Governed Vibe Coding: The Vibe to Value Workshop for Building Databricks Apps

Summary

Governed vibe coding: maintains intent, structure and standards while using AI to build apps.
Workshop app: choose a coding tool and build on Lakebase, Lakehouse or Genie spaces.
Agent skills: encode best practices to maintain governance while accelerating AI-assisted development.

Governed Vibe Coding: The Vibe to Value Workshop for Building Databricks Apps

Talk by Varun R, Sr Manager, Databricks
Vibe coding means building software by prompting an AI coding assistant rather than writing most of the code yourself, and governed vibe coding adds the intent, structure and platform standards that keep quality, security and ownership intact. This video introduces the Vibe to Value workshop, a four-hour hands-on session for building Databricks Apps, and demos the flow end to end: deploy the workshop app from GitHub, pick a coding tool such as Genie Code, Cursor or Copilot, define intent, generate a PRD, then build step by step with customizable agent skills.
Developed with the Databricks engineering community and more than ten customers over six months, the workshop closes the gap between creating software with AI and owning it in production. The live demo shows the workshop app deployed from GitHub: pick a coding tool such as Genie Code, Cursor or Copilot, choose a certified use case such as travel and hospitality, generate a product requirements document, and build step by step on Lakebase, Lakehouse, agents, Genie spaces or dashboards. Progress is gamified and tracked, and agent skills such as the AI Dev Kit and App Kit encode best practices your organization can customize. The video ends with how to sign up and bring the workshop to your team.

Chapters

FAQs

What is vibe coding?

Building software by prompting an AI coding assistant rather than writing most code yourself. Governed vibe coding adds the intent, structure and platform standards that keep quality, security and ownership intact.

What is the Vibe to Value workshop?

A four-hour hands-on session for building Databricks Apps using a coding tool like Genie Code, Cursor or Copilot. You define intent, generate a PRD, then build step by step on Lakebase, Lakehouse or Genie.

What are agent skills?

Agent skills are customizable modules that encode your organization's best practices and platform standards, so AI coding assistants generate code aligned with governance, architecture and quality standards.

How do I sign up for the workshop?

Deploy the workshop app from GitHub, available now. Choose a certified use case (travel, hospitality), pick your coding tool and work through the hands-on session with gamified progress tracking.

Full transcript

`00:00` Introduction: vibe coding in the enterprise

Good day everyone. Thank you for joining the session on "vibe coding" and best practices in this field. Ahem. I hope the summit is going great for you . I'll start with a simple question, or maybe imagine this situation. You are a corporate customer, and imagine that everyone in your organization has access to a super-sharp chef's knife. Everyone in your organization starts cutting. Your HR, purchasing, data analysts—they all start cutting very quickly because they have access to one of the best knives.
Imagine what will happen. From the outside, it seems like you're cutting everything very quickly. Vegetables are cut very quickly, but that's the problem. The problem is that no one knows the recipe. No one knows which cutting board to use. They don't wash their hands before starting, and worst of all, they don't know what they're going to serve for dinner. This is exactly how we feel when "vibe coding" and AI assistants are used in corporate environments to write code.
We need to understand how to change this to move faster with the right controls, get the most out of "vibe coding" while creating the real business value your enterprise customers expect . So, with that in mind , allow me to introduce myself. My name is Varun Rao Bhimavarapu. I lead the travel and hospitality business at Databricks. Today I'm going to talk about a framework we've developed within the Databricks engineering community, talking to over ten customers over the last 6 months, to create a framework: how do we apply best practices for "vibe coding" while still ensuring the right constraints, support, and governance?
As I said, it was designed with customers in mind. This was a problem that our customers were asking about , and we essentially built an entire framework around it. Unless you've been living in isolation from the world, you'll agree that enterprise software development has reached a tipping point. Our AI assistants are completely changing the way we build and deliver software. As they say, building software is now becoming much cheaper than discussing how to build it.
It's just reality, is n't it? So, according to Gartner, by the end of 2026, it is predicted that the number of "civilian developers" will be four times greater than the number of professional programmers. And 75% of all corporate code that will be developed will actually be created with the help of some kind of programming assistant. So, although the speed has increased, the real demand has not decreased. According to a recent survey, we see that the demand for data applications will increase 5 times by the end of 2026.

`03:39` Trust, quality and governance challenges of AI coding

Now, with all this pace of development, right? The main challenges for implementation are trust, quality and governance. And if you dig deeper, the question arises: how do I test what I'm building? How do I make sure the quality of what is being created is accurate so that I don't spend more time reviewing code than creating it, which is what is happening now. And the third is management, right? How can you be sure, because when using programming assistants, the area of possible attacks increases.
So, how to actually reduce risks, ensure security, and improve the level of governance in this area. So I want to share my opinion: creating something and owning something are very different decisions. AI definitely solved the first issue, but it didn't address the second, which was ownership, management, and actual deployment in a production environment. With that in mind, I want to introduce you to a framework we developed within the Databricks engineering community called wipe to value .

`04:58` What is the Vibe to Value framework and workshop

This is a framework that is delivered in a 4-hour workshop format, and anyone in the Databricks organization can help you run it. He provides three key insights during this workshop. How to write code with clear intent, right? How do you determine the right set of intentions for what you are trying to build? The second is going through the full life cycle from idea to real implementation in the gaming environment. So, to also maintain engagement.
And finally, how to apply Databricks best practices, using code testing tools and skills to ensure that everything you develop meets the governance and security standards you have planned. So, I'll give you a few examples—we've been running this program for the last 6 months. Here are some quotes from customers who have already tried it. At the end, I will show a QR code if you would like to register your organization for this workshop.
I'll just show you some of the feedback that clients sent us after the four-hour workshop. These are all the codes that the client has shared, including, for example, leaderboards. We are launching a leaderboard for user engagement, which you also get access to along with your CSAT score. So, if you are a leader in an organization and you say, “I’m struggling to figure out how to get teams to adopt this new development method while maintaining the necessary constraints? » This will be a great start to this journey.
So I'll move on to a live demonstration. I hope the demo gods are on my side. Good. First of all, this is an application that is already available to you on GitHub. It is available and is called "Whiteboarding workshop application". It is in a public repository. You can install it in your workspace and go through everything I'm going to talk about. Let me quickly show you how it works. When you deploy the application to your environment, you will see a workflow like this.

`07:05` Demo: deploying the workshop app from GitHub

Your team will be able to go through it and say, “ Okay, where do I start?” As you can see, there are many different tools for coding, and every organization has its own, right? You've probably heard a lot about the Genie Code that we talked about, especially during the keynote. So, whether it's Genie Code, Cursor, or Co-pilot, you can choose any coding tool you like. Let's say I choose Co-pilot. You pass the prerequisites.

`08:07` Defining intent: choose a use case and goals

And this is where the magic happens. This is where you define intent. As I said, the first thing to do is determine the goal of your case. What exactly do you want to create? You can come in here and say, "We're giving you a sample." Let's say it's a booking app that explains what characters exist. What are the main functions and features? You can also choose ready-made certified use cases, tested by our team of engineers and go-to- market specialists.
Right now it's for travel and hospitality, but we plan to expand it to all other industries. Each of these is linked to a set of goals that you can work on during the 4-hour workshop. You can also create your own goals. Here you can take, for example, an image. One of our clients took a screenshot of his business questions, uploaded it here, and the system automatically generated a goal that could be refined and used as a basis.
So, once you've chosen your scenario, you go through a cycle of defining exactly what you want to build. Are you trying to create an application? Do you want to build an application with Lakebase or Lakehouse architecture? Do you want to have a gold, silver or bronze level? Maybe you want to create multiple agents? Want to create a Genie space? Or an AI agent? Or maybe a dashboard? You are free to choose the path that suits you best.
Accelerators for creating end-to-end agents, Genie spaces, or even accelerators for data engineering are also available. For example, if I choose this option, the system will show what the overall logical architecture looks like. You can hover over each item to find out what it is . If your team is new and they don't know what a Databricks or Lakebase application is, they can quickly get familiar with it. And then you just move on, step by step .

`10:15` Generating a PRD and building step by step

The first step may be to create a PRD—a product requirements document. The system takes your goal and automatically generates a document with product requirements. Once the document is created, you can take each piece of it and transfer it to your development tool. This could be Coda, CodeX, Cloud Code, or any other environment. You get points for each step. As I said, it's a game process. So, your team starts interacting with this.
You have an interface development phase . This is the next step. And then you start to see what's happening. For example, I chose to build lakehouse- architecture, artificial intelligence and agents, and then you have the iteration and refinement stage, because you start with an intention , you build the first version, but you also need to refine it, right? You want to create your own restrictions or add a new brand.
This is how you improve this use case . I'll show you a few things. As I said, it's a gamified experience. So every step taken by your participants is tracked. Therefore, at the end you will be able to see who completed the training, what steps they completed, and what results they received. You can also see how agents' skills are used. For example, as I said, a lot of things are built into the agent skills, like the AI dev kit or app kit that we provide.

`11:28` Gamified progress tracking and agent skills

These are all skills that are tested by a team of field engineers. This is already built into this process. So, your employee or client may not necessarily know about these skills, but they can use ready-made best practices. I'll give you a quick example. In this case, you have these skills for building a reference design. Here are all the skills that are added to it as you go through this stage. You can look through them and see exactly what each one does.
You can view the source code. And here's the best part. Each organization will have its own enhanced version of this skill. For example, I have a skill that explains how to tag a specific lakehouse instance, but you may have a separate tagging policy or a different cost control policy. You take the basic version and expand on it. This is where the process comes in, which helps you use what is called a " skill builder."

`12:30` Customizing agent skills for your organization

You can say, "I want to be in the process of creating a skill instead of trying to build an app." So, as I said, you have different paths, and depending on your choices, the needs of your organization or team, you can choose the path that suits you. So before I close, I want to take you back to the kitchen analogy I gave at the very beginning. The fact that people had a knife and cut something with it. Now imagine that you could change this by implementing a certain process, a certain structure.
So, imagine this is creating freedom with a structure of how to actually take it and get the most out of it. So now people know exactly what the recipes are and what cutting board to use. Make sure they have the right cleaning supplies or wash their hands before starting, and ultimately know exactly what is being served for dinner, which is what really matters to your business. Here is the QR code. If you want to scan it, you can register for this seminar.

`14:04` How to sign up for the workshop

As I said, this is a 4-6 hour workshop, depending on the path you choose, and I look forward to your feedback . Before I finish, please take a few minutes to share your feedback, this is part of the survey. It also helps us become better. Thank you all.

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.