When 75% of Your Builders Aren't Engineers: Enterprise Vibe Coding Meets Databricks

Summary

The Replit and Databricks integration enables non-technical business users—including HR, sales, and RevOps teams—to build governed production data apps on Unity Catalog data, with 75% of integration users coming from non-technical backgrounds.
Apps built in Replit deploy directly to .databricks.com with workspace-level OAuth, Unity Catalog permissions, data masking, and full platform-team observability, giving security teams the trust model required to approve enterprise deployment.
A customer case study in this video describes a realty company that achieved 10–15x revenue growth by enabling non-technical staff to build and ship governed apps on enterprise Databricks data.

When 75% of Your Builders Aren't Engineers: Enterprise Vibe Coding Meets Databricks

The Replit × Databricks integration was built on a bet: that non-technical business users, not just developers, should be able to build production data apps on enterprise data. That bet is paying off. 75% of users building with our integration are non-technical, and they're shipping governed apps that deploy directly into Databricks with full Unity Catalog permissions, data masking, and platform-team observability. this video covers what happens when you put vibe coding tools in the hands of LOB and data teams at Fortune 500 companies. We'll discuss the architecture that makes this possible, the trust model that convinced security teams to say yes (apps deploy to .databricks.com), and the adoption we are seeing in leading companies using AI tools with a data foundation set in Databricks.
Talk By: Asif Bhatti, Head of Product Partnerships, Replit ;

Chapters

FAQs

What is the Replit and Databricks integration?

The Replit and Databricks integration connects Replit's AI-assisted app-building platform to Databricks, allowing users to build full-stack apps that query live SQL warehouses and Unity Catalog data. Apps deploy directly to .databricks.com with workspace-level OAuth and full Unity Catalog access controls, without requiring engineering team involvement.

How does the Replit integration maintain security and governance?

Security is maintained through Replit's own code scanning, deployment to .databricks.com with OAuth authentication, and Unity Catalog access controls that enforce the same permissions as any other Databricks workload. Platform teams retain full observability over every app built and deployed by non-technical users.

Why are 75% of Replit and Databricks integration users non-technical?

The integration was specifically designed to empower business users—PMs, analysts, RevOps and HR leaders—who have ideas and data access but cannot write code. By abstracting the engineering layer, Replit enables these users to build apps that would otherwise require weeks of engineering work, reducing the bottleneck on technical teams.

What customer results are referenced in this video?

This video references a realty company customer that achieved 10–15x revenue growth by enabling non-technical users to build governed apps on enterprise data using the Replit and Databricks integration. Early adoption results from Fortune 500 companies are also discussed as part of the public preview launch.

Full transcript

[00:08] Hi everyone. Um I'm Asif Bhatti. I help lead up our product partnerships here at Replit. And there's something really fundamental changing about the way that enterprises are now enabling their lines of
[00:20] businesses to experiment and innovate. If you look at the labs as an example, they have research teams repeatedly running on different problems from a technical standpoint, but that
[00:32] experimentation allows them to operate like a lab. But if you think about businesses that aren't like hard sciences or AI research labs, how can they experiment in a way
[00:45] that then allows those budding ideas to become big lines of businesses, big product areas. So that's where we're what we at Replit are starting to see. Our partnership with Databricks has
[00:57] really been one that of making sure that the other 70% 75% of the organization that historically doesn't build and has ideas is empowered to do that very
[01:09] quickly while being governed centrally. So I'm going to hope that this work All right, great. So I think the real thing is is like the pool of builders that we
[01:21] now all are working with has just increased four times. And so that's PMs, that's analysts, that's rev ups leaders, HR leaders. In fact, I was actually working with
[01:33] uh one of our customers here at Replit and their CEO goes directly to the HR and finance analysts and asks them, "Hey, I want you to build a dashboard that reports all of these different
[01:45] things." And for that analyst, they may or may not have the right credentials to access that data, let alone even build on top of it. And so when we are starting to see that happen, the real power is is that now you have all of
[01:57] these non-technical folks being able to build and generate and not do that in a shadow environment. Not Not shadow apps, not shadow IT.
[02:12] So, what the old world, I guess, shows is that there is a bottleneck because your technical team, your engineering team, your data science teams, those are the ones that are going to get all of these requests coming through. Whether that's
[02:24] inventory management, a portal to get tracking done, a CRM sync, forecasting model. They're happening in spreadsheets, they're happening on BI dashboards, and there's not really an
[02:36] easy way for the technology organization to manage it manage the inbound, let alone actually be able to produce the outflow of like a revenue dashboard, an exact dashboard, or an op portal. And
[02:50] ultimately, what ends up happening is either the the squeakiest wheel or the loudest voice in the room ends up getting these resources to build the things that leadership or your team needs. And so, what
[03:03] we are working is to how do we enable we at Replit with Databricks are enabling it is how do we make sure that this drowning that happens is not bottlenecked by a single team, yet the technology organization has a single
[03:17] way, a single point to govern all of the applications that are being built. And really, really manage not only the controls of which users are logging into those applications, but also how that data gets surfaced, interpreted, etc.
[03:35] So, if we move forward, like, what is this new reality? The new reality is is that your HR team, your sales teams, your RevOps teams now are builders. They're
[03:47] contributing to applications in your technology stack in in your technology organization and your technology stack. They are um experimenting much faster because they're closer to the customers, closer
[03:59] to revenue streams, and in that experimentation, they can actually glean very quickly, "Hey, the this tooling that I've built or this this process that I've redesigned is actually a better way to drive top-line growth."
[04:12] Or, said differently, say drive cost savings as well. And we see that in in a couple of different examples. We have a uh a a large realty company a largely focused on
[04:25] um digital realty management, and they have now used Replit as a way to empower their field teams, non-technical team members, to build the best way to route online leads to local realty uh
[04:39] realtors, and then optimize those realtor outcomes based off of the zip code, the the price of the home, the types of home they've sold. And it's driven 10 or a 10 to 15-fold, I think, last
[04:51] time I checked, increase in their top-line revenue from 10 million to now 50 million. So, what the big thing that ends up happening now is your time to delivery is increasing. Your cycle in which your
[05:03] organization can learn gets faster and faster and faster, and which in a world with AI, that's really, really important. So, what does this mean and how does this work? So, I'll go through the
[05:16] details here in a little bit, but in Replit, what Replit allows you to do is have a si- simple IDE live coding interface to start taking your words and building applications
[05:29] from them. In fact, this slide deck was built on Replit. So, I you you can just describe whatever you're thinking. And for this conversation, I was like, "I've worked with my Databricks teams in Slack
[05:41] channels, in uh Gmail, and we have our Google Drive full of files. And how do I pull that together to share what we're working on? And in fact, I didn't really pull it
[05:54] together. I asked my Replit agent, "Hey, connect to all these sources, and pull all the information over, and let's talk about what this storyline, what this document should look like." And it took the life of what you're seeing right
[06:06] now. But, when we do that, in this instance, I'll ask, "Hey, build a revenue dashboard on our Databricks data." And so, our agent is going to be able to help take that, understand and
[06:19] decompose it as to what resources do I need to provision. I understand it's a dashboard that's going to live on top of Databricks, so we'll constrain the Replit agent such that it doesn't do things that would not work in your
[06:32] Databricks environment. We would leverage our connector to actually start to port over a masked data set that your data your Databricks administrator has set, and then use the IDE, the the Replit
[06:47] interface, to build the solution, the application, the dashboard that you're looking for. And with that, what ends up happening is that your data is never leaving your Databricks environment. It always stays
[07:00] there. All that the Replit agent need is what the Databricks administrator has allowed, whether that's masked data or schema. And so, if we keep going, what that then starts to look like is
[07:12] you have with Replit a one-click deploy. You don't have to think about if you're using like a CLI to take that package and um figure out, "Hey, I want to upload it into this hosting environment or deploy it into this." With Replit, your
[07:25] non-technical builders do not have to take on that cognitive load. Your engineers, on the other hand, may want to do that. They may want to be a little bit more um prescriptive with how an AI agent is
[07:38] building. But, for your non-technical people, for them to be able to just say, "Hey, click. I'm going to send it over." and then your Databricks admin gets to see all of the apps that are running, being used, how to govern them, where
[07:51] they're living, what they're doing. And the benefit of that is now again, going back to the example of HR, of sales, of your RevOps teams, they're not going to know how to engage in a terminal or how
[08:05] to set up a database properly or what security guardrails that they should check before they publish something. And so with this, you can one-click deploy into a a staging environment on your end or
[08:17] wherever you want to pull in all of your non-technical builders' applications, and as a Databricks administrator then govern them in the most effective way. One One is of course what information's going in and
[08:30] out, but then two is are there redundancies in applications as well. And so what that can look like then is your lines of business as opposed to static slides, they have real dashboards, slides or applications or
[08:45] mobile mobile apps that host real information, and I never have to They never have to or I'd never have to rebuild any of this again because it'll always live on top of the the data
[08:58] environment that is being governed. So, how does this actually work? Um so our deployment architecture is two-pronged. Of course, in Replit, you're going to build it and describe
[09:11] it, but the way we do that up front is your administrator, your Databricks admin, is going to provision a connector on the Replit admin side. That connector is going to be It'll have a client ID and a client secret, and then that will
[09:24] establish for your entire enterprise how information should flow from Databricks over to Replit. And we have two types of connectors there, machine-to-machine and user-to-machine.
[09:37] That means nothing to a lot of people. Machine to machine, think about it this way. If you're going to have a single dashboard that if anybody in your organization logged in, or a single application that anybody logged in, they will see the same thing. The same set of
[09:50] data sets. But user to machine, U2M, is actually a lot more interesting because now your sales leader, your CEO, can go to the IC analyst in a fully different org that
[10:03] doesn't have access to say salary data or um sales revenue data, and they can still build holistically for that executive without actually getting the ability to
[10:16] see that data set data itself. And that's all happening within that top layer. It's they're building with the the the agent, the rapid agent with is building with the knowledge that um this
[10:28] is what that data data environment looks like, and when a user logs in after deployment into this application, that it is based off of their credentials that the dashboard or application or slide deck
[10:41] should populate. So, what that starts to materialize is that it when it's living on Databricks, your admins now have visibility into the the
[10:53] into the inference load and compute spend that those applications are taking on. If there's like an AI chatbot, if there's other things that are enabled on that application. It also gives them the ability to reconcile if
[11:06] there are redundant applications cuz a lot of times you're not the only one that had the idea, but you might be the one that executed on it. And so, from a Databricks standpoint, the benefit really is that you get governance not
[11:18] only on what information is pouring in and out, but then also how to reconcile all the bill that is happening across your organization. So, security is really really important. Um
[11:31] so, there's couple of things that happen on um on our end as well as on Databricks end. So on our end when you deploy on the Replit side and you're porting it over to Databricks, we run a series of scans and checks that are not only bespoke to
[11:45] Replit, but also through some of our security partners that we've brought in to make sure that applications are not sending information or having client secrets that are sensitive. So that is one scan. When you port it over to
[11:58] deploy into your Databricks environment, then it start your Databricks governance layer within Databricks kicks in. It's quite simple. The deployment URL will be .databricks.com. So now it's behind all of your OAuth
[12:12] layers, your uh your Unity Catalog controls for governance. Your data stays put and you have full full observability as the administrator on how information is flowing in and
[12:24] out. Who's logging into the application? If it's the executive, they can see it. If it's the analyst, they cannot. And I think a lot of folks think about vibe coding as this thing that's going to spiral a lot of different tools,
[12:36] create shadow IT. I mean AI can as we've all seen, have garbage in garbage out. But because of the ways that we've worked with Databricks, we've constrained the build motion on our end such that you're not able to have like
[12:50] garbage in or garbage out. Like the build won't work. You need to have more questions, more clarity from the user before our agent starts building. Similar to shadow IT, it's not. It's going into your Databricks environment
[13:03] and so you're governing all of those applications singular or in a single spot. So what's really really exciting for us is we've been working at this problem now with Databricks for I think 2 3 months
[13:16] to try to pull this all together from an integration standpoint. And we've had an amazing run with a series of early partners that have started to use this. And what we've seen with those partners is that when they enable the non-technical organization to
[13:29] experiment, to take their ideas and actually build with them, their ability to shape and direct and respond to the market as a business is much faster. And
[13:41] we I think launched in early an early preview of it 3 months ago had an immense response of it, but today we're going a lot of that business and
[13:53] I'm super super excited. Today we're opening all of this up for public preview. So, if any one of you all are using Replit, it is now available for you to try, to get your line of business
[14:05] to build on Replit and then deploy into your Databricks environment. There's no waitlist now, there's no setup required. I would encourage you any and all of you if you want to see this actually come come to life to visit us right down the
[14:19] expo floor. We're right by where the puppy park is and so you can visit some of the pups and then hopefully veer your way over to to our team. But, with that I think
[14:32] when you enable the other 75% you're going to find a lot of immense value for your business quickly coming and this integration I think hopefully now brings your technology leaders along the ride
[14:44] to make sure that any of these those applications are safe, secure and that data then lives where it's supposed to live. So, thanks everyone.

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.