Claude Code vs. Databricks Genie for data apps

Summary

Victoria, principal technical evangelist at Databricks, presents a framework for choosing AI coding tools—from rapid prototyping with Lovable and Replit to production-grade applications with Claude Code, the Databricks AI Dev Kit, and Genie.
The video demonstrates building financial data applications that extract insights from SEC filings into Delta tables stored in Unity Catalog, then query results using natural language through Databricks Genie.
Genie's two-times higher success rate for complex Databricks workflows makes it the preferred tool for end-to-end data applications requiring governance and scale, while Claude Code and the AI Dev Kit are better suited for writing and deploying production application code.

Claude Code vs. Databricks Genie for data apps

Choosing the right AI coding tool depends on your use case: rapid prototyping or production-grade control. this video walks through selecting between Lovable and Replit for fast iterations versus Claude Code, Databricks AI Dev Kit, and Genie for end-to-end data applications with governance and scale.
Learn how to build financial data applications from scratch, extract insights from SEC filings into Delta tables, store data in Unity Catalog, and query results with natural language through Databricks Genie. Understand when to combine tools for maximum efficiency, the benefits of each approach, and why Genie's two-times higher success rate makes it ideal for complex Databricks workflows.
🤝

Chapters

FAQs

What is Databricks Genie and how does it compare to Claude Code for data apps?

Databricks Genie is a natural language query tool for data on the Databricks Data and AI platform, while Claude Code is an AI-assisted coding tool suited to writing and deploying production application code. This video explains that Genie's two-times higher success rate makes it ideal for complex Databricks workflows, whereas Claude Code and the Databricks AI Dev Kit are better suited for end-to-end data application development.

When should I use Lovable or Replit instead of Claude Code for AI-assisted development?

Lovable and Replit are best for rapid prototyping and fast iterations, while Claude Code, the Databricks AI Dev Kit, and Genie are designed for production-grade data applications that require governance and scale. This video walks through a framework to help practitioners select the right tool at each stage of development.

How does the Databricks AI Dev Kit help build financial data applications?

The Databricks AI Dev Kit supports end-to-end data application development, including extracting insights from sources such as SEC filings, storing results in Delta tables, and managing data in Unity Catalog. This video includes a demo showing how to build a stock dashboard and scale it using the AI Dev Kit.

What is agentic engineering and how does it differ from vibe coding?

Vibe coding, a term coined around 2025, refers to using AI tools for rapid prototype generation, while agentic engineering describes the newer practice of using AI to build and operate production-grade applications. As noted in this video, Anthropic reported that approximately 80% of their production code is already AI-generated, signaling a shift from prototyping to agentic engineering.

Full transcript

[00:20] All right, welcome back everyone to the Summit Live studio right here in Moscone Center. Shout out to 160 plus countries around the world watching. And actually at the Fox Sport Lounge, they're having a watch party of the World Cup right now. So good luck to your country, but I am
[00:36] very, very excited and pleased to have Victoria and we're on the same team here at Databricks. But welcome Victoria, tell us a little bit about you and your role here. Yeah, well, thank you so much. So exciting to be here today. So my name is Victoria. I'm a principal technical evangelist at Databricks. So
[00:54] in my role, I build demos. I really go after our product launches, so I think about our team as a first user adopters, right? Like we build those tutorials to help our customers to adopt these new features and teach about new features. So really love sharing
[01:11] knowledge with data and AI practitioners. And today we have a very interesting topic to talk about because this ecosystem is evolving so fast. Like a year ago in 2025, Vibe coding term was coined, right? Like
[01:27] it's just like a brand new thing. But now, like recently Anthropic said that 80% of their code, production code, is already being AI generated with Cloud Code. So we are no longer talking about Vibe coding those prototypes. We
[01:44] actually talking about coding the production apps. And now this year, we're utilizing a different terminology, agentic engineering. And as these tools evolves, every company
[01:59] realized that they need to provide this AI assisted coding in their platforms. And what we see now, the emergence of those tools, which is fantastic, but it also creates kind of like this dilemma when you have too many choices. Like,
[02:15] well, now I don't know what tool I should use, right? Even within Databricks ecosystem. So, today I want to talk about kind of like a framework going from prototype to production and how to utilize different AI assisted
[02:31] coding tools to really help to improve this selection choice. Yeah, which is super important now since every couple of months something new comes in, something old phases out. So, what you're about to talk about really future-proof. But, what I love is of all
[02:46] the sessions, you know, separated from the the keynote, this is like really practical and I get asked this very question a lot. Like, how does Databricks fit in with that whole vibe coding? Where are we helpful amidst the ecosystem? Yes. Super excited to jump in.
[03:01] Yeah, so um first I want to kind of like a way of the land. We can think about different things. First is prototyping and building those AI powered applications. And those AI powered applications, they can be production
[03:18] apps, right? So, think about lovable Replit. You can prototype something very quickly and you can also deploy those production apps. So, then we have Cloud that is more for technical users. Right. And you also can build production apps.
[03:33] You can also prototype. And then we have inside Databricks, Genie, which of course was also talked today uh during keynote. So, now how do you make this um selection? The way I think about this is if I am building an application and
[03:50] if this is just my idea, I want to get something very quickly as an app. I will go to the stores like Replit or Lovable that are built specifically for this to build prototypes. Those tools, they have
[04:06] integrations to Databricks, right? So you can connect those AI apps to your Databricks Lakehouse. You can connect them to your Delta tables. You can write, you can read, and you can build on top of your data foundation.
[04:24] And what's good about this? It's very easy. All you have to do is just, you know, like a simple prompt and you actually get nicely polished UI. So it's really easy to get started if I am just exploring different ideas. Maybe
[04:39] I want to bring it to my manager. This is where I would start. Just go to the specifically purpose-built AI-powered apps and try them. Yeah, I I tried Lovable myself among many others. Like I used to work with professional baseball and just make a website that has, you know, the top 100
[04:56] baseball players and their salaries and it makes it very very quickly. But it's very limited as I found out. It was great to make a website, but how do I take it to the next level? Exactly. That's such a good point. So So let me show you this very what what you
[05:12] just exactly described. So let's say I want to build an app that will Let's say I want to do some kind of analysis, financial analysis. Give me a ticker of publicly traded company. Okay, how about Walmart and the ticker
[05:28] is WMT? okay. retail company. So let's see. Let's see if Okay. Yeah, so you can see right away I built this app simply by asking prompts. Now I have this very
[05:44] nice stock dashboard. And then what I also created here is pulling SEC filing data. And the way that it's been done, the logic lives inside the Laravel, right? So, Laravel builds up its builds a logic, and then utilizing
[06:01] um utilizing uh Delta our Databricks integration, I can write the data. So, let's say when I click extract, it's going to go call an API, and then pulls this data and write
[06:17] it to Databricks. So, that's good, you know, I have some idea, I build this, and then I can go to Unity Catalog and review that this data is extracted. But now, as you said, when you start thinking about um
[06:34] but what if I don't want just extract information? Maybe I want to pull financial statements, like get a PDF document like this, Nice. and utilize uh some of our functions like AI parse document or AI query, and
[06:50] now at a scale processes information. So, I want to get into the granular, right, information state in PDF. Well, now you're talking about building more like end-to-end because you need to think about data engineering, not just
[07:08] visualizations, right? Mhm. So, this is where you would find that this was a great starting point. It definitely helps to get started, but then as you think about end-to-end,
[07:23] we need more tools, we need more integrations. And one of the way to start doing this with AI-assisted coding is by utilizing Databricks AI Dev Kit. So, Databricks AI Dev Kit is an open-source project. I think it has like
[07:39] what, 1.7 uh stars. Um so, So, you can It provides It provides over 50 skills and tools. So, let's say if you want to build a job or maybe you want to um
[07:57] store those data into volumes or you want to create notebooks, you want to run notebooks, right? Now, all of the skills are available in DevKit and you can utilize your favorite AI coding assistant to connect to this DevKit and start
[08:14] building. So, I often, for example, use Cursor because this is more like, you know, I want to see what kind of code is being written and I can control more. So, the way I think about lovable, Replit, all of those apps that we
[08:29] uh we talked about they are more for non-technical users, but as you increase complexity and you want to really think about your entire pipeline from data engineering to storing data to data visualization to building those
[08:44] different AI tools, that's where AI something like AI DevKit will uh will really help. So, uh to give you a uh side-by-side comparison, so let's say this is also an app that I built with AI DevKit. So, all I did the same thing. I
[09:01] went to code code. I said, I want to have this app because it was just a simple dash app. It doesn't look as fancy, but let's Let me also pull here, do the same thing. I'm searching for more Walmart and
[09:18] it will get me the same information. But now, in addition to what we discussed that pulling the visualizing data or saving data, we can actually have those jobs created, right? And jobs initiated. That's because we
[09:34] have a lot more access to our ecos to our platform and integration. So, once it will load, we might have some internet uh issues. Uh we should see the same kind
[09:50] of chart. And let me switch back. And I used to work at financial companies and it take usually about 2 to 3 weeks for people humans to do the research. Exactly. Because there are so many like
[10:06] those documents, some of those 10-K statements, they are they can be 100 pages, right? And to process them, that's that's really amazing if you can build like an app so I guess it will help you to get insights very quickly because once you store the data into
[10:22] Delta table, then you can use something like Genie and then start asking questions, right? Because now this data saved in Unity Catalog. Actually, let me show you how it's saved in Unity Catalog. So, as an example, we um the data that was pulled was stored
[10:40] here in Unity Catalog and I can just simply, you know, let's say ask for for a ticker and I will I will be able to see this data. And then I can also use Genie and then ask questions using
[10:56] English. Not SQL, right? This is great. What does this filing say? What, you know, Exactly. How do our competitors Yeah, like calculate for me P/E ratio, you know, all of those different things that you can just simply ask questions. So, now
[11:12] where do we go from here? What Basically, this AI Dev Kit allows you to do, you can now take any of your favorite AI codes and tools and connect to it. And now you can build this end-to-end workflows.
[11:30] Now you can say, "Well, this is great, Victoria, but Databricks also built Genie." Mhm. Then, where do you see the difference, right? So, the way that I see this, when you build
[11:47] with AI DevKit and you utilize something like Cloud Code, it has MCP it utilizes MCP connections to go to Databricks and execute the same It utilizes those Genie models, right?
[12:03] They don't really fine-tune based on your Databricks data. So, when you say, "Build me a Databricks app," the model goes, tries to identify which tool to use from those available MCP tools, and start building this app. And
[12:19] as I was building the the apps, sometimes the model would make some assumptions on how Databricks app is working. Then, it would try to deploy it, fail, then come back to me, and I had to troubleshoot this. And this troubleshooting cycle, it might be,
[12:37] first of all, not uh cost-optimized because, you know, going back and forth with those models can be become very expensive at some point. Um and also, with Databricks Genie, we found that we have two x
[12:52] better success rate Wow. because Genie understands the data and your permissions on Databricks. So, sometimes my application would actually fail because the permissions were not set up. Right? But then, if you have utilized Genie, Genie understand, "Oh, you need to have a service principal permission.
[13:09] Let me provide this type of permissions." So, it really helps if you really think about, "Oh, this is what I want to build. I have this particular idea on Databricks." That's where Genie really shines because it's optimized for your own data. That makes, yeah, ton of sense. That's
[13:25] why Databricks Unity Catalog and this whole vibe coding, that's so central. Yes. So, exactly this. And uh with Genie, uh let's say the way that I see sometimes, if I have end-to-end where I
[13:41] want to create an application, I want to have um jobs, I want to create the dashboards, sometimes I would start with more more uh agnostic tool, let's say like code. I would create the specification file. It
[13:56] would get me the rough draft. And then, when I need to troubleshoot and when I need to start building those specifically optimized functionality, then I go to Databricks. And then, I will just try to solve all of those additional requests, troubleshooting in
[14:12] Genie, because first of all, it's at no cost, right? And you will just get much faster to your outcome. So, yeah, they're telling us just a few minutes left, but my this is my se- my favorite new chart for the whole
[14:27] day. This is like simplicity and genius. Yeah. Um so, everyone here, pay attention. Print this out, put it on your desk, um frame it. I love this. Yeah, so just to summarize, basically, if you are someone that is not
[14:44] necessarily familiar with coding, uh you don't have the developer background, tools like lovable or Replit like By the way, actually, Replit now launched so that you can deploy Replit apps on Databricks. So, Nice. um it's great way to get started. It's
[15:01] it will give you a lot of different functionalities. Then, if you find constraints, this is where you can have two choices. You can either go to um any AI coding assistant with Databricks AI Dev Kit and build utilizing your own
[15:17] favorite IDE, utilizing your own coding platform, or you can go to Genie and utilizing Genie code continue building on Databricks platform. And I almost certain that once you try Genie and you
[15:33] will see how fast and how much more efficient and how less troubleshooting you have to do with Genie, you will probably start defaulting more to Genie. So, I hope this was helpful. It it made a lot of sense to me and hopefully to everyone else. So,
[15:50] and anything else otherwise we'll we'll leave it at that. Um anything else? I think that that's pretty much it. Yeah, yes, if you are more familiar with coding, those are the two other tools that I recommend, but otherwise
[16:06] I think the most important things is really to um learn by doing. I am a huge advocate of learning by doing. So, those are resources for you to start. This is links for Databricks AI dev kit, links to the Databricks integration, and also
[16:22] you can try Genie code for free. It's available in Databricks free edition. So, thank you so much. you, Victoria. Such a pleasure to have you on. Appreciate everything you do. And

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.