David DeSanto Breaks Down GitLab’s Latest AI Innovations

This is Textron tv. Hi everyone. Welcome back here to techron tv.

You know, I, I feel like I just saw this next guest, but it wasn't, it was actually almost, I bet, a month ago already. Uh, David DeSanto, chief Product Officer at GitLab. David, welcome to Text Drunk tv.

I, I feel like we just spoke, right? We did. It was about, maybe about a month ago, but yeah, it's great to see you again.

Yeah, It's good to see you and it's good to see you home rather than in your, uh, road warrior mode right where you spent most, most of your time. So, yeah. Good to be home.

Um, David, it was a month ago, and then, you know, folks who follow us on Text Drunk tv probably, so it can't see if you go up on text Drunk tv. I followed up our conversation with you at CubeCon by, uh, speaking with Ashley Kramer GitLab, uh, CMO Chief Strategy Officer over in Vegas. And that was last week, I think, uh, AWS reinvent.

Um, it was, so, you know, we've been getting a steady stream of GitLab, but it's at a good time because you guys got a lot going on here. Um, we do, I I want to jump right into it and, and talk about Duo and, and what you have going on, but before we do, I learned something last week that it's no longer the 22nd of every month that we release software. It's now the third Thursday Correct.

Of every month sort of a Thanksgiving thing. Yeah, well, we did that. Yeah, we did that to, uh, not have team members have to work through the weekend to get software out In No, it makes all the sense in the world, man.

Yeah. Well, you know, that must have been, I I never thought of it until, you know, Ashley gave me the reasoning and I said, yeah, that makes a lot of sense. Anyway, so for all of you out there who like to follow, you know, month in and month out, the latest GitLab release third Thursday of every month.

Alright, David, well, I don't wanna let you, I don't wanna steal your thunder. Tell us, tell us a little bit about the latest news with GitLab. Yeah, so, uh, this month we are excited to have GitLab duo's code suggestions feature that's, uh, AI code completion, uh, move to general availability.

And it's actually on the 21st because that's the third, uh, Thursday of this month, uh, in December here. And with that, what we're excited about is the improvements we've made. There's a really nice, uh, usability enhancements.

We've continued to improve its performance and its accuracy. And I personally just had a little bit of fun building a lightweight web app just through our code to comment functionality, being able to describe why I want something to do, and it creates all the code for it was actually really slick. Wow.

The, yeah. The other thing that, uh, we announced a couple weeks ago was that Duo Chat, our natural language AI assistant moved to beta, and with that we introduce what we're calling code tasks where you can actually highlight a block of code in the chat window, put like slash explain, and it'll explain all of that code for you in natural language. Um, what I'm really excited about with, uh, duo Chat is the fact that keeps on getting more powerful every week because it's a AI assistant, it's learning from it's engagement with you.

It's also learning from the project in which you're asking questions about. And it's all done with a privacy approach to it where it's not learning your code, but it's learning about the types of questions and how to better answer 'em based off your conversation. So watch out, uh, that's gonna move into, or stay tuned, watch out.

Sounds kind of scary, but, you know, stay tuned. Do a chat's, working its way to, uh, general availability here in the coming months as well. That's fantastic.

So, you know, as I mentioned, I was in, uh, reinvent last week, and of course 60,000 people thereabout, and each one of them probably used the phrase AI a dozen times at on average, if not more. Yeah. So it was all AI all the time.

Um, I, I interviewed a whole bunch of different vendors and, and they all had, you know, AI in their story. What I liked about the GitLab Duo version is it's, it's sort of horizontally integrated across the entire GitLab offering, right? Because GitLab itself is kind of a platform, right?

You have, you know, from shift left and it's obviously security, but it's coding and it's testing, and it's CICD and it, you know, it's, it's in a, it's a DevOps platform. Yeah, it is. And duo you're building in and every, at every stop right, at every station here.

And, and you know, and I guess it's the same underlying AI technology, but, you know, twisted or twist is not the word, but optimized for each use case, if you will. Yeah. I mean, for those who are aren't as familiar as you are with GitLab, GitLab is a DevSecOps platform.

So we support everything from planning to coding code review, CI ICD, security scanning, uh, unit test, testing, uh, deployment and monitoring of that application all from a single application. And so to your your point, when we decided to start adding AI to GitLab about two years ago, we knew we had to do it differently. And so we decided to embed it, to your point, Alan, across the entire DevSecOps lifecycle.

So today GitLab Duo has 14 features that make it up. We talked about chat and code suggestions, but it also can help you with planning. Uh, the most recent feature we added is CICD troubleshooting assistance.

Wow. It can help with root cause analysis as to why your pipeline has failed and make recommendations on how to fix it. And so if you look across the entire SDLC developers only spend 20% of their time coding.

That other 75% is all the tasks around them, your security team or operations teams. And so we wanted to optimize GitLab do it to make everyone more effective. 'cause GitLab's, uh, one of our, you know, core tenants and our mission is that everyone can contribute.

Well, if you make your developers faster and no one else, they can't contribute as, as effectively as well. And so, you're right. One thing that's different is we've taken that approach to AI like we have with DevSecOps and delivering software.

Excellent. So there's 14 different, is it modules, is that the right word for duo? Yeah, we've been calling them features, but modular Features.

Okay. Or, uh, capability is, yeah, there's explaining code, refactoring code, uh, being able to help with merger Quest writing the description of what the code change is helping get through code review. Uh, there's my favorite, one of my favorites is explain this vulnerability where it explains the vulnerability the scanner found in natural language for the developer and in the programming language they're using.

And they're able to then see what that looks like and then see an example of how to remediate it so they can fix the code and not ship that with that vulnerability in it. So, lot, lots of things all the way across the SDLC. Love it.

Excellent. Excellent. David, one of the, you know, I've, I've been getting smarter about AI from, as a result of all the people I'm speaking to about it.

One of the issues that I've seen pop up is the idea of the security of what I input and output into the ai, right? Mm-Hmm. How do I know the, if you know, I may be putting proprietary IP into my prompt, I may be getting proprietary what I think is proprietary ip, you know, out of it.

How do I, as a GitLab user, how do I, you know, sleep at night knowing that that somehow is not making its way into other people's usage? Yeah, it's a, it's a great question. So, uh, we actually set ourselves three core attendance when adding AI to GitLab.

We've already talked about the first one, which is AI across the entire software development lifecycle. The second is being, uh, transparency and privacy first. And the reason why that's important to us is GitLab is an opens core source available company, but we're actually trusted by more than 50% of the Fortune 100 to secure their intellectual property.

And so if we're applying ai, we have to apply it in the same way. So we don't use customer code or any of their proprietary details to train and fine tune the models that our customers use. And so they're trained on sample sets and data sets that are not that customer's code.

Reason why that's important is now you're not introducing that intellectual property into the training where it can inadvertently be recommended to someone else. And we're able to show our customers what we're doing because that transparency first thing is we document everything. I, people who don't know GitLab as well as, again, you or I may know Alan, don't realize like GitLab has our handbook live on the internet.

We have everything we're doing documenting our documentation. So if you wanted to see what models GitLab do is using, you can go to our Docs site and see the different models we're using for each feature. And that's how we're able to then earn that trust from 'em.

But then also we're documenting how those models were trained. And so now they're able to see in our source code, we're not using it and we're documenting the models and how they're trained. And you can see we're not using it there either.

And so the cases where we've had to, to use customer data, like our suggester reviewers, we're running a dedicated copy of the model and the local inference for just 'em, so that way that data is not getting cross-contaminated between customers. And so it's really allowed our large regulated customers to begin to adopt GitLab. Uh, we actually shared on our earnings call, which was actually yesterday, a quote from NatWest and their use of GitLab Duo and a company like that, a customer like that of ours who's heavily regulated, they need to be able to trust that the AI is not going to introduce intellectual property that's not theirs or leak their intellectual property.

And what they actually shared was that they've seen a really big boost in their productivity using GitLab Duo, and that it's not just their intermediate and like associate level engineers, software developers, but it's also their principal and their staff level are getting a, a big boost. And across all the customer feedback we've gotten so far on GitLab Duo, they're seeing an average of a little over 50% boost in efficiencies using GitLab Duo. Wow.

And that's because it's across the entire SDLC. I love that. You know, and this, that goes David to something that I tell people, you know, when when they, you get to he pennies who say how many jobs AI is gonna take away, I, I don't buy into that.

My experience with technology innovation over the course of my career says that no, it doesn't eliminate these things don't eliminate jobs. They, they 10 x people, right? And I think that is exactly what you just hit here.

50% productivity boost. That's huge. It's huge.

Um, yeah. I mean If you think about automation, the intent of it is isn't necessarily to replace a job, it's to make you more efficient. And there's so many mundane tasks that end up happening as part of software development that what if you could get your developers, your compliance team, your security team, focus on the big strategic item, and then those things that are more automatable or doable via AI are then handed over.

You're now making not just the company more effective, but you're making like cut your employee engagement better. 'cause you're now working on more important partner tasks. Absolutely.

And more high value stuff. Um, David, I had another question. Again, I got a little smart from talking to all these people at Reinvent.

So a lot of the vendors I'm talking to are trying not to get locked into any one LLM or particular chatbot backend, and they're building their AI product line so that you could plug into multiple or different, you know, chatbots and, and LLMs. I'm wondering what GitLab's doing on that front? Yeah, that's a great question.

So, uh, that's actually the third of our tenants. Uh, when we were adding ai, I Should get an award. I think I hit 'em all.

Okay. Yeah. Yeah.

That wasn't even planned. I, I think it's great, but what, what it is actually for me, Al, is it's like reaffirming those tenants that we set almost two years ago at this point. But the third one was best in class ai.

And what this means we're picking the right model for the right use case. And, uh, earlier this year we had hit 16 different, uh, AI models we're using to Power Duo. And what we're doing is we're saying, what is that thing that we need to do and what is the best model for that use case?

And so we are partnered today with Google Cloud philanthropic Oracle, like Oracle Cloud, but then we also build our own models and use open source models. And so we see that we need to select the right model for that use case. Even that means that to o Chats, which is models in the backend or reintroduce a new model.

I think the, the big risk for a lot of organizations is they kind of take, and the best way I can describe this is like the, they say if you have a hammer, everything looks like a nail. Sure. It feels like people are grabbing an LLM and then everything is that nail as opposed to saying what is the actual right tool for that use case.

And so we're very proud of the approach we took. It's allowed us to be able to say, Hey, that feature over here, maybe that's better for a in-house model. Or, Hey, what we're doing over here, we've optimized this open source and it's actually better.

And so we'll continue to do that. We don't think a single model can do everything for DevSecOps and we're, we're seeing the, the power of duo by using multiple models depending on the use case. Got it.

David, if you don't mind, I want to turn from GitLab specific to maybe AI in general. You know, we're coming up to the end of the year. We have our predict our annual look ahead virtual event.

I think it's January 18th, but a preview if you will share with our audience kind of what, what do you see here for 2024 with all of this AI and, and, and do you want to get specific about, you know, GitLab with it? That's fine too. Uh, yeah, I mean, may just talk about trends I see happening going into next year and throughout the year.

I think the, the first thing that jumps out to me is bias and ai. And you can see it with some of the GPTs that have been created now that people can create their own. Uh, there's been history around some of the AI services and having inadvertent bios today.

Models are trained on the internet. And I don't know about you Alan, but I don't always find the internet the safest and most like welcoming place in the world. And so as we build more complex models, we're ingesting more data from places like the internet and that's adding an, uh, inherent bias.

'cause that bias is there and we as a software technology world are gonna have to figure out how to remove that so that way people are getting the best experience. I think the other thing is, I think AI is gonna continue to transform, transform DevSecOps today as an example from our annual DevSecOps survey, which I know you love. Uh, when they come out, uh, 50% of people said they're using AI and testing.

Yeah, I think that's gonna get up to 80% next year and get close to a hundred percent the, the following year. And then maybe the, the last thing and, and I have a couple more predictions, I'd love to continue to share them with everyone who wants to hear them. But the, uh, last one really is the, uh, threat to intellectual property that you talked about earlier.

I think that we are at this point with AI that even though it's still kind of nascent and it's still growing and everyone talks about its power, but how there's a lot left unlock, we're gonna, we're kind of seeing this world where we were with social media a decade, two decades ago, where we have to ask ourselves, is privacy more important or is the efficiency it more important? And social media went be whole. You don't pay to have an account, but they're using your personal data to make the service better.

And AI's kind of at that point now where there's people who are like, I'm willing to give up some of my intellectual property that makes the model better. And then you've got regulated customers and people are, we're about their privacy and we're gonna struggle with that going into next year. 'cause a GT five a cloud three, they're gonna need more and more data and where's that data come from?

And I think we're gonna have to watch out for intellectual property leakage and vulnerabilities that get introduced because of the training, data training on things like a Stack overflow or location where insecure code is written, where people are asking for help and that now gets taught by the model. Yeah. You know, so one, one of the things I learned in law school law professor told us was that the law always sort of trails behind technology innovation.

And so I don't think our laws, our regulations though, you hear noise and they, but what we hear is more aspirational or inspirational, but it's not real yet. And so I, I think we we're in this period where that's lagging behind that's gonna lag behind the state of the art and not just the next year, but probably for the next few years. Um, so it, it's gonna be a little bit of a wild, wild west, right?

Because a, the whole thing with ai, it's moving from that cute wow, isn't he smart kind of stage into the terrible twos, if you will. And, and there'll be bad people who, who do bad things. And it's just unfortunately the nature of the beast.

Um, but that being said, again, the the we, we can't let what could be bad blight out all of the good that comes out of this. And, and, and we need to just, you know, I'm, I'm not one of those people who think you, you could do a six month moratorium or something, right? You can't put the genie in the bottle.

No one's gonna play by those rules. Um, so I'm, I'm excited. I'm, I'm excited to see where, where it is, you know, where, where it's going.

Yeah. I I'm with you. I, I'm, uh, I, I feel very strongly that AI is gonna make a big impact in people's lives in a very positive way.

Mm-Hmm. Uh, some of the stories that I've, I've seen over the last year, we'll call it post the chat, GPT launch, uh, a mother shared on a podcast, I like that, uh, using ai she's been able to build better, uh, meals that she can prepare for her kids who have allergies and health issues, right? You've heard doctors being able to identify things like cancer a lot more effectively or tumor more effectively.

And so I I'm with you. I, I believe it's got a really strong future. The thing that I always talk about is we also need to talk about the trade-offs and the risks.

And I agree with you, uh, regulations usually trailing sometimes decades behind innovation. And I think we're at a, a good point where there's a lot of engagement now more than I would've expected from regulatory bodies around the world. And we just have to all together figure out how to make AI the great thing it can be without risking privacy, intellectual property and people's safety.

Right? And I believe we can do that. There's a balance.

Yeah. We, we need to find that equilibrium. David, we're, we're probably over time, just real quickly for our audience, we do have a virtual event, I think it's December 12th called AI in Action, where actually a lot of the sessions touch on a lot of the stuff we've spoken about here, David.

And then, as I mentioned again in January, I think it's the 18th is our Predict 2024 event. And AI will be top of mind there as well. So I invite you, they're both free.

com, you can take it from there. Um, but David, a pleasure seeing you. Enjoy your time home.

I'm sure we'll see each other on the road sooner than later. Um, I'm sure we'll, but if I don't speak to you before, have a really happy holiday and New Year's and we'll be in touch. Yeah.

Thank you very much, Sean. Thanks for having Me. com, obviously, but where can they go?

com/gitlab hyen duo. But if you just Google GitLab Duo comes up, Best way to find It, everything we talked about and more. Um, so yeah, please go there, but thank you for calling that we, we kind of missed just telling GI I can go read about Please.

Yeah, No, it, it just occurred to me how do I, what am I leaving people with? Yeah, David, always a pleasure my friend. Be well, enjoy.

Likewise. Happy holidays and thanks for having me. Thank you.

David DeSanto, chief Product Officer GitLab here on Text Drunk tv. We're gonna take a break and we'll be right back.