Sponsored by: Open AI | Build Working Applications Faster With OpenAI and Databricks

Summary

OpenAI Codex is an agentic coding assistant that reads context from across your tools — including a native Databricks connector for querying your lakehouse — and works through an iterative build-review loop that always keeps a human in the loop before any code is merged.
Codex can generate, iterate on, and security-scan code across the full product development lifecycle, and reusable Codex skills let teams codify repeatable work patterns so that common tasks across every sprint are accelerated consistently.
The combination of Codex and the Databricks Data and AI platform is shown live connecting to Databricks Lakebase, querying data through Genie, iterating on application features, and deploying securely to Databricks Apps — moving from idea to working app in a single workflow.

Chapters

FAQs

What is OpenAI Codex and how is it different from a regular AI chat assistant?

OpenAI Codex is an agentic tool for software development that operates across the entire product lifecycle — reading context from connected sources, generating code, writing tests, iterating on features, and scanning for security vulnerabilities — rather than just answering questions in a chat window. It uses a build-iterate-review loop that always surfaces work for human inspection before anything is merged or deployed, making it suitable for production development workflows. The native Databricks connector allows Codex to pull in live data and schema information from the Databricks Data and AI platform as part of the context it reasons over.

How does the Databricks connector work inside Codex?

The Databricks connector for Codex allows developers to query their lakehouse directly from inside the Codex interface, pulling in data and schema context to inform the code being generated. This video demonstrates Codex using the connector to connect to Databricks Lakebase, query data through Genie, and use the results to build and iterate on an application. The integration means developers do not need to leave Codex to explore data or validate assumptions about the schema before writing code.

What are Codex skills and why are they useful?

Codex skills are reusable, codified work patterns that teams define once and apply consistently across every sprint or project, capturing institutional knowledge about how specific types of tasks should be approached. For example, a Databricks skill for querying the lakehouse can be built once and reused by every developer who needs to connect to governed data without re-specifying the same setup steps each time. This video explains that skills are one of the most valuable productivity multipliers in Codex because they prevent repeated work and ensure consistent patterns across a team.

How does Codex handle security when generating application code?

Codex includes automated vulnerability scanning that reviews generated code for security issues before it reaches the human review stage, adding a security gate to the build-iterate-review loop. This video explains that this means developers get a pre-screened diff to review rather than having to audit raw AI-generated code for vulnerabilities themselves. Combined with the human-in-the-loop review step, this makes Codex suitable for building applications on sensitive enterprise data such as that stored in Databricks Lakebase.

Full transcript

[00:07] Good morning. How is everyone? I am here to talk a little bit about Codex and Databricks. Um and what some of the amazing things you can do with the Codex product.
[00:26] So, of course, complete your surveys. Very important. Uh so, I want to talk a little bit about how you can use OpenAI's Codex to build applications faster. And how you can also reach into Databricks and use all
[00:39] of the information, the data, the stuff you've got in Databricks to also accelerate the speed at which you're building those applications.
[00:51] These two products work incredibly well together. We have a native connector directly into Databricks that allows you to query information. I'll show you a little bit about that and how that works in a second. And on the OpenAI side, we have amazing models and technology that
[01:03] allow you to do everyday work faster, that allow you to build applications and code faster, and that really help accelerate every aspect of the job developers, data engineers, data analysts do within your organizations.
[01:22] So, Codex is OpenAI's agent for work. This is every kind of work out there, whether you are building code or applications, or when you are just creating a report or a slide deck or an Excel spreadsheet with information
[01:35] to present to someone. The interesting thing about Codex is it is the easiest tool you will ever start learning to use because at its heart, it really is just a a chatbot interface
[01:47] with a prompt box. And you can ask it how to use it. You can ask it to change its own settings. You can ask it how to do anything you would want to get done with Codex. So, if you take nothing else away from this presentation, know that when you sit down with Codex, when you
[01:59] start up this application and you're shown a prompt box, if you're not sure what to do, just ask and Codex will have that conversation with you. It'll work with you. It in many ways, it's like having an assistant that helps you get
[02:12] things done faster and that works for you and presents you with something that's ready for you to review at the end. So, when you're looking specifically at
[02:24] how Codex accelerates building applications and building software, this is where it functions across the entire product life cycle. It can help you with just about anything you want to do and when it's at its best, when it's at its most powerful, what it's doing for you
[02:37] is reading context. So, pulling information from anywhere. This is where you can get Databricks information. You can pull in information from Slack or Teams or whatever you use internally for uh chat
[02:49] for chat. You can reach into your email or your docs to read through requirements, to read through user research. Helps you go through all of that, analyze that information, assemble a plan,
[03:01] go out and actually build something, so write the code for you, test it, and then only present it to you when it's ready for review. And this is really where I I think Codex is is special and different and kind of a next
[03:14] step in agentic engineering, agentic coding because it'll execute this in a loop very, very competently and iterate until it gets you something that's of the
[03:26] quality you requested, that passes the gates you specified before it presents it to you for your review. And that's true whether it's code or slides. You can ask it for a slide jet deck. You can ask it to to review the deck over and
[03:38] over and until it reaches a certain quality bar, keep iterating, keep working until you actually get to that review step where you want to take a look and you want to decide what to do next.
[03:52] So, the way Codex works is it coordinates agentic work. So, this is not just a prompt response pattern where you're talking to a chatbot and you're engaged in regular chat. Codex will actually do
[04:04] multiple back-and-forths with the model for you in the background. So, you give it a prompt, you ask it to do something, the way it operates is it goes back through there and it will constantly invoke tools, get
[04:17] information, think about the information that it's received, go back and invoke more tools, do web searches, search through your email, whatever it needs to do in order to get the job done, constantly looping on that until it's
[04:29] gotten to a result that is ready for you to review or ready for you to take the next step. Now, in order to do this, it also has concepts like reusable skills. The The Databricks skill is one of those. Some of those are packaged with
[04:41] connectors that allow Codex to go into something like a database and find information, go into your email to find information. Some of them um are just skills that teach it how to do a repeatable task
[04:54] over and over again much better. Um someone described this to me what when I first started as almost like the scene in The Matrix where all of a sudden Neo learned to do kung fu by plugging in a skill. We've got over 100 different
[05:06] skills in Codex that do exactly that. Teach it to write blog posts better. Teach it to do data analysis better. Teach it how to do product design and product planning. Teach it how to build an iOS app. So, all of those things are
[05:18] things you can just go in, turn on, in some cases authenticate to whatever data source you want to pull from, and then continue moving forward, continue to to do the work you want to do. You can also take something you do
[05:30] collaboratively with Codex. So, let's say you write a report that you need to write every week and it's got a very specific format, you want to show share information in a specific way, you can iterate with Codex until you get the
[05:42] output you want the first time, and then ask Codex to create a skill that codifies everything from that session you just had with Codex. So that the next time, next week when you want to do that, instead of you
[05:54] needing to iterate and go back and forth, Codex will remember everything you asked for and exactly your preference for how you want that report presented, and it'll invoke that skill and do it in one shot, even if it might have taken five turns or six turns the
[06:07] first time to get it in the format you exactly wanted. So skills are incredibly powerful, not just the out-of-the-box skills that we're giving you, but also the skills you create yourself from your work with Codex and from the learnings that you have with Codex. And then you
[06:20] can share those within your organization, with other individuals, or with entire teams, so that you're really taking that organizational knowledge, that task knowledge, and spreading it more broadly. Then you can take something like that
[06:32] and fully automate it. So you can tell Codex, "Every week on Friday, I want you to run this process using this skill and generate this Slack draft for me that I then just hit return on to paste into a
[06:44] into the the chat." Uh So Databricks adds to that a whole a bunch
[06:56] of additional context that you have from all the work that you all the information that you already have stored in Databricks. So Codex can then take all of that information from within Databricks, it can help you analyze it,
[07:09] you can ask questions. You know, we use we use Databricks internally within OpenAI, and whenever I need information about what's going on, whether it's things as simple as business metrics, how are we doing against our goals, or whether it's deeper insight about our
[07:21] customers, what are they asking for, what are the most requested features, what are the issues that people are running into. You can get Codex to analyze all of that data for you, and either give you the raw data or give you
[07:33] a set of slides that describe a chunk of research it's just done across all of the information that you've got in your data lake, in your data in your data store in Databricks. Once you've done that, you can make a
[07:45] decision about what you want to build and how you want to build it. You go use Codex to build it, and then you review it, or you get your your team to review it, and then you you've got something really cool. You've accomplished something, and all of that is set up
[07:57] with the appropriate identities, permissions, human approval steps built in. In terms of get kind of diving a little bit deeper into the overall architecture, here you can see a little
[08:09] bit about what Codex does directly and what we can do using Genie inside Databricks and the interface directly into that. So, we have a Codex plugin for Genie and for Databricks. Single click, you can turn that on. You
[08:22] authenticate to your Databricks through all the normal gates, and then you've got whatever access you should have, whatever access you're entitled to, to get at that data, do that deeper analysis, really dive in conversationally with Codex to do that
[08:34] analysis. So, now I want to show you what this looks like in action. Um so, we've got a demo I'll walk you through. The scenario here is you work at a company called Blossom Bank. Blossom Bank um is just a
[08:48] a bank like many many banks out there that provides service to consumers, has its own interface on the web, has a mobile app. So, I'm going to talk a little bit about uh building a new feature with Codex for the mobile app
[09:01] for Blossom Bank just based on all of the the requests that you've gotten, all of the people out there who've asked for something very specific. And the idea here is just a a predictive finance model, a predictive banking
[09:13] model that consumers are asking for that this bank wants to provide built using the Codex uh app. As you can see right here, this is the this is the main Codex screen. As I was saying, you've got your list of sessions
[09:26] or projects on the left side. These can either be just general chats where you're having a conversation with Codex. It's it's working against your local files, but it's not attached to a particular project or a particular repo, or it can be something that's grounded
[09:38] in a repo where it knows that that's what it's working on. That's where it should get the files. That's the unit of of commit when it operates. Then down here, you've got the pane where you enter what you want to work on. So,
[09:51] let's say you go into Codex and you're working on this mobile app, it can pull up an iOS simulator so that it can test your application as it works as it builds things. You've got the ability to to have it go
[10:04] through and operate that app as if it were a user. So, it can do very thorough testing of the work that it does, of the features it adds, of the changes it makes. So, it's not coming back to you with code that might be broken or things that are half done. It's coming back to
[10:17] you with something that's finished, ready for a human to review, ready for you to add your input, your taste to the app that's being built.
[10:30] Here, you can say review the recent PRs, rollouts, what it what's happened recently. And Codex will then go through, exhaustively research, give you a summary of what's happened recently, what those PRs are, what those rollouts have been.
[10:45] You can ask it look at recent incidents and give you a nicely formatted summary of what's happened, what the resolution was, what the issues were, what the root causes were. It'll do the same thing. Just look through all of your data, look through your operational systems, find that information for you, present you a
[10:58] summary that's useful to you that you can read. Now, let's say you need to provide a brief on what you've done this week. It can also automate that for you. As I
[11:11] said here, we've got a skill that was created from the last time you created this brief. You just tell Codex to execute that skill and it can automatically go through and create that brief for you super easily.
[11:23] So, you'll you'll see depending on the way you have it configured. If you have it set up for everyday work, it hides some of these tool calls, particularly things like get calls or bash calls. If you're a developer and you want to see that, it'll show it to you so you see
[11:35] exactly what's going on, how it's working, what it's doing on your machine. And you can set it up so that it it either is asking you for approvals for each step or it's going through and doing this stuff fully autonomously,
[11:47] fully automatically. Um and then you might want to automate something so it happens automatically on a schedule at a certain time or based on a trigger. And then you can do that using this create automations window. And the cool thing about Codex as I said
[12:00] earlier is you you can use this window or I've actually never used this window myself. You can just ask Codex to create the automation based on the work you've done with a text description and it'll do that. So you don't even need to go in
[12:13] and work through a menu like this. You can just say Codex, please automate this so every Friday at this time you automatically generate this draft and it's ready for me in Slack so I can just review it, make minor edits and hit go.
[12:30] And then, you know, looking back at this, you can ask it to go query Databricks Genie directly via our plugin. It'll go find the information you need, do what it needs to do, and it uses that skill to to to make that request in a
[12:42] way that's super smooth, makes sense. It's got all the context it needs in order to explore that data autonomously without needing direct guidance from you on exactly what fields to query, what data to look for, how to find it.
[13:00] Go through this a little bit faster. Sorry about the the demo technical issue earlier. Um
[13:13] So now here's where you can use Codex directly to plan a feature two. So you can say go through this local repo, go through these data sources, and write me a spec. So you're not even necessarily writing that initial spec yourself. So one of
[13:25] the one of the really great things about Codex for software developers, one of the things they discovered very early on using this is it's not just for writing code. It actually can automate everything around the building part for
[13:37] you. So, it can help you with your research. It can help you write specs. It can help you write reports to other people within the organization. It can help with code reviews. And that's that's one of the other things I want to make sure we show before we're done,
[13:49] which is So, this This is where it'll it'll write it'll create the feature for you. It'll write the code, show you what it's done at the end.
[14:06] And then again, it can directly or bring this up for you to test yourself when it's ready for that step. Or it'll actually autonomously operate and test using an iOS simulator if it's a mobile app, using a browser if it's a browser-based app, using your computer
[14:18] with computer use if it's a direct app on your machine. Um and then actually, let me jump back to slides real quick. The other thing I was going to show in the demo that we've run out of time for a little bit is we also
[14:30] have the ability for you to in a completely autonomous way set it up in GitHub or in GitLab, whatever you happen to be using. So, it'll review every commit that comes in or every PR that gets issued and automatically provide
[14:43] its additional feedback for your team reviews and alongside team things. We even have a Codex security capability where it'll do a security-focused review on your each of your commits as you work so that you're
[14:56] checking for security issues and for security problems in your application code. So, just to summarize real quick, that demo took or should have taken if we were able to run through the whole
[15:08] thing, noisy signal, a bunch of information from disparate sources. It can go through Databricks to find information you have in Databricks. Can go through Jira or Linear, whatever you happen to use internally for project and
[15:20] issue management. Can look through feedback you're getting directly in in a customer support system, help you plan out whatever you need, build that feature, fix that bug, resolve that problem, and
[15:33] then give you as the person who's working something that is ready for your review, ready for you to take a look at, decide whether or not there are any changes that need to be made, and iterate, or if it's completely ready and ready to go, just say, "All right, this
[15:45] is ready, issue a PR, we're going to review with this with the team, and deploy, and get this out built and into production as soon as possible." So, Codex uses Databricks Genie, we talked about that a bit internally. We've got a native plug-in, native
[15:58] connector directly into Databricks, so that you can take advantage of all this really easily. You just at mention the Databricks Genie skill, Codex will reach out to that, explore your data, get you what you need.
[16:15] Uh and then, I don't think we have a final slide, so that Oh, no. It also will use the Databricks CLI, so it'll reach out and make use of that very very nicely. Codex and all of these agentic tools really love CLIs. It makes it easy for them to get at that information, use that information really really easily
[16:28] and really quickly and efficiently. Thank you guys so much for attending. Apologize for the demo issues. Uh and we really hope you'll give Codex a try. As I said, it's the easiest new tool you'll ever start to use. Just ask it if you
[16:42] have any questions about how to use it efficiently. You can even at the end of a session ask it to go through all of your interactions and say, "How could I have done this better? How could I have used Codex in a in a more efficient way?" And it'll tell you, it'll make those suggestions. So, really encourage
[16:54] you to to download Codex, give it a try, or stop by the OpenAI booth right over there, and ask any questions or or get a live demo right right there from from one of us. So, thank you so much for attending.

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.