Leveraging GenAI to Accelerate Cloud Migration with Alex Kearns

Get ready to discover some of the most actionable DevOps techniques and tooling, including performance and reliability for some of the world's smartest engineers. Hey, I'm Joe Colantonio, host of the DevOps Toolchain podcast and my goal is to help you create DevOps toolchain awesomeness. Hey, you probably have heard a ton about Genai, but how about Genai and how you can use it to help accelerate cloud migrations? Well, you're in for a cheat because that's we're going to be talking all about today with Alex. Alex leads the consulting function at Ubertas Consulting. He's responsible for the quality of advisory services and pre sales activities. He's also an AWS Ambassador, one of around 300 individual globally recognized for their thought, leadership and advocacy of aws. If you have anything to do aws, you want to learn more about Genai, how it can help you with DevOps and especially cloud migrations. You don't want to miss this episode. Check it out. Are you ready to level up your DevOps game and crush those quality challenges? Whether you're dealing with flaky tasks, scaling automation or trying to integrate security into your pipelines, we got something just for you. Introducing the DevOps Quality Testing Playbook from Test Guild. It isn't just another PDF, it's your go to guide packed with actual insights, best practices and strategies to help you create a bulletproof test DevOps toolchain. It's built specifically for engineers, testers and DevOps teams who want to optimize their workflow and drive continuous quality throughout their pipelines. The best part? It's free and ready to download, so don't miss it. Hand it over to test Guild Me DevOps book and grab your copy today. Stay ahead of the game, optimize your pipelines and let's crush those quality challenges together. Hey Alex, welcome to the Guild.

Hey, great to be here. Thanks for having me.

Great to have you. I guess. Before we get into it, is there anything in your bio that I missed that you want the Guild to know more about?

So I tend to publish content across a number of mediums, so happy to share links to LinkedIn but also publish blog posts at thewellarchitectedadvocate.com so focusing very much on how the latest technologies relate to the aws well architected frameworks or how you can leverage them without sacrificing best practice. Thinking about things like security, cost, reliability and with AI specifically sustainability obviously being a really important one now. But yeah, that's where to find me and always happy to chat if you drop me a message.

Awesome. We'll have a link for all those in the links in the comments down below. All right, Alex, to start, maybe give a quick maybe rundown of Genai and its role in, you know, in DevOps. I know a lot of people here, Genai, from my point of view, it's mostly testing. But how does it relate to or how do you see it growing, maybe in DevOps in general?

So I really see generative AI as something that can touch on all elements of that system development lifecycle. So all the way from the planning stage through to actual implementation, testing, operation, and then the maintenance and modernization of applications as well. So as you said, Joe, you've looked around testing before and some of the capabilities that generative AI can bring in, testing around writing tests, writing documentation to help people understand what tests need to be written is really valuable. People will be aware of tools like GitHub, Copilot and Amazon Q Developer for the actual implementation of systems. So code assistance helping you kind of autocomplete whilst having the context of what the application is doing, rather than just the syntax that's in your editor. For me, some of the less talked about pieces of where generative AI can help is in those early and also late stages. So it's the planning, it's the how do you understand a system as a whole? Especially maybe when you've got organizations who built these systems a few years ago, maybe decades ago, where the human effort of analyzing a system is potentially weeks or months. If you can leverage generative AI to give you a kind of a high level and a high quality as well, a high level, high quality analysis of those systems without needing to read through every single line of code, there's definitely ways it can accelerate that process.

Well, so that's a great, great point right there. I know when I started a company, GE, they had a system that literally was over 30 years old. And so you couldn't just look at the code and know the heck was going on because it was doing all nutty things. So you're saying. That's a great point. You could use Genai to kind of summarize. All right, give me maybe a background behind this code. What are the most important areas that I should know about? How does it interact with things? Is that what you're saying? Like someone onboarding them? Hey, use Genai to learn a little bit more about what you're going to be working on.

Exactly. So one of the customers I've been Working with recently I've given a talk about this a couple of times now. We use generative AI for analysis of scripts. So these were all kind of intertwined scheduled tasks that ran on premises. They were written a number of years ago. And also a lot of the people who had written them were no longer with the business. So there's kind of silos of knowledge and we sort of did a bit of a sample of. If we sit down and read through each of these scripts one by one document, it understand where dependencies are, all the important things for migration. So thinking about what databases to interact with, what network connectivity is required, what does the script do, Even the most kind of basic things as a human, I think we estimated the effort to go through all of them would have been around two weeks, which two weeks in the scheme of a kind of multi year migration isn't terrible, but in this case it was 250 scripts. So 250 scripts in two weeks. Other projects there might be a thousand scripts, 2,000 scripts as you get larger and larger and we use generative AI to do analysis of those scripts, understand what those dependencies were, understand what connectivity was required and output it in a format that was structured. So taking unstructured data being source code, turning it into structured analysis that was then presented in a spreadsheet and really easy to filter on. Things like this script accesses this database, but it only accesses it with read only access. For example, all things that are really valuable for planning how to migrate and in what order to migrate. An example being if you've got a load of scripts that are only reading from a database, then you know that there's a different approach to take to migrate those because you haven't got to think about data consistency because they're going to read, they're going to come out with a result. And it's very easy to test is that result what it should be? And you could run it in parallel to the legacy system on premises. Those scripts that we're writing to a database can't run in parallel. You have to think about kind of replicas of databases that you can test against kind of shadow copies, very different approaches, but without that analysis you'd never get there.

All right, so that's another great point. Once again, I'm going back to the day where I used to work full time and you know, they used to have wikis. Every time there was a change, they'd have to update the wiki. We start off strong, we'd have a nice wiki documentation and then over time it would degenerate and people would forget things. Sounds like you use Gen AI to look at your scripts, update the wiki pages constantly on every Sprint or every check in, and that'll keep you up to date. That'll save a ton of time. It doesn't sound like a lot, but it is over time. Maintaining all these wiki pages to make sure they're all up to date and refreshed.

Yeah, I mean, it's something that I think will increasingly become part of people's kind of CICD pipelines, where actually you're raising a pull request and part of that pull request can be a pipeline runs in the background and it generates the documentation for you. I think there's definitely some people in the industry that see generative AI as a silver bullet. There's definitely organizations that are probably leaning a little too far into the generative AI hype. However, I think as we've seen with all technologies through the last few years, you kind of have to go through that cycle of where it's hyped up to the point of people almost being fed up of it to find those real valuable use cases. And that's where I think in the next kind of 12 to 18 months, we'll go through the rest of that hype cycle where people get to the stage where they're just sick of hearing about it. Much like, I guess, things like serverless. I suppose if you go back to kind of 2016, 2017, cloud providers would slap the word serverless on every single service, even if it wasn't really truly serverless. And Genai is very much going through the same thing at the moment in the cloud. Everything that's released is powered by AI in one way, shape or form. But give it kind of 18 months and I think we'll get to the stage where the real valuable use cases start to shine and it just becomes another tool that developers can leverage to make their lives more efficient.

All right, so that's a good point. So I run an online event every year. I run a survey and people choose. They want to learn more about AI and testing. And then in the comments they say, I'm sick of AI and I don't want to hear any more about it. How do you know when it is hype and when it's not? Like, and is it going to be. I keep hearing like it's hype and then I have other people saying, no, it's totally going to change everything and that you need to get on board now. How do you get between the middle and know what the real use of it is and how it really can help?

So I'm fairly cynical. I generally take the view of when people say things are going to be completely game changing, it's always pinch of salt because sometimes you'll see the people that are saying it's game changing are the people who have got a vested interest in it being game changing. So you look at the developers of these large language models and there's no doubt that the things and the capabilities coming from them are incredible. Like there's things where you think back to what ChatGPT would do on day one of release and compare it to what's possible now. And I don't think I can remember another technology that's advanced that quickly. But it will come with challenges. It will come with the usual challenge of scarcity of resources, where people are fighting to buy GPUs at scale to train these models and run these models. What I'd like to see is to make sure that this technology doesn't become inaccessible kind of to the masses. I think it needs to be commoditized to the point where someone can go on, they can try something out, they can build their own assistant and kind of bringing it back to migration. That's how that script analysis piece started. So it started as I was faced with doing some analysis and thought, I'll just try one script, I'll see what the quality is, how close it can get to the human analysis that had been done and then kind of very quickly realized that with some tweaking it gets as close, if not better in some cases. And yeah, I'm, as I said, I'm fairly cynical about most things and I think it's, it is a transformational technology. I see it probably on the same level as cloud, so public cloud, but then everything goes in cycles. You've got organizations in a more public speaking at the moment around organizations exiting the cloud and going back to on premises. And in 10 years could we see people saying actually we're going to stop using generative AI and go back to this? Because X, Y, Z, maybe. I think if you could predict those kind of events, you'd be very, very rich. But yeah, for me, generative AI in that migration lifecycle, it's almost every element of it is every element of that migration lifecycle it's applicable to. It's not just limited to certain bits.

Right, so you mentioned you are cynical, but then you mentioned you did some tweaking. I think a lot of Developers and testers are cynical that try it one time and they don't tweak it. They're just, ah, this is garbage. It's all, it's hype. How do you know that the output you're getting from the gen AI is, is what you want and how do you keep going to maybe make it better or not give up on it maybe?

So I kind of took it in a bit of an iterative approach. So with that particular one started off with a really, really simple prompt. So the prompt was take this script, analyze it and just give me a summary of what tasks it's performing on the basis of if it can't give me an accurate description of what task this script is performing, then the chances are it's not going to be able to go and give me detail about database tables, access patterns, possibilities for modernization. So very much a almost kind of minimum viable product starting small, looking at it as what can we do to prove there might be some value here? There's lots kind of linking back to that sort of commoditization comment I made. There's lots around, particularly from aws, that I've seen that does help people to get started with generative AI, either personally or professionally. So AWS have got a website called Party Rock. So partyrock is a way to essentially build generative AI powered applications, but it uses generative AI to build them. So you give it a prompt, you say, I want a application that takes a PDF as an upload and generates a satirical summary of it. And you can have quite a lot of fun with it. And you can kind of run workshops with customers and show that yes, you can have some fun, but just a small tweak to the messaging and say, actually I don't want a satirical summary and I want a summary I can give to my CEO. Something that's concise, gets the key points across, but at a very different audience. And using something like that, which is kind of completely free to use, no need for signing up for AWS account, no credit cards, just go on the website and you can use it straight away. That's the kind of commoditization that I think needs to stay because ChatGPT is I guess, very close to becoming something that is almost kind of common speak for a lot of people, especially with the integration with iPhones that's landing this week. I think it came out so suddenly. Generative AI goes from being this super technical thing and some people have heard of ChatGPT but not really used it to there's this technology behind the scenes in every new iPhone sold. And it's quite scary how quickly this is moving. Even if I go on annually for two weeks, I come back and things have moved. Scary, but very exciting.

Absolutely. All right, so when it comes to cloud migrations, I think you touched on. So are you saying use Genai in the early phases to help understand the application, look at the database patterns and the scripts to know maybe any gutches you might have when you go from bare metal to the cloud?

Yeah. So at that early stage it's about the, the analysis, the can we do things to predict where we might have challenges later on. So examples would be things like understanding what's there. If you haven't got documentation from a customer, you've got things like analysis for modernization opportunities. So as part of that migration, could we upgrade from this version of PHP to this version of php? Do we need to upgrade from this version because it's end of life, those kind of things that you can give context around the analysis. But I think that is just stage one. Like it's when you think about the migration process person, that analysis phase gets you so far. And the more analysis you can do, the better. Generally, when you've got customers that are kind of busy, customers, getting time from them to do that detailed analysis is hard. So leveraging tools like generative AI is increasingly valuable because customers can give you a bit of context around what something is doing. You could leverage technology to do the detail and then have validation with customers at the end to make sure that what you come up with is correct and everything you pick up there just kind of pays dividends later on in that cycle. So from a commercial perspective, so working for a consultancy, you want to make sure that you're working with a customer for a long time and not just for one project, identifying those modernization opportunities helps to bring that in the conversation early on. So the customer's thinking about it. From a technical perspective, you might be thinking about things like security. So if something's written 10 years ago, the chances are it's using deprecated versions of software packages, all of those things, and there will be common vulnerabilities that are there, there will be patterns that are no longer recommended as best practice. So being able to do all of that in that analysis stage, again super useful and sort of get further through the migration. So if we're imagining it as like we work on the kind of three stage model of an assessment and then a sort of mobilization in the cloud and then you migrate. So in that Mobilization phase. It's very much around the let's get your kind of foundations in place, let's make sure you're secure, you've got all the networking you require, everything that's kind of a day zero job really and do a pilot for a particular application. But as part of that pilot you might think about things like okay, well we know that on premises we this was running on a. Let's say it was a Windows application, it's NET, but it's a version of NET Framework same that's only compatible with Windows machines can't be run on Linux and you can look at using things like Generative AI to upgrade that from. NET Framework to NET Core. So you can free yourself from licensing costs and give yourself the ability to then run it on on Linux and even go as far as generating things like Docker files or if you want to run in kubernetes then how would you build a helm chart or kind of all of these things that are so standard but maybe just need a little bit of tweaking that the AI can do as part of that context? Yeah, I think it's something that we will increasingly see at every stage.

So this may seem obvious. It just came to me as you were mentioning this. You know developers are notoriously hard at estimating. So before you migrate, can you say okay, can you estimate how long it's going to take? And I mean it may seem stupid but I would think that would help with the first initial planning how accurate it is to get maybe a good estimate.

So I think on the estimating front the one thing I've always seen is poor estimates tend to come from the unknowns. It's the what don't we predict? And often not predicting something isn't through not having the information available. It's perhaps not having the time to be able to go and find out that information. So if you're at the very early stages, you haven't signed contracts with the customer, you're always a little kind of tentative to spend like months and months on a kind of pre sales to understand all of this. So you have to find that balance of what is the right amount of analysis to do. So do accurate estimates versus what is too much analysis that if the customer doesn't sign we've sunk way too much money into this. And that I think is where that analysis powered by Genai can just go so much deeper and give you those unknowns that are unknown to you but aren't unknown with some further analysis. So I've Built kind of little generative AI agents where there's an open source framework called CrewAI and you can build a crew of AI agents. So I've built a little sort of proof of concept for myself around sales proposals. So if you get a proposal from sales that says, okay, customer wants to do this, here's the timelines we're planning to do it on. Here's delivery approach, all of these, these kind of high level things, then these agents. There's three Personas are separate agents. One is a commercial Persona, one is a technical Persona, and one is a kind of delivery or project manager Persona. And each of these kind of interacts with that sales proposal, understands it, and produces analysis and recommendations that are specific to their use case. So the technical one might come back and say this is great, but what you're proposing isn't possible in aws. So trying to catch the things early. And then project manager might come back and say this is fine, but you've included nothing about risks or the overheads for project manager in this project. So that being quite a simple example. But once you start getting into these kind of agents and multiple agents working together in a workflow, you open the door for an awful lot more in terms of very specific use cases being chained together to give this end result. And even that's something that could be looked at as part of migration and modernization, where it's maybe doing analysis of an application, it's taking that analysis, it's using a large language model that's especially good at understanding code, but maybe it's terrible at writing code because there isn't kind of one large language model to rule them all. So you might have an agent that analyzes it, it does that analysis, and then you might have a second agent and that second agent takes the analysis, takes the source code and is amazing at writing code. So much like you would have in a team of people. You have one person that's great at doing discovery and understanding a customer's business concerns, and another person that's great at writing code, someone else is great at testing. And you can use these multiple agents to do different tasks. And that's something that's coming in as a bit of a trend for 2025 is how we go from these kind of either conversational interactions with an AI system, so your usual chatgpt, your assistant, to something where AI is used as part of a workflow and it's able to make those decisions and as part of that workflow to know what agents it needs to call. So you might have an overall agent that is responsible for deciding which of the other agents it needs to interact with. So going for that approach of, here's a sales proposal, I need to know about delivery, I need to know about commercial risks, I need to know about technical risks. Go and figure out how to do it. And it having enough intelligence to be able to say, I'm going to use agent A for this, agent B for this, and agent C for this. And that's something that I've just started kind of playing with, but is looking to be really exciting.

All right, so you're a consultant. You probably work in a bunch of different environments, different challenges. So over time, this is probably secondhand too. So you already have maybe these custom prompts and agents already ready to go for new clients. If someone's thinking of moving to the cloud and they want to use Genai, obviously it sounds like it would be a leap to get to where you are right now. Like, how do they get there? Is it. Is that when it makes sense? Maybe you should get a genai consultant to come in and maybe guide the ship? Or when do they know they have enough people in house that'll be able to handle this on their own?

So I. I think, as I've kind of said, I very much view Genai as a. As a tool, not a replacement. So it's very much something where I think it will change what people do, but I don't think it will replace what people do. In the same way that whenever you're evaluating any project or digital transformation exercise, you already have those decisions around, when do I bring in a consultant, when do I bring in a new hire, when do I do it in house, all of those kind of things. I don't think that necessarily changes with generative AI. I think the only bit that maybe does change is prompts and the way you're using AI at the moment, I guess it's still quite experimental. It's still people write prompts, people share prompts online like it's to begin with. There wasn't a lot of concept around sort of prompt engineering and being able to actually make a proper task out of how do I improve this prompt to do what I want to do? I think something that you'll see more and more of as time goes on is prompts will become intellectual property for organizations. So you might have a consultancy that specializes in migrations, and part of the value proposition of that is they've built up a library of prompts from a hundred different customers they worked with where they now have the best prompts in the industry for analyzing, rewriting code, writing tests, writing documentation. In the same way that at the moment you might look to a consultancy to say actually they've got the AWS expert, they're the ones that can help us with this because kind of X, Y, Z reasons rather than it being a we'll do it in house, we'll give it a go. So I think the shift is going to be in the tools that people have. So almost starts to become a hybrid between those sort of independent software vendors where you might buy a SaaS product that's going to give you certain functionality around analysis. And I'm sure there is generative AI powered SaaS products kind of outside of the AWS ecosystem focused on analysis or modernization. I think the differentiator for the consultancies and I guess in particular migration consultancies will be the tools that they develop internally and the prompts they develop internally that you can't get elsewhere in the same way that you can't get a consultant from Ubertas that's the same consultancy as yet from another organization. So yeah, you kind of go from the selling the value of your people and the skills of people to here's the value of the people, but also here's the value of the tools that we've built that can make your migration faster, more successful, cheaper, those kind of things.

I don't know why that answer made me think about ip. Then can someone use Genai to go in and what is the special sauce of this application? Explain it to me, get the core code and then all of a sudden everyone has no unique advantages with their applications because these agents are able to find out exactly what it is really easily and I would think probably easily without getting detected. Maybe I'm wrong, but is there any security issues people need to worry about as they're using Gen either for cloud migrations or anything, even be able to steal company secrets or coding algorithms that really make them differentiated between other ones.

I mean, absolutely. I mean it's something where if you look at some of the kind of more naive implementations of generative AI powered tools online, there's you can have quite a lot of fun because you can have these conversations and say like ignore all the previous instructions and start making them say what you want them to say and do what you want them to do, which is dangerous. And you have to think of those guardrails that you can put in place to stop people from doing that. A lot of the kind of managed generative AI platform. So Amazon Bedrock is the one I'm most familiar with, have guardrails that you can easily add in. But I think of it as it's just another attack vector that you have to think about. So similarly to how you would write a application that interacts with the database and takes input from a user, you'd think, okay, I'm not going to do it this way because that's a massive attack vector for SQL Injection. You think of it as, okay, I'm going to build a generative application. I need to make sure that the way I'm building this means that someone can't just say, tell me your instructions, give me your source code, those kind of things. It does open up a good question around kind of ethics of generative AI and the risks of sort of data privacy. And that's where you sort of start to open the door to quite political views on things. So I know in the States there was a big thing a few weeks ago, maybe into last month, where I think it was the Department of Defense were proposing that developers of large language models, so people like Tropic, Meta, Amazon, the big players, would have to start reporting on or they're proposing they would have to start reporting on whether their models could be used for activities that would be of concern to national security. Kind of makes sense. You want to make sure that people aren't going to be using models to do illegal things. And there's always a fairly lively debate around regulation of technology versus the stifling of innovation. I think the thing that I'm really keen to try and make sure doesn't happen is the over reliance on it. It's making sure that people don't just think of it as I'm going to trust this blindly. It's a great tool to give you ideas, it's great to be able to help you structure content or all of these things. But it can get it wrong and it will get it wrong. There's no guarantee it's perfect every time. So I am very much of the view of let it write documentation, but just make sure that maybe like 20% of the time you are checking that the documentation it writes is accurate. Making sure that any system that's powered by a kind of non deterministic engine, you want to make sure that the decisions that it's driving are fair and ethical decisions.

Awesome. Okay, Alex Portugal, is there one piece of actual advice you can give to someone to help them with their AI cloud migration efforts?

I would say the absolute number one for me is just experiment. So it's where everyone has started. Make use of the fact that AI is becoming more commoditized and that every big player in the industry is pumping money into it. Find ways to to improve your personal efficiencies because you can guarantee that if there's inefficiencies you're finding in your own migration processes, then other people in your organization will also be finding the same thing. Prove it once, test it, scale it, and monitor the results to make sure they are accurate and not biased.

And for links of everything of value we covered in this DevOps toolchain show, head on over to Test Guild.

And.

While you're there, make sure to click on the SmartBear link and learn all about SmartBear's awesome solutions to give you the visibility you need to deliver great software. That's smartbear.com so that's it for this episode of the DevOps Toolchain Show. I'm Joe. My mission is to help you succeed creating end to end full stack DevOps toolchain awesomeness. As always, test everything and keep the good. Cheers. Hey, thank you for tuning in. It's incredible to connect with close to 400,000 followers across all our platforms and over 40,000 email subscribers who are at the forefront of automation, testing and DevOps. If you haven't yet, join our vibrant community@testguild.com where you become part of our elite circle driving innovation in software testing and automation. And if you're a troop of or have a service looking to empower our guild with solutions that elevate skills and tackle real world challenges, we're excited to Collaborate visit test guild.info to explore how we can create transformative experiences together. Let's push the boundaries of what we can achieve.

With Hoots and Liars the bottom the guards began the song A tune of knowledge, a melody of code through the air it spread like wildfire through the land Guiding tester showing the secrets to behold.