Viktor Van Den Berg: Agents at Work: How GitLab Is Redefining AI Adoption for Engineering Teams

This talk covers the GitLab Duo Agent Platform and its capabilities in redefining AI adoption for engineering teams. The speaker, Victor van der Berg, explains how GitLab integrates AI into its comprehensive DevSecOps platform, streamlining the software development lifecycle from planning to deployment. The Duo Agent Platform features an ecosystem of AI agents that assist with various tasks, including code reviews, security analysis, and pipeline management. These agents can be customized to fit organizational needs, enhancing collaboration and reducing context switching between tools. Through live demonstrations, the speaker showcases how AI-driven features can automate workflows and improve overall productivity in software development.

Okay, welcome everyone to this next session. Welcome. We're going to talk about GitLab agent platform today. The title of this talk is agents at work. How GitLab is redefining AI adoption for engineering teams. My name is Victor van der Berg. I'm a solution architect at at GitLab and I'm very very happy that you're all here. Before I get started, a little bit of a disclaimer so you're

aware of that. Let's not spend too much time on this. So, what we're going to talk about today. We're going to talk about AI, about Duo agent platform. We're also going to talk about what's GitLab in in what is GitLab? Maybe a good thing to start with and if you look at this slide, maybe you recognize this this this. So, many organizations, software development looks like this.

Multiple tools that you let's say stitch together, that collaborate, that deliver your software from the idea to production. But all these tools you need to manage those different tools and that can be can be quite challenging. There's maybe also some overlap there, right? You need to switch between those tools so a lot of context switching. And also, it can be pretty challenging manage all these tools. Also

like from an governability perspective, but also if you look at security for example, integration challenges. So, this might be a challenging way to build software. So, it's another way to do it. Well, that's how we as GitLab look at things is with our comprehensive AI-powered DevSecOps DevSecOps, excuse me, DevSecOps platform. So, what does that mean? So, what we have is a platform that covers the full software

development life cycle. So, all the way from idea, from the planning of the work, all the way to delivering the software at the end, that is all covered by one solution, and that is GitLab. So, GitLab is focusing on facilitating the full DevSecOps cycle. That's what what we are doing. So, not a patchwork of different tools, just one solution that does everything for you. Of course, we

are also able to integrate with external solutions. So, for example, if you're using Jira Jira for planning, we also integrate with that. But, you can also do everything with just with GitLab, right? And what we have, and you see that on the slide as well, is a single data store that contains all the information about your projects, the applications you're building, your source code, how you organize

your work, security, deployment, everything is in there. And that's AI-powered. So, AI is not like a add-on solution that you need to configure separately and install separately. No, it's a foundation of our solution, so it's integrated into the GitLab platform. And the big advantage here is that the AI AI solution has access to this single and understands the broad context or the complete context of how you

are doing software development. Um so, we give a lot of context, we provide a lot of context to our AI solution, and AI is also working not only on code generation, but is available in all the aspects, all the different stages of the DevSecOps cycle. Yeah? The AI AI solution that GitLab is providing is called the Duo Agent Platform. It's a foundation of our platform. So, it's

integrated again, integrated into the DevSecOps platform. It consists of an ecosystem of what we call AI agents that you can use in your daily work. You can ask questions to and also something that we call flows. And a flow is a let's say well, standard workflow that an AI agent is tasked to do a specific task, is focused on a specific goal and will work towards that

goal. And I will talk a little bit on more a little bit more on that in a in a couple of slides. The whole AI solution that we have is fully customizable uh for your organization. So, you can customize how AI is working. That's what you can do. Uh but you can also like add new custom options to the AI solution. So, you can add extra agents.

You can define new extra flows. And what's also interesting to note is that it orchestrates seamlessly with any other AI development tool that you're using maybe as of today. Cuz that might be a question at this point. Okay, we heard a lot about AI. We heard we see a lot of different AI Why is this different? So, if we look at existing AI tools and there are

a lot of them available. Um many of those tools are mainly focused on let's say code generation. Building capabilities, building features. That's what you do with with these tools. And that's of course very valuable. Different tools are very capable of doing that. But the question is also from where is the the real productivity gain? Is it just only about code generation or is there more where we

should think about? And that's where how we look at this. So, if you look at software delivery, it's not just only about coding, right? The coding part is in yellow on this slide. AI has made it easier to code, but the process around the coding, that's also important. And here is where Duo Agent Platform, our AI solution, can help you to uh accelerate software delivery. Right? So,

if we we are talking about planning the planning of your work, with GitLab Duo Agent Platform, AI is there. Of course, the coding, you can use our tool or maybe you want to use another tool. And then we're talking about creating the merge request, creating the commits, security scanning, uh managing and troubleshooting your pipelines, um all the work that needs to be be done before you can

deliver the software, that is where we see a role for AI. And it's not again not only important to accelerate coding, actually to accelerate the full software So, how do we do that? So, again, we have our platform, the GitLab platform, managing the full SDLC. GitLab Duo Agent Platform is integrated into that. Yeah, so it's a layer inside of the platform. It consists of different agents, agents

that you can solve can consult can ask questions to. Uh we have uh um let's say generic agents, but we also have what we call specialized agent that have a specific expertise. For example, we have a security analyst agent that is um very good at um solving or um exploring security vulnerabilities, explaining security solving them, for example. We have a specialized agent that is focused on planning,

for example. So, you can ask how to plan your work, and the agent can help you with that. Right? So, different agents that can help you with specific questions, specific tasks. Another important construct in the platform is flows. I already mentioned them. So, a flow is more like a automated workflow that can be triggered inside of the GitLab platform that is tasked with uh or that is

focused on delivering a specific goal. That's what a flow is doing. So, for example, very popular flow is the code review flow. So, what's that doing? Code review, of course. So, that means if you commit to a merge request in GitLab, automatically the code review uh flow is uh triggered, is executed, will do the review of your commit, and will provide you with feedback almost instantly. Takes

you maybe a couple of minutes. So, that's an example of a flow. Another flow is, for example, when a pipeline fails in in the platform, then you want to fix the pipeline. So, we have a fixed CI/CD flow or fixed CI/CD pipeline flow, and you can trigger that flow to solve issues that you have in your pipeline. If your pipeline is fai- failing, you trigger the flow,

and the the the the AI will try to solve the issue that you have, create a merge request, and it's up to you to approve the change, and to uh well, actually merge the merge request, because that's with GitLab Duo Agent platform. AI can act autonomously, but there's always a human in the loop. You need to any changes that AI wants to do to your application, to

your environment, right? That's important to understand. GitLab Duo Agent platform is available on GitLab. We have different deployment options. If you are using GitLab, you can use gitlab.com. That's our SaaS option, so you can go to there's a free option there. You can create account yourself and start start exploring the solution. Uh of course, gitlab.com has Duo Agent platform uh available. But next to gitlab.com, the SaaS

option, we also have a dedicated option. So, what's that? That means that you are running GitLab in a private managed service. This managed service is delivered on Amazon. So, you have a private tenant on AWS, and we provide you with a managed GitLab instance. That's another option. And third option, also very popular, is the self-managed option. So, it is also possible to download GitLab, install it in

your own data center or in your own cloud, and then use it. For the different deployment options, we also always provide um the AI solution as well, Duo Agent platform. So, you can can use it in those different uh deployment options that we that we have available. So, what does this look like? You have GitLab, you have Duo Agent platform. And then you can say me like,

GitLab certifies the work of your coding agent. So, we have our coding agents. Maybe you Duo Agent platform for that, because we can also help you there. But maybe the work is coming from other agents. It's input to to the platform. It input to GitLab Duo And for example, DAP Duo Agent platform can do the code review for you. Can take care of pipeline repairs. Can take

care of security remediation. Can take care of the planning of your work. different scenarios where you can think of and of course, yeah, the options are unlimited because you can also create your own agents, your own flows, and yeah, create whatever you want, right? A lot of a lot of options there. So, our customers are using uh Duo Agent platform as of today and we see some

uh let's say popular use cases, top use cases. So, what are those use cases? So, first of all, like Agentic chat, just your let's say coding partner, the the chat option that we have available. It's everywhere throughout the product, in the interface of GitLab, either in the UI, in the web UI. It's available in the IDE. But, we and we also have a CLI available, so that's

one Um automated code reviews is a very use case that we see. Fixed pipelines are related or I already mentioned that. Uh then we have something that's called the software development flow. So, let's say you have a a issue, work item that describes a certain feature that you want to develop. Then there's a very or there's just a button in the issue that tells of the or

that says says issue to MR. And that means that AI will read the issue, will see what you're asking, and then will just build the feature for you. You will get the result in a new merge request. You can test the feature, maybe change it a little bit if it's not completely what you expected, and then merge the new feature into your into main branch. And just

use it. Uh another use case that we see is converting pipelines. That's also something we can do with AI. So, this is about customers that are maybe are moving away from for example, Jenkins, CircleCI, have an existing pipeline there, want to convert it to the GitLab format, AI can help you there as well. And the security use case security remediation, you can use AI to well, analyze

your vulnerabilities, uh do triage there, maybe explain vulnerabilities, resolve vulnerabilities, those kind of things that is also what we can use AI for. Different use cases. Okay, great. I think enough slides. Let's go to a live environment and let's see what this looks like in a real GitLab environment. So, let me uh change my screen settings here. Give me 1 second. Yeah. Super. And here we are.

Move this a little bit. So, here we are on gitlab.com. This demo is on gitlab.com. Uh this is a a demo project that I'm running. I'm uh I'm actually um quite a fanatic runner. I like to run, so that means that I need to record all my activities into Strava. Excuse me, [clears throat] into Strava. So, that this is a little bit of a >> [snorts] >>

test project based on Python that is interacting with Strava API. And it gives me some information on what I've been doing. Uh it can create some statistics, some nice graphs, and of course, I don't know if you are on Strava as well, but the most important thing of course on Strava is that you get as much kudos as as you can get. So, there's like a little

bit of a kudos graph. So, great stuff. So, here is my my project. I'm here on on gitlab.com again. And here on the right, I have some options available and I can easily click this chat box open. So, I'm now able to interact with the chat agent. And for example, I can ask a question, "Please tell me what this project is about." Duo Agent platform is is

context aware. Context aware in the sense that of course it understand the concept context of the projects that you're working on and all the all the constructs that are there, but it also understand the context as in "Where am I in the interface?" Right? If I'm looking at a merge request or at an issue and I'm asking question about that issue or merge request, it understands that

okay, you're looking at this merge request, so you want more information on that. So, it's not telling me yeah, okay, this is a fla- Flask based Strava integration dashboard. And it tells me some information on on what this is doing. Great stuff. So, this is the UI. Of course, I can also also go to my IDE. So, in this case Visual Studio Code. And here on the

right, I have the same yeah, chat box. Uh me clear that. Oh, sorry. Not clear. Re- Oh, there we are. Reset was the command I was looking for. And [snorts] reset. Give me 1 second. Open it again. You know, think it's just working. So, "Who are you?" And it is just responding. So, this is straight straight from the IDE. I can also use it there. The only

thing that I I need to configure or install is the GitLab extension into the IDE and then I also have access to uh to uh Duo Agent platform. And the third option is uh is on my command line. There is a Duo CLI that I can use and this is also connecting to Duo Agent platform. Uh we also have an integration with Open Code. So, if you'd

like to use Open Code, we can also uh or you can also use that as an interface with uh with Duo agent platform. So, for now let's go back to the uh interface here. Uh I've asked what the maybe I want some more information, some more extensive information about this project. So, what I have done is I have also created uh some work items here. So, these

are yeah, let's say my issues, my tasks, the things that I need to do. And you see here a option latest onboarding information. So, what you see here is that I'm asking here to a specific This is a an AI developer onboarding documentation generator I created myself. So, this is a customized option. Hey, give me a um um um give me some information on this this project.

I'm a new developer. I want some onboarding information. And this is the following a predefined prompt that I have put into the system. And this prompt is creating more extensive information on this specific project. Right? So, now I get more information on how to get started, how it is being built. It's using Kubernetes. It's using Python. It's using Flask, how it's relates to each other, etc. etc.

right? So, this is another option what you can So, what I also can do then is ask, for example, open the chat agent and now Sorry, I need to open this one. So, I just opened the generic agent, right? For generic knowledge. There's also a planner agent. And a planner agent can help me with planning my work. So, I have my work items in this project. what

tells me what I should be working on, but I can also ask the planner agent, "Hey, I have 4 hours available. What would you recommend me to start working on based on existing work items, right? now it will just explore the work items that are available, and it will tell me or at least help me from hey, maybe you start working on this or that or whatever.

So, we leave that for now, just to give you a bit of an indication how that works. Comes back with some recommendation, hey, you can do this, start working on that, etc., etc. So, let's have a look at the work items. I just talked a little bit out about the issue to merge request or also sometimes called the development workflow. Um so, here we have a task

and that says implement dark mode with a team toggle. And this is a description of a new capability or a new feature that I want to build. And now there is a very easy way to build this because here you see generate MR with Duo. And that's all what I need to do. So, what this will be doing is analyze what's in the issue and then start

building this specific feature/capability. This will take a couple of For those of you who know GitLab and are using GitLab, this will spin up a runner in the background, and that runner will will execute the um yeah, let's say the flow, the workflow that I'm I'm using right here, and it will build based on the requirements I've described, will build this feature for So, we're we're not

going to wait for that because it will take a couple of minutes, but of course I prepared this demo. So, if we look under code and then merge requests, code and merge requests, yeah, there we are. We will see that this feature has already been implemented before because I ran this yesterday evening. So, here is my dark mode implementation. And this is being built by the agent

So, it tells me what the changes that it made, the files that it changed. There are two commits apparently, etc., etc. Okay, that looks looks good. But maybe I will still want to probably review what's being built. Can also use AI for that, right? Because then I go to reviewers here. I click edit. I type GitLab Duo, And now what will happen is that GitLab Duo will

also help me with reviewing the changes that are being made in this case by this developer flow. Of course, if the commits were coming from a colleague, you can also start this review. You might be wondering at this time, "Okay, that's that's nice that GitLab Duo is doing the review for me. But can I customize that?" Well, the answer to that is yes, you can customize that

because in the project, in the repo, you will find a folder called dot GitLab called dot GitLab {forward slash} Duo, and this contains a MR review instructions YAML file. And this file contains the review instructions the code review that GitLab Duo is doing So, think about it, right? Normally, when you do a code review, you need to wait for someone who is available to do the code

review. The lead time maybe takes 1 or 2 hours. Then the person needs to review the code, eh your colleague may may may cost another or may take another hour, 2 hours, and then you have your result, right? And with this, it only takes a couple of minutes that you get feedback on on the changes that are being committed to the to the merge request in this

in this example. Um so, this is running right now. Uh again, to speed up the demo a little bit, I did the review already on on another merge request, that's this one, so we can quickly see uh what this will look like. By the way, what you also see see here, right? I used uh Cloud Code in this case to build this feature. I have different options

there. I can also integrate with Cloud Code and ask from the GitLab on the same like issue that we just saw, the description of the feature that we're building, I can ask Cloud Code also to build it, and that's what you see here. Uh and what we also see here, here is the review. So, I did that 2 hours ago. What you see here is I reviewed

this merge request and um uh I left several recommendation addressing code quality and best practices. what you also see here, maybe let me enlarge it a bit a little bit more, zoom in a bit more. According to the custom instruction in CSS review, so you see here that code review uh using the MR review instructions YAML file as the foundation, as input for the actual review of

the code that has been committed, right? So, that's another use case. Another thing that we can use uh Duo Agent Platform for is for example to fix a pipeline. So, here we have uh a merge request that has a failed Uh here's the merge request, here's the pipeline. You see a big red cross. So, then there is here the Tanuki icon, fix pipeline with Duo. There's also

a button over here. Click And now Duo Agent Platform will help me analyzing this pipeline, what went wrong. And it will either come back with a let's say recommendation how to solve the issues that we're being faced, or it will even, depending on what went wrong, create a new MR for me and propose some fixes how to fix this specific pipeline. Yeah, that's another So, as you

can see uh yes, we can help you with code generation, building features, but yes, we are also taking care of other stages in your software development life cycle. Also, for this specific demo, I prepared it. Um and what you will see, where is it? Here. We have a new merge request. It's 2 hours uh old. I ran it 2 hours before Uh what you see here is

that it's telling me, "Hey, I see that the pipeline failed. I think this is uh going wrong. I propose this change. Please check if this is what you want, and then commit it if you agree with it." It is blocked because this merge request is in a draft state, so I need to change that, but then uh this will solve my uh uh, pipeline issue that I

have here. Yeah? So, different options there. This These are all workflows and agents that are available out of the box. You also have the option to define your own agents flows. For that, we provide also an AI catalog where you can store those agents and So, for example, if we look here under agents, you see enabled. um, let me see. The planner agent is there. Security analyst

agent, all with a the little the tanuki icon behind it that are what we call foundational agents provided with the product included. But, if you look under managed, I also have two other agents. For example, a Kubernetes best practices advisor and a readme agent that create readme files for me. Yeah, different options there. Uh, the same count for flows. You can also, uh, we we have out

of the box flows available. Um, for example, the code review, fix the ICD pipelines, but you can also define your own flows like the developer onboarding document that I showed you. Again, different options there. Okay, we have a few minutes left. So, let me, uh, go through the last couple of slides that I have available for [snorts] you. I think already talked about this. So, this is

the the flows, the code review, fixed pipeline. But, also like SAST false positive detection can help you there. It save you a lot of time again, right? So, I was talking a bit about triggers. So, triggers is starting flows, so AI flows in a automated way. And this is how we see that you can orchestrate AI in in Uh so, a trigger based on a code review,

that is a trigger, but also if a pipeline fails, then you can trigger it using a fix. Uh or you can trigger it if the pipeline fails and run the fix CI/CD pipeline Other triggers are coming soon, and this will make AI even more automa- autonomously uh in a sense that you can trigger it based on events that are happening in the GitLab platform. Yeah? So, that

that is what you see on this specific slide. >> There is a lot more to talk about, so if you would like to hear more about this, maybe you want to talk about customizations, agents.md, maybe you want to talk about skills, we fully support that. Or you want to see how this makes sense for you. yeah, we we really like it if you stop by stop by

our booth. Our booth is right over there in the back of the of the room. I'm together there with my colleagues. If you want to learn more about GitLab in general, we have some good content online, of course, specifically about We have a prompt library, that's quite interesting quite interesting. So, the prompt library contains a lot of prompts that you can use to yeah, enhance your AI

layer in GitLab, right? That's what this is about. and then there's a nice interesting website or webpage about the top use cases that we see uh um at at our customers that that that are using Duo Agent platform. Uh that's about it. I think we're also running out of time. So, if there are questions, please come over to our booth, we're very happy to talk to you.

Thank you very much for being here today and I hope this was a useful spend of