How AI Powers DevOps and ITSM at SKILup Days 2024

Hi, I hope you're having a fabulous skill update today. I'm Helen Beal. I'm gonna be talking about how AI powers DevOps and ITSN today.

Um, so as I said, my name is Helen Beal. Uh, I'm head of the Ambassador program here at People Cert. Um, as you know, this includes ambassadors for DevOps Institute, ITIL Institute, and Language Cert.

I have a number of other jobs as well, but today, let's get on with the content. So let's talk about the relationship, first of all, between DevOps and ITSM. So they're very closely related.

In fact, the founder of DevOps Institute, Jane Grohl, used to talk about the harmonious polygamous marriage of ITSM Agile and Lean and how that led to DevOps, a kind of love chart, if you like. So there is a lot of crossover as the naturally is in our world with all these different ways of working. They all intersect and interconnect, but let's try and kind of pull apart DevOps and IT TSMA little bit.

So DevOps really emerged when Agile was being used in software development, but not on the IT operation side of the house. And that difference caused a lot of friction and DevOps was born to try and resolve that friction. And DevOps now focuses really on this rapid and reliable.

So it's about throughput and quality of software and services, whereas ITSM is really about the effective management of those IT services on the IT ot operations side primarily, but increasingly on the development side. And actually we're connecting DevOps and ITSM more and more. DevOps has an agile iterative meth methodology, and ITSM is a bit more structured and process oriented.

So if you use a framework like itil, you'll be very familiar with this emphasis on defining processes and SLAs and having, uh, instant management and other management routines to ensure that stability and reliability that we're looking for in our runtime services. The goals of DevOps are about accelerating software delivery and improving quality, whereas IT services or IT service management really wants to ensure that those services are aligned with business needs and deliver value and have the high levels of availability and performance. So there is some crossover.

Um, we do want quality availability, performance stability on both sides of the house. There's a lot of cultural emphasis in DevOps. Um, whereas ITSM is perhaps a bit more about adhering to the processes and having controls and accountability around them to make sure that things don't go wrong or if they do go wrong that we know quickly how to fix them.

We have loads of tools in DevOps that any of you DevOps engineers out there will be very familiar with. Our DevOps tool chains, uh, are often quite complex, um, heterogeneous environments of, of many different things, connecting together to, uh, to help us automate the root of software from ideation to customer experience. In ITT m we have mainly tools around service desk management, which are of course a component, part of our DevOps tool chain as well.

Um, and then within that we have capabilities that manage things like instant management. So there's a lot of crossover, um, between the two. But I've got a few things which were a little bit fun to say.

And here's one which I wanted to talk a little bit here about this, this differentiation's like, oh, we recognize DevOps stuff. Oh, we recognize I ts M stuff. We also have these types of AI that we now use with machine learning, which is basically in this case, we'll be showing, um, the AI hundreds, thousands of different pictures of things, um, of cats and dog for examples so that eventually it knows, um, it can look at any picture and say, oh, yeah, that's a cat, or that's a a dog.

Um, and then what we've got here is a robot. So actually I'm a bit more of a dog robot. Um, like it can be cat people or dog people.

Here. We've got a robot that prefers dogs a little bit as well, but very interesting, um, way that machine learning works and of, of course you can, there's some great tools online. I did a, a really good course last year, um, which is Google's AI for anyone course.

And in it takes you to a tool that you actually, um, train the machine learning to recognize tic-tac toe. So you actually do tic-tac toe into your webcam over and over and over and over, um, until it learns what the different things are, and then you can put anything in front of it and you can do it with pens and books and glasses and all sorts of stuff. Um, it's really good fun, really good application of ai.

Um, and we do have these different types of ai. So I've just really been talking about that kind of machine learning, that pattern matching type AI there, um, which is our more of our traditional ai. So this is what we've used a lot in the past.

Um, and you know, there's a lot of debate. It's like, what's the difference between automation and ai? When does it trip over from being like a thermostat that senses that the house isn't warm enough and has some script in it that tells it to actively turn the heating on to something that is like an AIOps tool, which is being fed data from multiple sources of monitoring and observ observability tools, um, and analyzing all of that data to look for anomalies to alert you that there might be an instant or a problem happening and to, to show you where it's happening.

So you to do some RCA or some records analysis. So we've got this kind of traditional algorithmic ai, and then we've got the thing that's really created all the noise in the last 12 to 24 months, which is the generative ai. So the chat, GPT open AI type tools, um, which have different goals.

So the goal of the generative AI, or gen AI, as we often call it, um, is to create the new and original content. So it is text and images and it wants to answer questions and things like that for us. Um, whereas in traditional ai, we're really, uh, analyzing data to help us do things like forecasting and predictions and things like that.

We have to supervise the learning quite a lot in the traditional ai, but actually in the generative ai, um, it's much less supervised. It can do much more on its own and find much more data on its own. We get different outputs.

So, um, the outputs in our traditional are much more, um, predetermined. So we're expecting we, you know, they're gonna fall into a certain number of classes that we've defined up front, whereas the output though, comes out of generative AI is a lot more novel and diverse and can often look very creative as well. And traditional AI consequently isn't particularly adaptable, whereas generative AI can do some quite surprising things with different prompts and things, um, with exhibiting this kind of degree of generalization as well.

So with traditional ai, we're typically using that for tasks like classifying things, predicting outcomes, optimizing processes and, and ways of doing things and, and various different levels of automation across lots of industries. Whereas Gen AI is, is kind of more in the creative fields, um, overall and for our kind of general, uh, discussions and, and generating documentation such like, but there's a lot of fear out there around AI as well, like whether it's actually gonna be good for us. Um, the, we are inventing it as humans to as a service to ourselves to make our lives easier.

But there's quite a lot of fear not help by science fiction books that have been going back for many decades that, um, you know, I was gonna say predict or, um, suggest imagineer, that there might be futures where, um, the AI could go a little bit wild and, uh, not be helpful for us at all. But there's already many, many different, um, use cases today out there of AI doing incredible things, um, incredible things in developing worlds to help people, uh, find, um, uh, or diagnose eye conditions and eye diseases, um, out in the field using mobile equipment where there aren't very many hospitals, for example. Um, lots of things supporting oncologists.

Um, today there's loads of, um, ways in which, uh, we are shortening, um, the work that we do and making things more accessible to everybody by, um, layering on top of our data these capabilities and pattern matching, you know, even when you're just buying a book from Amazon and it recommends you another book or a product that, um, that you might be interested in, that's ai, um, using predictive services to help us there. I sit with Gemini Open, um, pretty much all day every day now. And whether I'm doing creative work, um, and asking it research questions perhaps, or I could be doing, um, more technical work and asking it to, to help me with definitions or coding or test examples and things like that, there's lots of ways to use it.

This is my favorite AI use case, and it's a nice example of bringing together that traditional algorithmic AI alongside the generative ai. So this is a beautiful book. Um, it's a nonfiction book, um, by a, um, a study of, uh, a ologist, a study of Ians, which are the, the whales and dolphins.

Um, Tom must, Tom's actually quite famous on YouTube. He can do a search. We do a search for, um, whale lands on Kayak.

You'll probably come across his YouTube video where he was actually kayaking with his friends, um, in a bay in California, very famous for having lots of humpbacks, and the whale did actually jump out of the water and land on his kayak with Charlotte. So, um, they thought that they'd had it, but in fact they both survived. The whale just kind of slightly missed, and probably on purpose, possibly on purpose, read more about it in the book.

Anyway, Tom was interested in Wales before, that's why he was kayaking in that particular zone. Um, but this did kind of take his interest to another level, this near death experience. And actually he was approached by, um, a couple of the guys behind the, uh, center for Humane Technology.

If you've ever watched their film, the Social Dilemma, you'll know exactly who they are. If you haven't, I recommend, uh, searching either out the Center for Human Technology and or watching the Social Dilemma, which talks about how AI in our social isn't particularly healthy. We're not going down that road today.

We're gonna stay firmly on our, uh, AI road here. Um, so what these guys are doing is they have got a project called Project Seti, um, like citation, and they are trying to learn how to speak with whales. And there's actually a few projects going on around the world, right, this at the moment, learning how to speak to animal species.

And they're all kind of similar because what they're doing is, um, trying to build language models of the different creatures. Um, it, it may sound like, oh, that's like a fun thing to do, but actually quite seriously, it's a really important thing to do because species are under pressure all over the world. We're in something called the Anthropocene, which, uh, basically is a time of the world where we are more threatened, um, uh, or more species are threatened due to human activity than ever happens before.

So it's just, it's a, we're in a period of species mass extinction at the moment for various different reasons. Um, like there's a lot of humans and we're encroaching into their territory and we're using more land for farmland and we're creating climate change and all sorts of different factors are happening. But the bottom line is there's a lot of species going extinct.

So actually being able to talk to those animals about what they need and things like that might help with that and keep the ecosystem in balance, because if we don't, there's a good chance that humans will go extinct as well. So I think we probably all know, for example, pollinators like bees. If we don't have pollinators, uh, anymore, we won't have any crops and we won't have anything to eat, and therefore we will also, um, suffer.

So there is a point other than it'll just be fun to talk to these animals. Now what we've had, um, in the past is the way that we've been using translation services is, uh, when we first built translation software, basically we would build a dictionary in English, for example, in a dictionary in French, and we would teach the computer that in a word like window in English is a finera in French. And then we teach it different like ways to construct sentences.

And it was all very complex. But then we invented machine learning. So then what we did was we fed, um, our AI machine learning, uh, lots and lots of examples of French, uh, text, and same for English.

And the AI itself would learn the patterns. So what you're seeing on the right here is just an example of, uh, word relationships in English. So, um, for example, the word queen, you could say, well, like it's the opposite of king.

Or you could say, well, it's usually associated with a, a head of the royal system or the head of state, um, and female. And that's the queen. And this is the kind of the difference between doing the literal translation, um, between, um, queen in English and brand in French.

It's actually the different machine learning. So they know it by the pattern, not from the word. So if they see a pattern in that sentence, they'll uh, know what that is.

Um, and the point of this is it's a much more accurate way of translating and it's a much more, um, it's, it's easier to do more languages with it because what you're doing is you're pattern matching. You are not doing the training at the word and the dictionary and the semantics and the, and the syntax levels. Um, you are actually letting the computers themselves figure out these patterns.

And we can do it with animal language as well. So what we have to do is gather all that data. So we have this other thing right now called citizen science, which is great 'cause it means all of us can take recordings, um, and upload them into various different systems and share them around the world, um, and gather loads of data that way.

So the steps go along, get all the data of the whales singing, which is how they tend to talk with each other. They're very, very loud as well, one of the loudest screeches on the planet. Collect all that data, apply the machine learning, learn the language, then potentially, um, build machines that can talk back to the whales so we can have conversation.

And then I did promise the connection with gen gen ai. Um, mine are in my cupboard at the moment, but if you imagine you with your meta, uh, wayfarers on with your augmented reality sunglasses and you are out on your boat, um, off the coast of Costa Rica and you're looking at a humpback, your a Wayfair things could tell you a story about that whale based on the data that it knows. So it could tell you what the whale's name is, how old it is, um, but whether it's maybe had calves, whether it's got a calve now, where it was last week, where it was last year, where it was born, um, you know, all sorts of different information you could learn, um, about it.

And you could do that in your back garden, potentially, like looking at a butterfly or a bird or whatever. So could be very interesting. So that's kind of my favorite current, um, application of both types of ai.

But let's talk a lot a little bit about where it's at right now in terms of what we are doing in our organizations. So as I said, we've got these two different sorts of ai. So our traditional algorithmic AI is now here we are saying on this, uh, this, uh, curve of, uh, adoption.

So we're at the slope of enlightenment. So we're heading up towards the plateau of productivity and we've gone through the trough of dis disillusionment. So this is the type of AI that's running AIOps predictive, um, selling online applications and those examples I've given you already.

But our generative AI, that chat, GPT, the thing that's gonna tell us stories and make art and things like that, this is just coming off that peak of inflated expectations. So it's caused a huge amount of, of chatter, um, loads and loads of hype. People are using terms like AI crowd out, uh, which I personally have experienced, um, in my work with value stream management, for example, which has been a tough year value stream management because no one's really wanted to talk about it because everybody's just been talking about ai, ai, ai.

That's all everybody wants to talk about. But let's get to the crux of what it can do for us today, um, because it's gonna come down that peak of inflated expectations. We are gonna hit this trough of disillusionment.

I have it already some days. As I said, I have Gemini open, um, all day every day, and it by no means gets everything right. The other day I was working on something to do with creative writing and novels and I was thinking about stakes and different novels, and I asked it what the stake of, uh, a particular character called Dexter Mayhew, um, was in, um, a book called One Day that many of you may be familiar with, with it was made into, or new TV series was made for it this year.

And the AI gave me an answer and I had to say to the ai, but that's not Dexter's stake, that's, uh, that's, um, oh, I just forgot, forgot her name. Emma's, Emma's stake, of course. So Emma, the other character, and it did say, oh yeah, you're right, that is Emma's stake.

And I was like, what's Dexter? And it had another crack at it. So this is like the big warning really when you're using Jenna io for everything is you cannot believe verbatim what's coming out of it at the moment because it's just trying to train off it what it can find that we already know.

Um, and we are fallible as well, but we'll get there. So it'll take a while, but we'll get, um, to a plateau, plateau of productivity, uh, with it over the next few years. And that's plenty.

We can do it today, which is, uh, useful. But we just take a look at some research quickly and understand a little bit about where the market is at the moment so you can actually position yourself as well. So this, uh, question here is what stage best describes your organization's current adoption of AI for it?

Uh, this is from the state of ai, uh, AI in IT report, uh, here. So you can do, um, a search for that and find that and look through the whole report. And what we're saying is that 12% have no investment and then we've got 21, 24, it's about 45% that are kind of in the plan.

And early expiration, 14% actually piloting, um, 28% that are actually integrating at the moment. So, um, you know, really on our way at the moment, by no means, uh, mainstream or fully deployed, and this is one of the things that we are afraid of as well, is this possibility that we won't be needed anymore and we won't have jobs anymore. Um, the thing is, we've been saying this for a really long time as we've been inventing technology and we've been inventing automation, and we actually, what happens is we still find plenty of things to fill the day with and people still expect us to work, um, at least eight hour days, at least five days a week.

So, um, they're unlikely to replace us. We're just gonna find other jobs to do. But of course, if you want to future-proof yourself, um, I would suggest you learn more about AI today, which you are because you are here.

So congrats. We are on the road, um, to upskilling into ai. Um, this is an interesting one I think here.

So the question is, if your organization is investing or planning to invest in AI initiatives and tools, where did the requirement originate? And nearly three quarters of the people said it. So this speaks to, um, a phrase that I often use, which is the geek has inherited the earth.

Um, and the point of this is that 40, 50 years ago technology or it wasn't really a thing, um, and then we became a thing, um, but we became very much a cost center and order order taker in organizations. Then around 1995, the internet arrived, and now we live in a digital economy. And that has switched it from being the sort of dungeon dwellers that we are used to seeing in programs like the IT crowd into strategic enablers.

And we are used to now seeing, um, banks have more develop, some banks have more developers than Google is a stat that's often, um, banded around. Um, but certainly we are now strategic enablers and strategic drivers. So it's no real surprise that we are driving AI adoption in our organizations as well.

And back to that idea about whether we are gonna be replaced by robots, we are not probably gonna be replaced by robots because we'll find other things to do. Um, but the idea is that we are creating technology that can work alongside us, the man and machine in tandem. Um, and actually it should be making our lives more joyful.

It should be reducing our cognitive load, should be making work easier for us, um, so that we can have a swim in the back of our car on our way to work. I love that idea. Um, but perhaps, you know, we can do, we can do more fun things and have some of that time back, that would be nice, maybe the four day week.

It's been piloted around Europe in the lately and it's gone really well. But, um, that's the topic for another day as well. So the kind of benefits that people are seeing, um, the top one, that data analytics, our traditional stuff, the chat bot is next.

Um, very popular in ITSM much here as is the workflow automation. Um, we're seeing some people's reporting employee improved employee experience, and then we've got predictive maintenance and security. So again, giving us a handle over when we need to make changes in our environment.

And then at the same level, our IT infrastructure management, um, concerns, data security costs, inaccuracy, our lack of experience, governance priorities, competing with each other, and privacy, um, privacy of individuals and customer data security, quite similar things. So we do have concerns, as you can tell here. Um, you'll notice that these numbers don't add up to a hundred percent because what you're actually seeing here is that 42% of respondents said they were concerned about that.

So, you know, 80, you could conversely say that 80% of the respondents aren't concerned about, um, privacy of individuals, but certainly not naugh percent across the board. People do have concerns about AI adoption. So let's have a think about how all of this locks in to our DevOps and ITSM world.

So here's an example, a a an abstracted level, um, thematically what a DevOps tool chain looks like. I've got it into five sections to try and simplify things. As I said earlier, DevOps engineers will know how many tools, um, can exist in a DevOps tool chain, but if we simplify and abstract it and the thematic it this way, it helps.

So our first section is about portfolio and backlog, and then our second section is continuous integration, where we create our code and incorporate artifacts, control versions, and build that code trunk based manner. And then we've got our testing, our delivery. And then, uh, this is kind of leaning into our ITSM piece here.

So the way that we are releasing and operating and then the way that we are running and observing, uh, and getting feedback to go back into, um, our vision and goals and aligning to what we're going to do next. So if we then take that and we can add on, um, the ai. So these are all the AI use cases that you can add to these different areas.

So we are actually gonna go through each one in a bit more detail. So I'm gonna skip straight ahead into portfolio and backlog. So the kind of things that we can do here and use AI to do, um, is we can do intelligent prioritization.

So we've got all of that data and in, you know, very large organizations where they've got portfolio management, those portfolios have got thousands of projects, thousands of products, and maybe thousands of platforms in there as well. And we need to understand where the greatest impact is gonna be. So using AI to help us give us insights there is incredibly important.

Um, and this can also help us with risks as well. So it can help us manage resources and help us identify where there might be a dependency, for example, that might cause a problem downstream for us. Um, that those resources we can use AI to, uh, help us resource with resource management and automate a lot of our routine tasks that we are doing, like progress updates and, and status updates and the entering data.

Um, it can help us with documentation, drafting emails, product su, project summaries, all sorts of different things. Um, the boring stuff. So let's get the machines to do the boring stuff so we can do the more interesting things.

Um, and we can look at our historical data and give us some, uh, forecast and predictions on what's gonna happen next, which will help with our decision making and the way that we balance our portfolios. The next stage on our five step DevOps tool chain is the continuous integration. So we've got tools already, like copilot, which many people are using today, um, which can save us a lot of time when we are coding and developing.

They can suggest code snippets and complete functions and even generate entire classes based on the context of the code being written. I can generate, uh, the code from natural language so developers can, can describe what they want in just like the way that they speak and the AI can actually make it into code for them. The AI can help us a lot with quality.

We're gonna move on to testing, uh, a bit more in a moment, but we can do refactoring, code refactoring and look at code smells, um, redundant, find redundant code, and look at potential performance bottlenecks. So, um, you can even take it a step further and allow it to automate, uh, your refactoring if you want to. Um, code search is a lot more powerful with AI because we, you can use natural language processing NLP and machine learning, which helps us more with the context and the intent behind search queries that we're doing.

Um, and similarly it can analyze the code and generate our documentation again. So, uh, I particularly like writing, uh, uh, content and using words, but I know that I'm in the minority for that. And developers in particular, lots of developers I've met really don't enjoy doing documentation, but our AI can help us with that so you don't have to to do it anymore.

Um, and it does it pretty well actually. It can do it during code reviews and uh, even do some translation for you, like really technical jargon into, uh, more, uh, plain language that can be used, um, uh, around the team if you are trying to share information about what you've built with people that perhaps aren't as technically minded as you are. And then we can also use it for intelligent build management as well, so it can optimize build configurations and dependencies and reduce build times from a testing perspective.

There is so much we can do here in terms of analyzing the code, change changes and uh, user stories and automatically generating test cases. Um, executing those tests in smart ways and analyzing the failures, identifying root causes, um, and then even more doing different code analysis at different levels. So, um, static code analysis, dynamic code analysis, it can look at, um, all your security stuff as well.

So it's got the capability of ex planning, like vast amount of security data, um, which really helps in a world where actually we're being attacked now by, uh, AI as well. So it's a bit of a battleground out there in terms of cybersecurity and you probably want some AI on your side to beat the AI that's on the other side from a delivery perspective. Um, again, we can use this, uh, technology to help us identify risks and warn us, uh, towards where things are likely to work or likely to not work and help us, uh, do smart releases or intelligent release orchestration.

Um, so we can really leverage things like canary testing, um, uh, to really help us control deployment and help us make sure that we don't mess up or if we're messing up we can UNM mess up very, very quickly, um, with our things like omic remediation. Um, and then we can do also things like analyze the pipeline execution data to help us identify where we've got bottlenecks and delays and waste in that um, level, which also, uh, speaks to another part of my world that I mentioned earlier, the valley stream management part. So this is where we can layer AI on top of the flow of our pipeline and discover where we can improve our pipeline.

So another question that we asked or was asked in this survey, uh, was whether a IT team was using, uh, chatbots and uh, most people said yes, it helps. Um, and then the smallest slice there is actually yes, and I wish they didn't. Um, and we've got a no and I'm happy they don't.

That's the slice that's kind of coming out. And then a few people not using ai. So you know, nearly half of people are saying, yeah, chatbots are good, they're really good.

I think when you're doing things like, uh, password resets and, and the really kind of like kind of standard day-to-day grind work, obviously when things, uh, become a bit more complex, we probably want a human at the end of it. And then, uh, finally our final section of our five step tool chain is around, uh, what's happening in terms of observing our runtime. So of course AIOps here, it's been around for quite a few years now.

Um, it's pretty easy entry, entry level stuff. It can really help if you've got a lot of monitoring tools, which most people do. Um, we've got some really great chat bots that can help here as well, that can help prevent the potential problems before you get any service disruption.

And we can then use it more on the customer facing end as well in terms of like customer experience, not just availability and whether there's problem in the instant, but actually how much people like what they're working with or what they're using and consuming in terms of your products and services. Um, and actually power sentiment analysis using LLP, um, across things like, uh, social platforms, um, as well as uh, referrals and reviews and things like that. Um, and also use it to help with the personalization recommendations that we talked about earlier on.

And this isn't just for Amazon, this is for you too. This is, um, you can use this inside your ITT SM tools for things like service requests and instant re resolutions as well. Um, and then we can use it across things like customer, uh, journey mapping as well and look at chat trends across multiple channels in the omnichannel, uh, as well.

So very, very powerful stuff. I mentioned value stream management. If you've got a digital value stream, um, it looks a lot like a, a devil's tool chain We d enough.

Um, so you can do lots of different things like you can map and visualize, uh, what's happening using AI and your IT tools connected to your devil's tool chain. And you can use patent detection, um, as I've just described, to identify where the waste, the waste is the same for bottlenecks. Um, you can prioritize work based on customer and business value outcomes expected and you can continuously improve.

So you can continuously aim to, um, reduce waste in your process and inject innovation capacity. There are places people don't want AI to go. Um, mainly ethical people, customer relationship things, uh, some sensitivity around customer data, ip, um, and kind of a strategic planning that's not really for machines is that that's for us to do, that's for humans to guide and lead, um, and we don't wanna hand over control, um, total control of quality to the AI as well.

So most people are saying that they're not concerned about how their organization is using ai, so people are getting pretty comfortable with it, um, as we move forward. So hopefully you are as well because AI is powering also the delivery of AI solutions. If we look at things like gen AI ops, we are using DevOps principles to help us deliver gen AI tools in our organization.

So doing things like lifecycle management, leveraging data ops and LLM ops, um, and model ops as well. And then wrapping that with security and governance risk and compliance, uh, around it to help with those concerns around ethics and privacy. Um, so I hope you've enjoyed my talk today.

Um, I am around for chat and uh, it's been lovely to have this opportunity to speak with you and I hope you are enjoying the rest of your skill update and I will see you somewhere sometime soon. Thank you very much.