have seen maybe in the last three
[04:07] four months we've launched the uh I think it might have been more than that maybe six seven months the MCP marketplace. So this has been um something that a lot of customers asked for is how do I make sure that Genie whether it's Genie1, Genie Asians or
[04:20] Asian bricks can be extended to an ecosystem of tools uh that I can use from partners. So we made it extremely easy to connect these third party MCP tools into your Gen and Asian bricks
[04:32] agents using the MCP marketplace. So it's growing very fast. We have already 60 plus MCP servers and every week we're adding uh even more for different industries and it seamlessly integrates
[04:44] these external tools and remote MCP servers into your data and AI platform via Unity catalog and Unity AI gateway. So it's the same uh framework you use for governance for access controls that
[04:57] you would use internally with your own kind of authentication uh works natively with these kind of external tools. So if you want to augment an a custom agent you're building with Asian bricks or you
[05:09] want to make Genie smarter with external kind of tools, you can do that with MCP marketplace. So this has been out there for a while now. Many many customers are using it and it's uh it's seeing a lot of uh adoption just to kind of makes it
[05:21] easy for customers to set up these MCPs and the uh idea here is to really unlock specific use cases for customers. So we really want to make it easy to scale this across many tools so that you can
[05:33] connect to your ecosystem whatever the use case is. So like I put some examples here in the slide but you can see in financial services you can have an MCP from Moody's for like credit research or codality from like real estate uh data.
[05:47] You can have like basic web search with like u.com and that would make your agents act give you the ability to combine your own firstparty data with these third party kind of data providers
[05:59] or tool providers. Same thing in healthcare. I put some examples here but there is like many of them out there. We put out a few blogs that show these examples in in real life. So this has been what we've been uh up to in in the
[06:11] MCP world and we really made it super easy for providers on the other side to kind of onboard to this so that anyone who has an MCP server and a remote one that that they have they can in a few
[06:23] clicks you can see here a snippet like a couple of clicks you can get up and running and list your MCP server in the marketplace. So the idea is to really increase selection so that when you go to the marketplace you can see dozens
[06:35] hopefully soon hundreds thousands of tools being added here as one place where you can govern all of your kind of integrations with the ecosystem. But MCPs are one way to integrate with
[06:47] external tools and external providers. They are a great way that allows providers to package up their intelligence and context and give you like a standard protocol to connect via
[06:59] MCP to their APIs or to their agent. They work great with all agents including Genie. They allow the providers to protect their intellectual property so they don't expose uh whatever is happening behind behind the
[07:11] API. But there are some use cases where there are some challenges with the MCP approach. So for example, the provider has to develop an MCP server. So they have to develop it, they have to operate it, they have to host it. It's often
[07:24] time it's a multi-tenant server. Uh so that's kind of work that needs to happen. Uh from a privacy perspective, if you're using a remote MCP, it's great because it just kind of works and all the questions are routed uh uh back to
[07:37] the provider. But that can sometimes raise questions of like do I want every single question to be routed externally to the provider of the MCP or do I want the answers to be uh uh basically generated locally and then finally uh
[07:50] with these kind of um uh MCP approaches remote MCP approaches um it's often times difficult to have a question that combines your data or multiple data sets from different providers or different
[08:02] sources into a single question that gets a single answer because by definition, it gets routed into different tools and then it's hard to have a question that maybe does a join between one data set
[08:15] that sits in some tool and then another data set that sits elsewhere and the answer actually is not um just two individual questions. It's actually one question that combines the two data sets and gives you an answer. So this idea of
[08:27] combining your own data with external data is is becoming even more important. So these are some of the questions we've been hearing and we're like okay how do we solve this and make it even more seamless both for providers who don't necessarily want to operate and manage
[08:40] and host an MCP server but also on the recipients who want to kind of optimize for privacy and unlocking the value of their own data in these kind of AI collaborations. So that's why we are announcing today the Genie Asian sharing
[08:54] with data control. So, it's an an idea that allows data providers to package up their data as part of a Genie agent and then when they share that out, they share a natural language interface that
[09:07] makes it very easy for the recipient to get started with that data set and then get started using uh basically an AI experience. And the data provider is not only packaging their data, they're packaging the data and the semantics and the
[09:21] knowledge that come with it so that the experience is already fine-tuned and ready to go. So that when I receive that agent, I can get started um questioning and the kind of time to value becomes significantly faster relative to just
[09:34] receiving a data set. And then I can combine that securely uh with my own data. So I have this Genie agent and I can add my own data to it and now it it becomes even smarter because I can ask a
[09:47] question about my data and the third party data set in one configuration in one setup and then finally it unlocks new ways to do asset monetization whether your company is traditionally in
[10:00] the business of licensing assets like a data provider or like a software provider or you're not in the business of licensing data and AI assets but you're looking for new ways to generate revenue venue and monetize uh your
[10:12] assets. That could be very exciting for many companies and many CEOs are looking for ways to use AI to do monetization. So this could be a new way to u package up your assets as part of an Asentic experience without necessarily revealing
[10:25] the underlying data or the sensitive data that sits under the hood. So from a recipient perspective and we're going to see a full demo at the end that TIA will will do. Uh but from the recipient perspective, it just looks like a
[10:37] regular Genie agent. So it's a similar the same familiar experience that you are using. Uh the uh Genie agents are prepackaged with the provider instructions and semantics and you can add your own data to the agent so that
[10:51] the agent comes already configured. It has all the instructions. It has all the semantics of it and then if you want you can add your edited add your data to it and it works with genie genie one uh all the genie integrations in slack and
[11:03] teams and other things all work seamlessly with it and the architecture is relatively simple is u the idea that uh on one hand you have on the left side you have the provider's data and the
[11:15] provider semantics and instructions and on the right hand you have the recipient's data and then they're both combined in a single agentic experience so that the end user can actually add their own data have been when when integrating with this and this is all
[11:27] done in a secure way. So from a provider's perspective, it's super easy to basically stand up an agent and then test it out locally and then once it's up and running, you can package it up and share it. And we have this idea of
[11:40] data controls that allow the data provider to decide what they want what kind of interaction they want to allow on that via that agent so that they can control how many questions are asked or uh the limit in terms of outputs that
[11:53] are generated. So I have a small video here that will actually show a provider experience. It's a very short video. Uh but basically I am here in my genie uh
[12:06] user interface and I am looking at a genie space and I can create one just like oh can you see it? Yeah. >> Uh I can create one and I can add my data uh to it just like I do on a regular genie space. So here I'm just
[12:19] kind of browsing uh the data sets. I have a world cup data sets. I'm going to add this uh world cup table in there. And as you can see, I will be able to create this genie space and I'll be able
[12:31] to share it the same way I share a table or an AI model or a volume today. I can just go and hit share and it will ask me of the sharing identifier of my recipient. This is the same way I do
[12:43] open sharing today or delta sharing uh for volumes, AI models, notebooks and all of the assets that we support. So this is consistent with that. And what is really uh interesting here for data providers is that the privacy settings.
[12:56] So if I go here as a uh creator of the space I can decide here is the koda number of queries per day that I want to allow. Here's the limit that I want to allow on in terms of output uh in terms
[13:08] of size uh and the bytes that are outputed in terms of rows and and megabytes. So this is basically the the overall experience. It just shows you a little bit the simplicity of it. And uh the idea here is just to really give
[13:21] providers control over their uh intellectual property but at the same time give recipients privacy in terms of the questions they're asking and the data they're adding to the agents. Um so this is uh the provider experience on
[13:34] the genie space. We are partnering with many data providers and and and solution providers that are looking at this. Uh Kitera Labs is one in the healthcare and life sciences uh space. they have solutions that have very valuable data
[13:46] sets for clinical trials and other use cases. So they're packaging some of these um uh data sets in um uh in genie spaces and they're looking at like uh leveraging this to reach more more of
[13:58] their customers. So that was on the Asian experience. So we talked about MCP uh collaborations with the MCP marketplace. We talked about okay now I want to share an agent that actually can combine my data with
[14:12] third party data in in a secure way for both sides. But there is this question of like okay agents are one way to package up a solutions but what if I want to share a full experience an intelligent application that kind of
[14:26] combines all of these things and provides an end toend user interface for the customer. So this is what uh Tia is going to talk about with the apps marketplace. >> Awesome. So,
[14:38] as some of you guys might know this, we kind of launched this late last year and it's something called data bricks apps. And as a quick primer, basically what they are are interactive like data and AI apps that you build natively and deploy natively within data bricks. And
[14:52] they kind of span across all different types of use cases. Like under the hood, you could have tools that you're using. You can have this as a chat interface that Aram was showing earlier. Um, you can have either like you can also have
[15:04] agents under the hood as well. Um, and they're all basically seamlessly integrated into the rest of the data brick stack. Um, and your data also is all secure and governed and everything goes through Unity catalog. Um, and it
[15:16] also is built on top of open framework. So for example, if there's Python frameworks like Streamlit, Gradio um, and then also Node.js frameworks. So that it's pretty much what your developers are already familiar with.
[15:28] Um, and this has really grown a lot over the past year or so. like we're seeing a lot of thousands of enterprises actually they deploy like hundreds of apps every single week um in production. So that's
[15:40] something that you know we've seen a lot of internal usage for. Um and we're also seeing one more trend and is that with AI it's actually a lot easier now to build these apps right you can vi code them you can see your developers instead
[15:52] of turning around in the span of months they can actually build something in maybe matter of days now. Um, but we're still seeing one gap though, and it's that to make these apps actually like productive or even effective, you have
[16:04] to have the right context and the right data and the right metadata flow into it. Um, but a lot of times customers, especially for external use cases, well, they're a little bit hesitant about that, right? Like I don't want to
[16:16] necessarily share my data with you into an app. Um, and I don't really know what you're going to do with it, right? And usually there's secure there security as well as deployment review cycles that take three to six months and it just
[16:28] really slows down this entire process. Um and we're basically left with a bunch of apps but not much to do um with connecting them back into your enterprise data. And that's why we're super excited to launch apps on data
[16:41] bricks marketplace. And basically what this is is you can actually now discover and deploy thirdparty apps from our partners. And you can even use your existing data bricks commits to access them. Um and starting this week you can
[16:53] actually browse if you guys also like this morning we just launched. So if you like go to your phone you open up open marketplace um you'll be able to see and we'll show this later as well. You'll be able to see our live apps in like
[17:05] production today. Um so we're super excited that you get to have access to these dozens of partners. Um you can deploy them and install them with a one-click very easy installation experience. Um everything is built in
[17:18] through uh governed through Unity catalog as well as Unity AI gateway. So if you want to configure say like permissions to your underlying uh model serving endpoint or to an underlying agent that your app is using, you can do that. Um and and finally like everything
[17:33] is IP protected as well. And what this means is you actually don't need to see the code complexities that are underneath every single app, right? So it strips away a lot of that so that you can get into production, you can get
[17:45] into the insights right away. Um and at the same time the provider source code is hidden and stays close source this whole time too. Cool. So that very last point I actually wanted to go a little bit deeper in. Um
[17:59] so what does actually mean by like oh the security stays closed source and it's living on my like workspace like what does that mean? Right? So here's just a little diagram for you guys. Um, I want you to notice that like once you install the app, it actually goes right
[18:12] into your own workspace, right? So, this it's no longer touching anything from the provider. Um, it's in your own workspace using your own serverless compute. Um, and obviously that's where your data already lives. So, your app
[18:25] just builds on top of that. Um and it also uh once you bring in like say any resources or any scope that the app needs um everything you can basically configure say even if you want say
[18:38] external APIs to be accessed from via your app. It's not like there's absolutely no egress. It's that everything is declared up front when you install the app and you have to give explicit consent um in order for this to
[18:50] be accessible basically. So let's take a look at who's launching with this right. So, we've actually been working with them for the past couple of months already. So, I'm actually like super excited. We have 20 launch
[19:02] partners with us. They span across four major categories. The first one is like identity and privacy. Um, and then we also have adtech, marketing, activation as well as data quality and reasoning. And finally, there's also developer
[19:14] tools and operational tools as well. Um you might like recognize major brands such as say capital one, meta, trade desk and then you'll also see specialist providers like lake fusion or data vant
[19:27] obsera and bright data right and that breath um of all these different providers and the offerings is really what makes data bricks such unique place that it brings together um this type of thirdparty intelligence that we're
[19:40] talking about right to where your data already is living. Um now let's just give maybe a concrete example across one per category right so if we talk about datavan I know this is
[19:52] probably quite small so maybe you can't actually see it um but what datavan does is they actually tokenize the patient data that you pull into it so that health records can be linked and you hide the underlying phi let's say the
[20:05] second one is let's look at meta um so it actually sends your first party conversion data straight to the meta API so that you can activate any downstream advertising campaigns or use cases you might have. And then third one, let's
[20:17] look at Lalom over there in the third column in on the bottom. So what they actually do is they help you AI generate metadata and it actually pulls that from different source system. So let's say I want to have something from my Jira or
[20:30] from my confluence or my internal Wikipedia um that my enterprise has. It actually basically taking all that knowledge also generates metadata for you to feed right into UC so that your
[20:42] data is AI ready. And finally on the right side if we look up um they help you basically track your pipeline health as well as the operational health um of your data living in living in data bricks um so that there's really a lot
[20:55] of visibility and transparency across your entire organization. Great. Um, and now that we have kind of some idea of like what the offering is out there, I just wanted to walk through kind of what is the process to create an
[21:08] app and list it on the marketplace. Um, so it's really not meant to be a complicated process as all at all. Um, so you start with you build your app locally and then you deploy it right into the data bricks apps uh platform
[21:21] and then you create a listing on marketplace. You add in your regular say like listing description, your attributes, whatnot and then you attach the app to it, right? Okay. And then the next step is you actually declare any resources or any scopes as well as any
[21:35] external egress endpoints that your app might trigger. And then you can also add on any assets. So maybe you have a readme file. Uh if you have companion notebooks, you can do that as well. Um and then finally it goes through this
[21:48] comprehensive security process um and security scan process that we have internally to make sure there's no vulnerabilities um and your data is not actually being shipped off to somewhere that you're not aware of. And then finally you can publish it and you have
[22:00] access to our entire customer base of more than 20k uh customers. And this whole time once again everything stays closed source. Great. So I know we've gone through a lot of slides. So let's just go and see
[22:13] how this works. U we've shown like pieces of what the experience is like on the provider side but let's put it together as a customer now. Um and here's a scenario right? So imagine let's say I manage the retentions team
[22:26] at a athletic footwear company called Stride Footwear and my CFO is asking me hey like which customers of ours is at risk and what do I do about it? Um so I'm going to show you guys how I compose
[22:39] this across the three different features that we talked about today. So third party intelligence on top of our first party data. Um and everything is running on my own govern environment. So
[22:51] let's jump over to marketplace. Let me go like this. Great. So this is our data bricks marketplace um in product experience. So
[23:05] you can see this very exciting banner um that announces the launch and then you can click into it. You'll see each individual offering. But let's say we click here and you actually can see there's this brand new filter that was
[23:17] actually introduced as well. Let's like click on it and I actually see a combination offering across first-party apps which we launched last year. Um so these are open-source apps that uh are
[23:29] mostly built built by our field team um to tackle like specific customer problems. But more specifically, let's go down to the bottom where you see a selection and these are some of the logos that you saw earlier um of
[23:41] partners that have launched their app. Right? So let's say I'm looking around. I'm looking around. Oh, if I want to figure out which customers are at risk, I actually have a point of sale data.
[23:53] Um, and then I also have CRM data. So, I have like 30 records, um, of how my transactions are going within my organization. I want to clean it up, right? So, I want to look for entity resolution uh, product. So, I come over
[24:05] here and I see, oh, okay, like Fusion looks pretty interesting. I can click in here. I can look at screenshots of what it looks like internally. I can see what the recommended use cases are. Um, and then here I can see explicitly like what
[24:18] we talked about earlier, the resources and the scopes that um are necessary and required by this app. And then there's also oh nice okay there's even an installation guide that makes it easy for me to just deploy out of the box. So
[24:30] I come here and I say okay I want to install this app accept terms and I go right into the data bricks page where I basically uh config and give access and
[24:42] bind my resources um to whatever the app needs right so here create a database bind it create a secret bind it um for the sake of time I've already deployed this app already um so this is like kind of what it'll look like after you click
[24:55] install deploy takes a few minutes to run and then you'll see this URL Well, that'll take you directly to the UI of the app itself. Um, so here you can see, okay, great. Like, yeah, I want to import my data sets, which I've already
[25:08] done. And then maybe I want to set some rules of um how my entities map to each column. So, I'll add like some attributes. So, you can see there's addresses, there's first name, last
[25:20] name, there's like the CRM account ID for example. Um, and then I want to map them to like based on the two tables that I have like what account or what column actually maps to what column in each table. And then maybe I set some
[25:33] survivorship rules as well. Um, all good. So then after that config is done, I come over here and I see okay, I can actually train entities and run this resolution directly, right? So I'll even
[25:46] look at the playground and see what is in my table, whether the prompts that are needed might take a few minutes to load. Um but also for the sake of time I basically
[25:58] what this does is it'll write back like a golden record table right into my UC. Right? So we can see over here if we go here under lake fusion catalog I have a golden record output um already generated for me. Right? So, uh, after
[26:12] all that, like what I have done is I've taken my dirty records that are hard to reconcile. Don't really know what they are. Like maybe I know Bob is here but Robert is there. Like who's who? I don't
[26:24] know. Um, but now I have a golden record that is clean, right? Um, and the next step is, oh, I actually now want to learn about the behavioral u and the behaviors of these of of my customers,
[26:36] right? Like are they about to churn? Like how sensitive are they to prices? Um, so I've actually acquired a thirdparty data set from a consumer IQ company and they've shared a genie agent with me with their thirdparty data in
[26:49] it. Right? So I'll come over here. This is just our like open sharing uh kind of page where I see oh okay I actually have this consumer IQ customer behavioral insights genie agent shared
[27:01] with me. So, let me go ahead and mount it actually. And I'll just say I'll just keep the same name. Might take a little bit to mount, but what I'm expecting to see here is a
[27:15] fully populated Genie agent essentially. So, it should have that third party data in it. It should have any uh sample instructions. It should have sample queries. So, let's see if it actually does. Um, cool. Oh, and even nicer, like
[27:29] Genie actually gives me a little AI description as well, so I know exactly what's in here. Um, great. Okay, so I see the data actually contains this third party behavioral table. And then in the instructions, I also see you are
[27:42] consumer IQ, a genie agent, blah blah. Great. So like everything is set up for me. I even see like sample queries to see like, oh, which of my customers are at risk and they're price sensitive. Like maybe there's a specific logic that
[27:54] consumer IQ likes to think about, right? So they've created a SQL query um for Genie to learn from and to apply future in the future as well. Um but I want to do one more thing. I also
[28:07] actually wanted to bring in some live data uh from the market about pricing as well. Like if I'm learning about price sensitivity, maybe I want to know like how my competitors are pricing their products live, right? So let's actually
[28:20] go to Genie1. Um let me just click here. And what genie one like the difference between that and what we just saw the genie agent is simply genie one you can think of it as the orchestration layer right it will intelligently pick which genie
[28:34] space is the best fit for the question you're asking at hand and it can even like switch between multiple ones and give you a combination um of responses across the genie agents that you have in your uh workspace. So here, let me just
[28:47] ask um what customers are the most likely to churn and are sensitive to price. So
[29:02] let's see what happens if we ask that. It's going to think for a little bit. I wonder if it'll know which genie agent to use. Hopefully it does.
[29:16] Okay, it found six results and great see exactly it's it's calling that Genie space that I just got from my third party provider. Um, and now it's asking that specific
[29:29] Genie agent the question that we just asked. This might take a little bit,
[29:42] but I'm sure like this type of like interface is super familiar to you guys. like nowadays. So bear with me for a little bit.
[30:00] Oh wow, nice. It's even creating viz for me, right? So I can actually see maybe a graph or uh a dashboard um of this result. Let's
[30:13] wait a bit more.
[30:28] It baited me. I thought I was done.
[30:40] Um, and we can see what the MCP offerings there are. So, we can filter right here for MCP servers. You see there's actually 63 servers available um, for install today, right? And this really spans across like you.com for
[30:53] example is like a app is a web browser. Um, and then there are things that are more verticalized for like specific industries. So, you see Fax, S&P, so finance. Um, I don't know why the logos
[31:05] are not showing up, but that's okay. There's like Kytherra, there's Informatica, um there's even like Census, MCP. So, it really spans across a variety of different use cases. Um and
[31:17] you can basically like set this up directly as long as you get like the right credentials and the bear token directly from there. So, let's see if it's done yet. Great.
[31:29] Oh, was this the right one? Yes. Yes. Yes. Okay. So, it tells me, oh, I actually have 1,200 customers about to churn. They're high risk of churning. Um, and I didn't even know
[31:41] that, right? And I actually see that of the customers who are about to churn, they also are price sensitive, right? And at risk price sensitive customers, 78% of them show declining engagement
[31:54] trends. So, I know that, oh, something is wrong. I know that like 1,200 customers are about to turn and they're price sensitive. So, I need to do something about it. But I also want to know like hey what are my like I don't what are my competitors doing. So I
[32:06] actually click on the plus here. I go to more connectors and this is where you can see the actual like you see connections that you have set up already and it spans across both external as well as your own custom or even your internal MCP connections. So here I'm
[32:19] going to just add in the one from you.com. So if you remember that was the one I was doing web browsing and I'll just ask it really quickly. Uh, what are my top competitors pricing
[32:31] their athletic? Well, can't type footwear as they don't even care about typos, which is great.
[32:45] Yeah. So, you can see now it's like triggering web search, right? So, it's no longer the context is not just restricted to whatever is living in my own environment, in my own workspace. like I can actually now have access to what is publicly available on the internet.
[33:06] Yeah, it says no internal data. So, it's going to go to external data and we might have to wait a mile again. But basically what will happen is it's going to tell me that all my competitors are doing discounts. Like Adidas is doing a discount, Nike is doing a discount. So
[33:18] what is my takeaway? Right? If I have customers about to turn, they're highly uh price sensitive and my customers are doing a discount. I also probably want to do a discount as well. And I was only able to get to that conclusion.
[33:32] Still taking a bit. Oh, there we go. Great. Yeah. So, it's telling me like for this like at risk customer segment, I should
[33:44] price it in this lower range basically, which is lower than what I'm pricing today, right? So, let's just go back to the slides real quick and let's recap a little bit of what we saw
[34:01] is that let me skip this uh using the lake fusion app I was able to do MDM and get my golden record and then using that thirdparty genie agent I was able to bring in thirdparty intelligence and basically enrich my first party data and
[34:15] finally using that external MCP server I was able to get real live market signals um and get the pricing of my competitors and give it urgency like hey you need to fix your pricing right now
[34:27] or else your customers are going to turn. So all in all the app made my data trustworthy. The genie agent grounded it in intelligence and then finally the MCP gave it urgency. Um and this whole demo
[34:39] basically was just done. I don't think I would have been able to get to that conversation or conclusion with just a singular piece of it. So this is really thirdparty intelligence built on top of your own firstparty data and it's all composed by you.
[34:57] Great. So just to close out kind of what we talked about today um talked about apps, talked about genie agents and we talked about MCP servers as well. Um and this is really two sides of the same marketplace, right? So customers they can compose that entire uh intelligence
[35:10] on their own data and then providers can distribute and monetize that intelligence um without having to rebuild it for every single individual customer. So they build once and then they can distribute everywhere.
[35:22] Um great. So for next steps, how do you start? Feel free to start going to marketplace. You can browse all the fund listings that we just talked about. And then if you are a provider and you're interested in listing any of these
[35:34] assets, come talk to us. Um, super excited to hear about your use case as well. >> Cool. >> Awesome. Thank you.

Learn more about the Databricks Data and AI platform.

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