GitLab’s David DeSanto on Embedding AI in Software Development

hi I'm James Maguire and on today's webcast we're talking about how artificial intelligence changes workflows in software development we'll talk about where we are now with AI and development and where we're going in the future to get insight into that I'm joined by David Danto Chief product officer of gitlab David thrilled to have you with us and happy Monday to you happy Monday to you too and thank you for having me absolutely so are where are we now with AI software development I I think that uh my sense of the development world is one where AI has made pre- many inroads certainly there's a lot of code being written by AI both by junior junior developers and Senior developers but what your sense is it still emerging is is it largely adopted in the development world yeah so I kind of consider it if you look at like a hype cycle with the Peaks and and the valleys and whatnot and so I kind of see this like leaving the trough of disillusionment and entering into to the slope of Enlightenment I think a lot of that has to do with like how quickly gen popped up for software development there was a point two years ago maybe months ago where everyone said I have to use AI in like development if I'm not I'm not meeting the goals of my board or my CEO and as it continue to mature questions like well how's my intellectual property being handled how is things how are things being stored and so forth and so that led to this world where there's a little bit of a pullback so to kind of look at it I I would say there's a lot of organizations using AI they may not have adopted across their entire engineering or for making sure they're going to be protecting their intellectual property they've got the transparency and privacy they need from the product and so forth but I'll definitely tell you it's going up uh we have a de secc op survey we run every year it's Global we had thousands of people respond and the people who responded that identified as a developer 78% of them said they were using some form of AI as part of building software where that was helping them write test helping them write code helping them summarize sections of code to better understand it so based on a survey we can really surmise that AI is fully adopted in development at this point if if the figure is 78% yeah I kind of think that you know we take a subset of the world right not every human being responds to our survey but I do think it's safe to assume that almost every organization has some AI in it they just might be somewhere in their adoption journey and not 100% sure that they're there yet to be 100% everyone's using it all right so given the fact that AI is pretty well adopted that 78 figure 78% figure does say that what areas are currently most impacted yeah so they're obviously the number one that everyone talks about is AI powered code creation this is either the cursor is in line it's doing completion of a couple rows or it's code generation where where they're doing like a comment decode or talking to a chat bot but in that survey one of the other things that popped up is using it for unit tests whether that is helping create them or helping run them you know there's a lot of companies that are starting to see that their QA needs an additional boost and AI is a great way to do that the other ones that have been interesting that have popped up developers TR to use it for generating documentation for the code whether that is Internal Documentation or maybe for an external product that they sell But ultimately what it's really resonating with is it's helping developers scale better than maybe they would have done by themselves and that's because they're getting that assistance from an AI tool interesting that AI is actually doing quality assurance so it's a it's a pretty high responsibility task is it is it really doing it yeah so I think it you have to look at it from the test generation standpoint right so it's running a unit test for a section of code that's a little bit more ictable than saying write something a test for something where you don't get to see all the source code or there's some other component to that really what I think what it comes down to is as part of the merge request where they're merging the code in they're able to leverage AI to also help generate the test case at that time and I think I'll speak as a former engineering leader uh before I went you know to the dark side and join product uh the uh the thing was that getting tests written was always a huge pain and if you're able to then look at the code and and have ai write it at the time you're committing the change there's a lot less or there's a lot less unpredictability a lot more predictability well all right what about the idea of embedding AI across the entire development life cycle it sounds like it's somewhat already there is is is that the ultimate goal of and and the best use of AI in development uh I absolutely believe that Gab believes that it's why we built gab Duo our AI powered Suite of de Ops workflows what ends up happening and this also came out of that survey and it it's honestly been the same for the last several years a developer spends according to this year survey 21% of their time writing code the rest is spent on all those other tasks that go around them previously it was 25% so it's interesting that it's gone down some on developers it barely speaks to the amount of roles we asked them to take on today but the other part of that is that if you make your develop team a lot more effective and you don't bring along the other teams we just talked about quality a second ago you security teams platform teams Ops teams you're not going to get the benefit from AI so with gitlab Duo and we just uh brought to GA Duo Enterprise we're focused on all those other things around developers as well while that's adding Ai and planning helping help you understand a conversation generate a better description of what needs to be changed helping you through code review that's where we started with AI in 2021 that's what the biggest bottleneck was from our customer interviews and then now today helping with security vulnerability is being able to identify them describe them to the developer natural language convert them by resolving the vulnerability for the developer and letting them see what was changed uh and then my favorite featur is probably a root cause analysis which is helping with cic pipelines and making sure they're always running green that's how you give power to an organization it's not just how much code do you generate it's what's the quality of everything around it as well and I always like think of it this way when AI started generating code it became about the quantity of code that was being created but again as a former engineering leader and I'm sure people listening to our conversation would say if you're Genera a lot more code does not mean it's quality code and doesn't mean it's addressing the use case and so you do that by helping everyone around the developer and that's why as I said it's been a focus of ours and why we just released do Enterprise to GA a bit more about uh Duo Enterprise so in other words it sounds like it's an overall planning workflow enablement uh program to talk a bit more about that if you would yeah absolutely so we kind of break it down into a couple different we'll call them tool Suites uh it's planning and templating uh Duo can help you create descriptions for issues I like the fact that it can also summarize conversations gitlab is an open Core Company our issue trackers is live customers been commenting on issues Community people commun communicate or leave comments on issues same thing with internal team members and I get pinged like hey what's your opinion on this it could be a hundred Pages worth of comments and you can have Duo summarize that for you I think it helps with code review some of that summarization is helping describe the change better than how you would have described it yourself uh helping summarize up someone's feedback to you in a way that you can go take action and then you move into the code creation code generation test creation chat so you can have kind of a partner working side by side with you and then security and compliance helping with understanding and resolving vulnerabilities helping you troubleshoot issues in your CSD pipelines and the last thing is really uh analytics reporting so you can understand the true impact of AI on your organization both at the project level where you can see what the Boost of the developer is but also as an organization understanding the broader impact what AI is having all right sounds like a rather a rich tool um all right so I talk about AI in the larger world I've heard more than one developers say they're uneasy about ai's role in development you know perhaps it'll take jobs particularly that it will make the life of Junior developer Junior developers harder because AI can more easily replace those jit developers your thoughts on that so I I may start with I'm not sure that's the right question we should be asking ourselves I think it's because with every new techn technology there always is some sort of impact to an existing Workforce or the type of work that people are doing I experienced that sort of my career in it and my experience and expertise changed as tools became more prevalent but I think the question that organizations should be asking themselves is how can they best harness AI across the software development life cycle we were just talking about so that way they can accelerate their Innovation and get their team members working on the more important strategic task asks you know AI today is doing a great job creating code but it's also really good at like building out templates and refactoring code and then those tasks that would tie up a developer's time they get to go work on the more strategic thing for you as an organization and so to kind of tell you that from the perspective of our customers um AI has become the partner for Developers for our customers and have a couple customer examples I can share with you as well but it really depends on the maturity you asked about Junior so what we're seeing is Junior and intermediate developers are becoming more like senior developers they're able to take on harder tasks that's because they can have the code explained to them when they get access to the code base they learn things like how are the variables being used where are they being passed to and those like critical things that as a new developer it's going to take you time to onboard they also prefer the code completion functionality because they can see the next couple of lines being printed out for them and that's help helping them better understand the why behind what's being recommended and how it'sit in the application but those benefits aren't just for them though I think Junior and intermediate level developers are going to get a really nice boost in their career because of it senior and principal level have been interesting to watch as well uh Duo has been out in the market for a while and we've a lot of customers have now adopted it as part of their software delivery and part of that power for them has been their principal level and Engineers can deliver larger strategic projects faster this is because they like things like code generation where they can describe in a uh like single line of code comment or multi-line comment here's what I need done and here's why and then they're really guiding the AI to create the thing that would have taken them a couple of days and I think the the best way to look at it is a customer recently shared that a project for their most senior developers were would have taken them a week so five business days uh with three of them and one of them completed the task in two days leveraging Ai and this is a major refactor of a microservice for them and so that sort of thing that would tie that developer down it's done faster they can now go to the more strategic work the organization has H okay well that is it's a it's a part positive Vision to be sure um with that sort of positivity mind let's look at the future of AI and software development I'm thinking that obviously we you know low code and new no code is not new but you know there are now low code know applications that will generate AI apps for you and and someone doesn't even need to be a developer at all forget even being a junior developer there's some Regional sales manager who might be building you know bottom end AI apps so like you know five years from now I can said you know build me this and in the we' be able to to do that um it's it's a wild world Ai and software development what do you see when you look ahead yes so there's really two parts to it that I see happening one's the first part we've talked about so far which is the we'll call it the acceleration by applying AI across the entire software development life cycle uh I think your spot on James like it this will bring in a whole another world of potential developers I I like to joke and say I know I don't I know I look like I'm 29 right clearly a little bit older than that but like even for myself I haven't been a full-time developer in a while I can now to your point come in and contribute and I think you're going to see a lot of Citizen developers popping up who can do that the best way I like the demo Duo uh to customers is I start with a project and I write just those prompts those multi-line comments and build an app in front of them and that's something that I think has been a struggle for the low code no code Market whereas today with power generative Ai and large language models you can feed enough context that it can now answer properly for you whether that's the right type of web service you want to use the right type of database all those things that it can learn from the project that are a little bit more challenging with low code no code uh but the thing that's more exciting for me when we think about the future is the move to autonomous agents and things like agentic workflows uh gitlab is very much focused on this right now we having our depc Ops World Tour uh event so if you want to come here about what gitlab is doing if you just goog git lab de psyops World Tour you'll see the cities will be in the next one's London and then it's Singapore uh but we're going to be talking about Duo workflow which we announced back at the our G lab 17 launch event in June and what de do a workflow is it's an autonomous AI agent that's knowledgeable of devc Ops and what this means for an organization is now instead of having a reactive promp based AI we have to go in and ask get a question or you have to click a button to get it to do something it can now take actions on your behalf and of course have guard rails around it and so what we've been demoing is the fact that do a workflow can grab an issue that it's assigned to and its backlog so the your regular development backlog uh create a development workspace make the code change commit it into the project and have it assigned to a human reviewer and the human reviewer can give it feedback and in real time it'll refactor the code based off the human feedback and of course then it learns like any other developer would learn and so we see that as the future software development is that AI is there to accelerate the typical work but with something like an autonomous agent it becomes a true team partner working alongside your existing team because there's still a shortage of developers in the world I feel like my entire career they keep on saying that and I guess I'm not 29 so it's been for a while like this is going to help fill that Gap and I think it's going to be the way Sofer starts getting created uh in the long run you it's really fascinating about the idea of the agenta Ai and how that that bot or whatever you'd ultimately call it is that that agent that autonomous agent can really do tasks on its own uh of course it is dependent on on the large language model it is it is feeding off of true it it that is true and I think this is where G livs taking a different approach in those companies to AI so we have three founding principles we've talked about the first one AI across the sdlc the second is we're transparent with what models we're using and how they were trained and the third one's what you're just touching on we choose the best AI model for that use case so something like Duo workflow is not using a single model it's using multiple models to able to deliver what it's doing and that's true about all of Duo today you know we're partners with Google and Fric we use open source models we write create our own models for some features and so you know right now do is using about seven different models for a Duo Enterprise because we're choosing the right model for The Right Use case and that's how agentic stuff's going to have to work work too I think if you just take one large language model you're not going to get the uh the depth of what you need to be able to have that workflo be successful and so I'm very excited about our partners and how they support us I'm very excited about what our broader Community does to contribute to git lab and I think it's going to be very exciting as we bring Doo workflow to Beta to more customers as well as GA David I think you said it uh fascinating stuff it's going to be a really interesting sector to watch uh please come back and talk with us again sometime I would love that James like enjoyed our conversation day and thank you for having me

This transcript was generated automatically from the video's captions and may contain errors.