Empowering Continuous Delivery: Unleashing Generative AI for Software Innovation – The CD Pipeline EP 9

Hi, everyone. Welcome to CD pipelines. For those of you who are familiar with CDD pipelines, you know that CD Pipeline is a monthly show brought to you in partnership between Techstrong Group and the C D F, the CD Foundation, which of course is part of the Linux Foundation.

I'm Alan Humel, c e o editor-in-chief for Textron. And welcome to our show. Every month we delve into different aspects of continuous delivery, whether it's the projects of the CD Foundation or different relevant topics that are weighing in on CD or the cd, uh, world and Market, such as we're gonna discuss today or something else.

We, every month we, we come up with some great topics, and we match that with a great panel of people to discuss topics that if you are involved in the continuous delivery world, are important to you. Let me introduce you to this month's, uh, panel as a matter of fact, and we're gonna jump into this month's topic in just a second. I'd like to first introduce you to Arvin Cannon.

Arvin is a director of software engineering over at eBay. And Arvin, welcome to the show. Why don't you expand on your bio a little bit for our audience?

Sure. Uh, thanks Avin, uh, for having me in the show. And, uh, yeah, I'm, uh, director of, uh, engineering at eBay.

And, um, I'm having a lot of fun, uh, leading the cans, uh, delivery group at eBay and as part of the platform measuring group, uh, and, uh, as a group like we will the CANS delivery frame, uh, platform for eBay, along with like other things like, you know, uh, measuring systems for Dora Metrics and all our, like, you know, uh, how, uh, the pull requests are being evaluated and several other sub metrics that can help, uh, teams identify, uh, bottlenecks. And there's also like, you know, test automation frameworks and things like that. And, uh, of course, like right now having a lot of fun exploring AI for, uh, various use cases, You and everyone else it seems, huh.

Next up, let me introduce you to Christie Warwick. Christie is a software engineer over at Google, and she had, she won the prize for the coolest shirt on today's panel. Um, Kristy, welcome.

Why don't you introduce, give you a little more background. Sure. Um, hi everybody.

I'm Kristy. I'm a software engineer at Google. Uh, I have a background in a bunch of different domains.

Um, I spent a lot of time working on AAA video games, um, mostly Call of Duty, and then I decided to switch to something a little bit more, um, sane and stable and started at Google. Um, I've been, I'm one of the creators of Teton. I've been working on that for about somehow five years now.

Um, and I also wrote a book not too long ago, uh, Gring Continuous Delivery, um, and very, so I'm very interested in the continuous delivery space and really motivated by anything that just, um, makes developers' lives easier and, and more fun. Absolutely. And the name of that book was Rock Continuous Delivery.

Yeah, rocking. That's what I was just making sure. Okay.

So check that book out, guys. Uh, let me introduce you and, and thanks for being with us today. Christie, our final panel member today, but certainly not least is, uh, he's been on a few panels with us here lately at Textron Group.

It's my friend Ravi Lockman. Ravi is product manager at Harness. And Ravi, welcome to the show.

How are you, man? Yeah, I Doing great. Yeah.

Excited to be here with the panel. Like, it'll be a great conversation. You know what, I don't wanna short shift you on your background though.

Why don't you share a little bit with the audience? Yeah, Sure. Hey, my name is Ravi.

Um, I kind of started my career as a software engineer. Uh, kind of went into platform engineering and distributed systems after that. And then now I'm a product manager at Harness.

So I kind of, uh, I'm a product manager over our observability solutions or integrations at Harness. I'm also our own developer education at Harness. So two very, like different things that we do.

Uh, big fan of the C D F used to participate more actively and excited to be participating again with the C D F. Uh, and yeah, like as a software engineer, um, I never knew how things gone to production until later in life, and so that's why I'm such a big fan of continuous delivery. So excited to be on the, uh, in the panel.

Fantastic. Last but not least is my co-host here on, on, uh, CD pipelines. And she has a whole bunch of different hats underneath that blue hair.

I'm gonna let her tell you all about herself and she's gonna introduce what we're talking about today. Lori LaRusso. Well, thank you, Alan.

So, yeah, so my name's Lori. I am the open source program manager at Jfr Holler at my company. We love, it's not the best church, but it, I'll take second best today, you know, over KAZ and Kubernetes.

I got it. Mm-hmm. Understood.

Um, so yeah, so I'm the open source program manager at jfr, which allows me to work for foundations like the C D F I'm their outreach, um, marketing chair, and then I'm also the marketing chair of the C N C F and do stuff with open s ss f and all kinds of foundations. Basically really love that I get to work, uh, for the majority of my time in a vendor neutral space on sort of emerging tech and like the hot topics. And speaking of hot topics, today, we are going to be talking about generative ai.

So where DevOps was a hot topic, now we're taking it up a level and we're spicing things up to say, okay, DevOps, collaboration, culture, all of these things work together. You've got developers and ops. Now, where does gen generative AI come into the continuous delivery pipeline?

Where does this work? And so let's get right in there. I'm gonna throw out a question and say, Christie, where do you see generative ai?

What is, like, when you think of it in terms of, uh, continuous delivery? What's the first thing you think about You? I, I have to go first with the hard question.

I guess the first thing, the first thing I think about is just the fact that it seems at least that generative ai, um, it has a tendency to, uh, make some mistakes or need some guidance. So my feeling is, at least at the moment, it seems like it fits well in places where you kind of need a little bit of, um, assist, you can assist a human with something, but maybe you don't completely replace what the human was doing because you might need to kind of spot check it a little bit or, or guide it or make a few corrections. So things where there's a lot of toil for people and something takes a long time, but you could have a machine do it much faster.

Like, I don't know, digging through logs or like taking an error and going from that to like, what does that mean? Can you explain it to me? Um, those seem like, at least, at least some of the maybe low hanging fruit around things that generative AI can help with.

I think it's great that you start with the low hanging fruit. Uh, Ravi, do you like have a take on that? Because I think people are always assuming it's the, it's the high level stuff, but with anything, you have to start at the beginning, right?

Yeah, yeah. I'll take, I'll take that question like in two parts or like maybe slide in a little bit of, uh, uh, just how I feel about generative AI or, or understanding. I think it's so new that from, from, at least from a product standpoint, we're still learning like all the capabilities.

So I'll tell a little bit of story. So the very first time I used generative AI was in March. Um, we, we were having, like, every year from university, we all get together.

The older we get a little bit harder to plan. And one of our roommates couldn't make it, so I used chat G B T to come up with a poem to tell him, Hey, these are the reasons why you need to come. And I was like, oh, this is really cool.

Like, it hit like really solid reasons. Like, Hey, you know, we only meet up once a year. And, uh, you know, like he was missing it for a certain reason.

And I, I gave that information to chat, BT and I made solid recommendations of like, oh, that is great. Kind of fast forward to several months go by, how do we fit that into the enterprise, right? Or how do we help our users?

Um, kind of like low hanging fruit could be just that, like, especially what, uh, uh, Christie was saying, like, Hey, how do we explain certain things that are happening? Or how do we, uh, especially from the harness standpoint, we're actively building, uh, a generative AI piece of our product. It's called A I D A, artificial intelligence development as systems, uh, ida or ida, depending, I say IDA comes from Georgia.

Uh, I say differently than other people. And kind of like the starting use cases for us are like, oh, can, can it scour the documentation and prompt you to give help on, hey, you rent this problem, this might be entered in documentation or in the community in some sort of form. Um, but also it's getting more advanced.

But yes, like explainability is one of the first things that, you know, it generally I can be really help helpful on explaining things in different tones and different languages and whatnot. Arvin, do you wanna jump in? Yeah.

Um, I, I agree with like, uh, uh, with Cty and Ravi, uh, in terms of the low hanging fruits, uh, and, uh, the way that, uh, I see like generative ai, um, is, you know, like looking at, uh, the DORA metrics, uh, primarily the lead time for change, right? And, uh, that's what like we focus on. Um, so trying to see like, how can it accelerate, develop productivity and what are the areas?

And that really starts from the architecture, right? Again, uh, I think like we can, uh, all, uh, agree that, uh, it's, it's more like used to generate a draft content at the moment. So, and then there is a, a huge response.

We shouldn't be a semi accuracy of the content being generated. And, uh, we have a huge responsibility to review that, correct it, and then go through all the review process that we have in place, right? So, um, I feel like, uh, it can really, uh, help even with like architecture, uh, starting from architecture to coding, uh, continuous delivery, um, testing analysis of logs, and, uh, even like, uh, we are seeing that it's like giving us good insights when we, uh, try to give it access to the DORA metrics, all the metrics that we are being measuring, and then ask it to identify bottlenecks and, uh, feeding the right patterns in the can delivery pipelines, right?

And then expecting it to analyze pipelines and, uh, tell us, uh, what could be improved various, uh, based on like different applications. So these are the, uh, use cases that, um, um, I'm like super interested in and like exploring at the moment. So I, I feel compelled to jump in here.

I, I think low hanging fruit are great. I think writing a poem, writing a song, or doing something in the voice of William Shakespeare's a nice parlor trick, but that's not what gets people excited. Well, maybe it gets like my sister-in-law excited 'cause she's not a tech person.

I show her chat, G p t like Ravi, write a poem, you know, and she thinks it's magic, but that's not why we're here today, right? And quite frankly, even documentation is a nice to have, but it's not why they're investing billions of dollars into this. You know, we, we were talking off camera, I'm working with a bunch of people very big into DevOps space who were looking at this problem.

We, we actually have a hackathon on it coming up. The real, the real thing is how can we use this to empower our developers and DevOps engineers to do more faster? And at the heart of that guys helping us write code, I'm not saying it's replacing, we we're not getting into that.

It's not gonna replace anyone, but it's gonna make every developer better, more powerful. It may not come up with finished code, but the, the fact of the matter is, today's code is, and we spoke about this in another show, Ravi, a lot of what a lot of what developers do is they sit there and they, they type in some stuff, then they try it, make sure it works, they go back. It's a lot of it.

Today we're pulling down components from Artifactory or, or Maven or, you know, APIs we're using. A lot of it is, is sort of rote kind of stuff. We're not custom writing the entire app.

So something like a, a generative ai, when it's empowered with the right LLMs, when it's empowered with the right information, can really take a lot of the mundane developer task, a lot of the mundane testing task and scripting off the plate and allow, and like turbocharge developers to and turbocharge our C I C D pipelines to do more better. And I think, guys, that's why we're here. That's why we're here talking about it.

No one's getting excited about the poems anymore. No one's even getting excited about the documentation, frankly, right? 'cause documentation's always, always the last thing we do anyway when it comes to C I C D and development, right?

It's always like, oh yeah, we gotta do the documentation. It's this aspect of, of being able to turbocharge developers that I think gets people excited and not just developers, the testers, the, the deployers, the platform engineers. How's it gonna help there?

And I, I'm not, look, I don't consider, you guys are much smarter than me. I'd love to hear what you think about it. I have a very, very concrete example.

And then, you know, we can take it in different, different areas. So like, like where, where you're under the gun is when there's a problem, right? Or production facing.

So like root cause analysis or RCAs, um, an easy one happened to me. So like part of the developer portal, uh, that I run at harness, uh, one of the packages dropped out of N P M, like, it just gave a crazy error, like the build fails or the deployment fail, right? So like, that was like the, the cause and effect.

And immediately because of like years of experience, I'm like, oh, this particular package failed. Let me go try to get a newer version of it. Maybe it got pulled for vulnerability.

Or like N P M said, Hey, this is, this version will not be hosted on us. And going through Drew going through the steps, like as, you know, as I was trained to do years and years of, uh, dependency management, um, I made a fix. But our AI development assessment actually was spot on saying, Hey, this package is can't connect, um, or we can't pull the package, like try changing the version, uh, of it.

And I was like, oh, it mimicked what I just did, uh, with years of experience. So I was, if a more junior engineer or someone who hasn't dealt with that problem before came up with it, it was, it was pretty spot on saying, Hey, this, this is a problem. This is a possible resolution.

And in plain English, not very, you know, it could have been a poem, it would've been funnier, a poem. I I try to bring it to the product team. They say, can we do a sonnet or dev?

But, um, yeah, that was, uh, the, that, that was very helpful. It's like having autopilot and an airplane or a car, like the individuals that still there, but the guidance or the, you know, the du reduction of toil, like, you're given like a playbook to kind of run and it's like, oh, that was like spot on what the, the resolution was. So, So I like your idea though of an autopilot.

So autopilots weren't, uh, like that whole idea was not open for all consumers to attack, right? So I think maybe if we reframe low hanging fruit to letting generative AI learn, like there, it's kindergarten right now, right? Like, yes.

Like, and to your point, Ravi, like, because you had years and years of knowing from dependency management what it could be, and it spit it out really quick, and then you cross-checked and verified, you're like, yes, you were right. But because I had that knowledge, I also got there. Um, so I think maybe that's the part that we miss, right?

Is that it is a, it's gotta learn. And, and, and for it to learn low hanging fruit or smaller, simpler tasks might be a good start. But to Alan's point, like, um, where, like if it's five years from now, where do you think like your companies or the industry really will see a massive impact?

Is it in the productivity? What does that mean? Is it in the testing?

Is it in the automation? Is it in vulnerability? Like, you know, security?

Like, where, where do you think the big, the biggest push is going to be in terms of investment? Uh, I mean, at least I think like, uh, it's on all of these areas, right? Like, uh, like I stated earlier, different organizations would have, uh, uh, various tools and frameworks and technology in place, right?

Like, so let's start from there. So starting from there, if a new person comes in and they want to build up something, and, uh, even with the discoverability of those things based on the use cases, right? Even that is like one place that, uh, AI can help, uh, where it, it helps me pick the right, um, technology and tooling available within an organization, uh, given, uh, trained with the knowledge right?

Now, again, like, uh, uh, the, there's also like this, uh, thing that we have to look at, like, uh, do we go with like a third party provider or like, you know, uh, take, uh, an open source, uh, l l m model, fine tune it if you have the resources in terms of people and compute. Uh, but again, like there's also, if you just have to like, you know, feed in, uh, your data and then, uh, get answers out of it, like vector embeds could help as well, right? So again, understanding all of the techniques, uh, is, is important even to start like experimenting with it.

And then, uh, I think it opens a, a whole new avenue in terms of like where we can use it, right? Starting from like architecture to like coding. Um, again, support is something that, uh, I would like to bring on upon, because again, platform engineering groups, uh, we really want to, uh, spend engineering time on building new features and solving problems versus spending time on support, right?

So that's another area where, um, uh, general AI can really help, uh, given the context of all the internal documentations and like all the open source knowledge that it has, being able to, uh, help users of the platform engineering, uh, solutions get answers quickly. And also being able to free up time for the engineering group itself to work on solving problems. And then, uh, of course, like I, I think you bring up something that I think is a very important fact that the audience hopefully knows, but if you don't, you should, when we're talking about generative AI helping us in these situations, and I, and I agree it's gonna help across the board in every single instance, in every single thing, we're not talking about chat G P t that has the entire internet from 2021 on it, because that's where you get hallucinations and poisoning and all of these things that we hear about, right?

We're what I think the future is just what you said. We're gonna have custom LLMs, whether that's an open source, L l m or or each organization creates their own l l m, right? Where, where we control the, the universe of, of facts that, that this AI has at its fingertips and, and then can help humans with, right?

So that you cut down on the hallucinations, you cut down on the poisoning, and you, and you, you, you have it, right? What, what's in here, right? We we're experimenting with this at Textron, right?

com, create an l l m just of that, or, or these look for c d F, let's take all the c d f pipeline shows we've done and create just a custom l l m of that and make and, and put a chat bot on it, right? Think about those kinds of uses. I think you are going to see, that's going to be the real, um, accelerator, is you're gonna see organizations just creating their own LLMs for their, for their AI uses.

Not, not everyone's, it, it's not gonna be like, you know how everyone uses Google search for Google, but you, you, you're not gonna search the entire, for most part, you're not gonna search the entire internet, uh, wide l l m and I think that's gonna make a huge difference in, in productivity. But Gloria, to your point, this thing, yes, it does need to learn, but generative AI learns pretty fast, right? Compared to how humans learn, right?

It, it's not gonna take years like it took Ravi. But again, you know, I, I guess I'm just a, I don't know, I'm an optimist with this stuff. I think it's funny.

Uh, you're an optimist. I think I tend to be kind of more of a, a pessimist a little bit about these things. Um, but, but putting that aside for a second, the learning aspect is interesting.

I think there's, there's the models learning, but then I think there's us learning how to deal with the models and how to interact with them. And like Ravi was saying, I think one of the, one of the things right now is that we're very early in the process and we don't really know how to work with the models that well, and things like preventing hallucinations. Like if you look at the ways that people are doing that, one of the ways is even doing queries where you're saying like, don't hallucinate or be really sure that this is right.

Like, things like that. Yeah. And that actually works sometimes and sometimes it doesn't.

And I feel like when you start seeing that, you realize that we're actually really early in the process. And so like the dream of like, you know, you want like self self-driving CD or something like that, it, it does seem like we could get there, but when you look at the reality of what we're doing today, that's where, that's where the kind of more low hanging fruit come in. Like, what could we do today?

We can do these, these more sort of deliberate things. But when the, the dream, like, I feel like we're not there yet, but I do wanna be optimistic. Like, I think we can get there, but we all have to kind of, we have to learn a lot about how to work with these models.

Well, I, so you're not gonna throw a hail Mary, right? And I don't, I don't mean to go into football, but we're not gonna throw a 50 yard bomb and, and hope, you know, and make that giant leap. One of my good friends, a guy named Brad Feld, he's pretty well known as a venture capitalist foundry groups, tech stars.

He started and Mobius, and he was at SoftBank Venture Capital. He's always told me 98% of what we see in tech is evolutionary, not revolutionary. And generative AI may be revolutionary, but the steps we're gonna take to make it, you know, Christy, to your point about, you know, auto autonomous CD or something, self-driving cd, to get there, we're gonna have to take evolutionary steps, baby steps.

Not, not giant leaps, but you know, when you look back, you'll see those steps adding up to a path. And, and, and I think, I think that's, you know, in my 30 plus years in tech, that's pretty much the norm, right? There's this great, you know, breakthrough, but then it takes, it takes baby steps and evolutionary time for it to, to get real.

Yeah. I think just like if we take a look at like what a CD pipeline is, like for me, it's, it's, um, very democratizing. It's, it's the culmination of expertise of many teams, right?

And so if we take a look at what is actually in the CD pipeline, there might be some application security expertise, there might be some observability expertise. It would be application expertise, infrastructure expertise. And it all comes together in this thing called a CD pipeline, right?

So the culmination of knowledge, um, there's, there's a DevOps term, but it's, it's old, it's an old measure called cams, C A M S, if you're familiar with the term. And Dan Edwards invented that term. Yes, yes.

Spot on. And so every like, industry has gotten better at all, except the last one, SS sharing as humans, we, we suck at sharing. Like if you look at like the elite versus the non, like the, I forget what the non elite, you know, the lower performing teams are called.

But the, it's very, the gap is not big between what separates the best to the worst, which means we are all bad at it. And one thing that I, I'm, I'm bullish on generative AI is like explaining, going back to, you know, however it needs to explain, let it be in documentation, analysis, phrasing things differently that, you know, for example, if there's a security vulnerability that one of these n number of scanners you have, because the AppSec team enforces that, how can it explain how to fix it or how to explain, like, why is this important to you? And, you know, one thing, it's going back to vin's point.

Let's say, you know, Arvin, like, let's say I, I joined your team, either you like, depending how much you like me, like congrats, or I'm sorry I work for you now, right? Like, um, there's a lot to learn as an engineer. Like we have a hu like ramp time to going up, like our tools are just more complicated to learn.

And the quicker that, like for engineer to become effective, like that was like our room was getting at like, how can we lower the bower entry for people changing teams? Or how can we lower the bower entry barrier entry for someone joining? And there everybody could probably help with that, right?

Explain this very complex flow in a pipeline. So there's the explainability versus the generation of the pipeline, but I feel like early use cases are just helping people learn a lot quicker, especially with Chris, Christine was saying like, Hey, like, teach me a little bit better. You know, this makes me think about, um, the whole idea of reference architecture and then also interoperability, right?

So if, you know, making sure all of your systems in your pipeline are working and then being able to then spit that out to maybe the next team that is trying to do something similar, like, so where do you see generative AI helping in this phase of sort of like either helping to create some sort of reference architecture for things moving on, like Ravi was saying to upskill or, or interoperability. How can that help make you realize what's working or what's not working, or if something blows how to, how to then attack? Alright.

Um, so maybe I can take a stab at that. Uh, so some of the things that, uh, I think like we can ex uh, leverage generative AI is for like, like you had, uh, called out, right? Like, how do we carry forward, like best practices from one pipeline?

Like there are, like, uh, again, if I take like eBay as an example, like we have like thousands of, uh, tens and thousands of applications, uh, with like thousands and thousands of, uh, hundreds and thousands of pipelines and, um, uh, that that's where like we also like, you know, uh, I I tend to think about like lead time for change all the time. Uh, because if, if I have to say, like, if I have to say very clearly to a team that, uh, hey, your pipeline's lead time for change is X and uh, if you do A, B, C, then your lead time for change can be y which is like better than X by, uh, certain percentage, right? So those kind of, uh, analysis, uh, is something that we can leverage our generative AI for.

And that's, that's an area that we are exploring as well. Um, and, and like, again, it's all about like training it with the best practices, right? What are the best practices where, and, and also like, you know, letting it know about, uh, the areas of friction as well, so it can, uh, really come up with a recommendation that makes sense, uh, based on, uh, each organization, right?

So that's, um, something, and then like, even to promote like, uh, pipelines as like best practices based on the lead time for change and like complexities and things like that. And, uh, we also like, uh, another area is like the deployment, uh, strategies itself, right? Like when we are doing like canary, if you're doing big groups like say 1%, 33%, 33%, or versus like 1% 50%.

Now I, I might be like super conservative and add like a lot of like baking time and like have a, uh, deployment strategy that I as an engineer like feel comfortable with. Um, but maybe the error are found like much earlier, right? And if that's the case, then I can go with a more aggressive deployment strategy, reduce my bake time.

So these are again, like, uh, things that it could, uh, help with. And yeah, these are some of the examples. Fair.

I, I think something like a reference architecture is certainly something that I, I don't think we're gonna wait five years or even three years for a, a generator of AI to help us, you know, uh, generate. I mean, 'cause that that's the kind of thing where it could look at past examples, pick out those patterns and what have you, and, and, and, and shoot up a reference architecture. I, I'd like to bring up something else though that I think gets lost in the sauce with all the, you know, the generative AI soaking up all the oxygen in the room.

And that is kind of, let's call it traditional ai ml ops, AIOps ML slash ai. That's what we, we were calling it AI all these years, though we knew it was mostly ml, you know, people are throwing that out. Like it's, it's, it's old dirt.

And, um, but we were just starting to get good at that kind of stuff, right? Going through large data, you know, stores and, and finding patterns and, and recognizing, I mean, the whole observability space is kind of built on a lot of it, right? Um, do you think there's a place where they, they work together in, in, in helping us with our C I C D pipelines and deployments?

Christie, I don't mean to pick on you, but hey, you wrote the book. Um, what, what do you think? I mean, I, I wrote a book about continuous delivery, but not about artificial intelligence.

So I, I feel like I'm probably slightly outta my depth there. I mean, I think this is not, this is not what you were getting at, but it did remind me of something that I was thinking of mentioning, which is this kind of funny meta topic you mentioned, like ML ops. Um, and, and we were talking about, uh, there being a lot of value in, in training a model on your own organization's data.

Um, and, and the reference architecture idea, like if you could, if you could train a model on, if you have a big enough company, you could train a model on, uh, you know, all the repos you have in your company, and they could give you, maybe it can create a better reference architecture, but that has this meta component where how do you actually manage the model itself? And then weirdly, continuous delivery kind of comes back again because you need continuous delivery for the model itself. Um, so that is not at all addressing your question and totally sidestepping it, but I, I think it's an interesting space for continuous delivery to not only be harnessing the, this power, but also kind of enabling it and making it easier for people.

Absolutely fair. Yeah. Absolutely.

Yes. Like, so for, for, for Alan's question, um, uh, what, what I, what I see it's for the, the LLMs out there, it's, it's making the AI or ML ops like better. Um, if you take a look at the algorithm, so the piece of the harness product that I own is actually exactly that piece.

The, um, durability integrations that how, like our particular platform can learn and decide what was good or bad, um, based on logs, metrics, trace and spans, like, or what it can observe and then spit it out and say just as good or bad. Um, it goes back to like timeliness of the answer, right? And so, like, let's say that we were deploying something and you know, we, we need to make it a judgment call, right?

So to, to vin's point, if we can like, just press a little bit harder that do, is it a five minute judgment call? Is it a 15, 20, 30 an hour, two hours? When do we have enough traffic to make a judgment call?

Or when do we have enough transactions that we we're pretty confident we should promote something? Um, it it's, it's more, uh, I would say more, uh, evolutionary than revolutionary, right? Like the, the AIOps models that a lot of vendors have it really, vendors have it, different CD vendors have it.

Um, we'll tend to the, the training will get better based on the LLMs. Like that's what we're observing now at harness that hey, if we switch our algos to something else, like can we give a better response more quickly? Or also, can we observe more than we were able to observe before, right?

So like the bandwidth or pipeline that we can ingest data could grow. So it, it's more just like, Hey, can we give a better answer than we could have given without, so more evolution. Fair Arvin, any thoughts on that?

Uh, yeah. Again, uh, uh, on, on the point of like, uh, training, right? Like, uh, there, there's also like one more thing that, uh, I've observed like the temperature, right?

So, uh, how, how much creative do you actually want your lmm to be? And uh, uh, specifically we trying to use it to take a look at the insights data like Dora metrics and all the other metrics that we have, and then, uh, come up with certain, uh, analysis on what should my application be doing on where should my focus be? Or like, you know, if I have like bunch of like, say a hundred applications, what are the top tens that I could actually, uh, move to a better performance, right?

Like in terms of, uh, delivery performance. And, um, looking at that, uh, we do see that like the higher temperature, it comes up with very creative answers, uh, but like, may not actually be the ones, uh, that's, uh, relevant, right? And again, uh, that, that's where like really understanding how to, uh, tune the temperature, uh, is also like very important.

And then also, uh, feeding in the right context, right? Like, uh, so what, what is important for my organization? So feeding that as a right context with like system prompts, uh, those are all like, very important to understand as well.

So that, uh, uh, again, like we, we have, we need to be careful about it because, uh, sometimes we might, uh, make it just say things that we like, uh, but we also need to be pretty careful about, uh, how we, um, play around with these parameters. And also the top key, But I like that, right, is the idea of being able to have it spit a bunch of stuff at you, but you have to be the one case by case to make the decision. Because isn't that sort of just like real life in general, like where you're saying, oh, this gets me there faster, this might get me there a touch slower, but it's actually benefit on the back end because of whatever.

So I, I think, you know, that in and of itself is kind of beauty, right? Because you're like, oh God, now I have to think more. Like, now you're giving me more options based on like your creativity versus my, like stubbornness.

And is it the right time to flip the switch and go your way or is my way still might be slower, but tried and true because like, I'm not ready or we're not ready for that sort of ramp up. It's cool. I, I, I love including the idea of creativity and software development because I think that's something that, that's what I love about software development.

And I think something people don't think about is it's such a creative endeavor. Like there's so many different ways of doing the same thing. And I think that's something where AI can really help us, especially generative AI is just producing new ideas, just new ways of doing things.

But again, you still need that kind of human involved to take that input and then turn it into something. But I think it can really kind of help us embrace, embrace the creative side of, of this, of this job. Christy, you know, we should bottle that because what you just said is exactly the point for all the naysayers and doomsayers, what you just said is, is, is what it's about.

It enables and empowers humans and humanity and our creativity. And I think that at, at the essence, that's what it's about, right? I don't think it ever, you know, we're not creating, you know, the Steven Spielberg movie or something, right?

It, it's not that, it's about enabling and empowering people because at the end of the day, that spark still resides in humans. And I don't think that spark ever resides anywhere else. And, and so I, I think that's, that really is, that's an important, important point that our, our audience should, should take from it as well.

Excellent stuff. I mean, when I think about it too, Alan, like, you know, and maybe, 'cause we were talking about back to school, and I have a seven year old, but when my kid tells me something, I'm like, wait, what? You know, like I know the right answer, but like, she gets there in her own little crazy way.

And then, and so maybe that's sort of like similar to what you're saying, Christie is like, sometimes you get stuck in like tunnel vision and yeah, there's 18 different ways to get there, but like, this is the way, and then it takes this extra piece to be like, oh, hold, hold on, I've got a viewpoint. And it, you know, and then you really do need to attune to it. So like, I think that's Alan Christie.

Like those are, it's growth, it's evolution versus, you know, all of these things. And I like that we're on this positive kick, you know, pessimism aside. Cool.

We're on the positive kick. Very cool. Hey guys, we, we are, we're coming up on time here.

Um, I want to give everyone a chance to kind of final thoughts on, on where you are on the subject and advice, anything you want to share with the audience. Why don't we go in reverse order? Well, Lori, I'm still gonna make you go last 'cause you're the co-host, but, but Ravi, you we'll go Ravi, then Christie, and then Arvin, and then we'll wrap up with Lori.

Sure. Yeah. I think, uh, this resonating with, uh, Christie was saying that, you know, it, it'll, it'll help unlock creativity.

Uh, le leaving with two notes, um, one I heard, uh, on Instagram, a song written by Frank Sinatra and Littlejohn, so both of them don't live at the same time, but AI produced a song and I was like, we are living in the future now. Like, this is the most creative thing I ever heard. It was, it was really hilarious hearing them two sing together.

'cause they're, they're not alive at the same time period. Um, and that's it. Like, Hey, you know what?

AI will help us unlock things that we never thought was, uh, was possible. And also a joke I like to tell people, if you're familiar with Terminator, uh, the movie Skynet, Skynet will be written in Java. So take it for what it's worth.

The machines will be sad and run with memory problems, and so they won't be able to take over more than a day, I hope, like before we need to get involved. So, uh, but yeah, excited to be in the panel, a great talk and, um, yeah, just happy to be here. Thank you, Ravi.

Christie, Let's see. It's such a big subject. It's hard to even know where to go with it.

I mean, definitely I think, I think there's just so much potential and I hope that everybody gets a chance to explore it. And I think that what that means is it's, we, we have to also embrace kind of a thing that we're not great at doing in software, which is being okay with failing. 'cause I think that because we don't know what we can do with this, we have to be able to try some things and then not all of them are going to work out.

And that kind of has to be okay. Otherwise, we're not going to learn like what we can really get out of this. So I guess I would just encourage everybody to try some things out and then, uh, you know, if it doesn't work, that's fine.

'cause we have to, we have to figure out what doesn't work in order to figure out what does. So, you know, with everything in software, like be okay with failing. 'cause that's how we learn.

Um, and we, we all need to, we all need to really be embracing this and trying it out in order to learn what we can actually do with it. Excellent arvid. Um, so again, um, I I think like, uh, it, it unlocks a massive, uh, potential for all of us in various things that we do.

Uh, the development lifecycle, deployment lifecycle everywhere. Uh, but I think it's also important for us to understand how to, uh, use generative ai, um, in a proper way, right? Like understand various parameters, uh, and the details about it.

And then, uh, also, uh, to train people on it, right? Like again, prompt engineering. Uh, it, it might, uh, feel like a, uh, odd term, but again, it is very important, right?

Like it, uh, the, what we get out of it is like as good as our prompts and like, uh, and also feeding in the right context. I think we need, it's important to train people on it. Um, and also, uh, not to be afraid to try it, right?

Like for various use cases, experiment, experiment, uh, experiment in a safe way, uh, take care of the data privacy and like security, and then still like, experiment and, uh, then go like based on like data-driven approaches on that. Uh, so that's, that's another thing, right? Like, so I, I mean I, I, I talk to my teams and then like, uh, one of the thing that we, uh, discuss frequently is like, uh, I mean see AI as a, uh, technology but not as a solution for everything.

Uh, and we don't have to try to force inject it into everything that we do, but like, let's see, uh, if it is something that can boost productivity, right? Yeah. Thank you Arvin.

Excellent. Laurie, take us home. So I think like all of this is so exciting, and again, Alan, thank you for letting the c D F be able to showcase this sort of ingenuity and these thought processes behind, you know, technology as it evolves.

And one of the things that is really exciting is that, again, the C D F is completely vendor neutral. So we've got eBay harness and Google here with us today, and they're all talking about something that then perhaps they can create a reference architecture and show other businesses small, medium, large, how they have been using AI and how it's worked for them. So it's all very cyclical, all very continuous delivery, you know, and it all sort of works together.

So, uh, again, Alan, thank you for letting us showcase, uh, tech in a different way and, uh, in a space where it's safe to say things suck and that we need to fail, but also let's be creative and, you know, not, not worry about too much. Um, so, uh, C D F Continuous Delivery Foundation, we have, uh, nine projects. Christie created one Tecton, we love her and we love all the things, uh, that are happening within the space.

So if you have any desire to learn more, go to CD Foundation and check us out, join our Slack channel, check out the projects that we host and always love to have, uh, more community involvement. Absolutely. And we're thrilled, we're thrilled to partner with c D F on this.

We, I, I have fun doing these shows and if you couldn't tell, I'm, I'm real bullish on generative ai. There's gonna be a lot more coming out of it. You know, we have Text Drunk AI where we cover a lot of this now, but we, we have some stuff going on.

I was telling the panel off camera and, um, we'll have a lot to say on this very soon. But until then, hey, next month is a new show, a new topic. Great stuff.

There's actually more stuff coming outta C D F, we're gonna have some news on at the end of the month as well. Check that out. A CubeCon and is coming up.

Uh, cloud native Con Q con's coming up in November. You should start making your plans if you haven't done so already. Lori, I C D F gonna be doing something at, at CubeCon this year, in November.

So we'll definitely be hosting a CD mini summit at Open Source Summit in Europe in Bill Bao, Spain. And I'm pretty sure we're gonna try to do something at CubeCon, which is in Chicago, which is amazing, uh, this November for sure. Fantastic.

Until then, though, this is Alan Shimmel for Tech Drunk tv. I hope you've enjoyed CD pipelines in partnership with the CD Foundation and the Linux Foundation. Have a great day.

Thank you everyone. One.