Why AI Cannot Replace DevOps Engineers (New Study)

Is AI going to replace DevOps jobs? This Is AI going to replace DevOps jobs? This is one of the most common questions I is one of the most common questions I is one of the most common questions I get and it's all over my YouTube get and it's all over my YouTube get and it's all over my YouTube comments. So in this video, I'm going to comments. So in this video, I'm going to comments. So in this video, I'm going to attempt to answer this question. And I attempt to answer this question. And I attempt to answer this question. And I have some recent data that is going to have some recent data that is going to have some recent data that is going to help you sleep better at night. So let's help you sleep better at night. So let's help you sleep better at night. So let's jump in. The data I want to share with jump in. The data I want to share with jump in. The data I want to share with you today is from an article by Quesma you today is from an article by Quesma you today is from an article by Quesma who created a very interesting benchmark who created a very interesting benchmark who created a very interesting benchmark that is very relevant for us DevOps that is very relevant for us DevOps that is very relevant for us DevOps engineers and SREs. engineers and SREs. engineers and SREs. Now what they did was they asked 40 Now what they did was they asked 40 Now what they did was they asked 40 models to add distributed traces to models to add distributed traces to models to add distributed traces to existing code bases using a standard existing code bases using a standard existing code bases using a standard method. Basically, the benchmark is method. Basically, the benchmark is method. Basically, the benchmark is giving them basic SRE tasks and then giving them basic SRE tasks and then giving them basic SRE tasks and then asking the AI models to solve them. And asking the AI models to solve them. And asking the AI models to solve them. And the results will surprise you. But the results will surprise you. But the results will surprise you. But first, the article does a really good first, the article does a really good first, the article does a really good job at explaining what distributed job at explaining what distributed job at explaining what distributed tracing is. So when an app runs on a tracing is. So when an app runs on a tracing is. So when an app runs on a single machine, you can often trace an single machine, you can often trace an single machine, you can often trace an error by scrolling through a log file. error by scrolling through a log file. error by scrolling through a log file. But when it runs across 50 But when it runs across 50 But when it runs across 50 microservices, which is very common in microservices, which is very common in microservices, which is very common in modern cloud-native modern cloud-native modern cloud-native application landscapes, that single application landscapes, that single application landscapes, that single request gets scattered into a chaotic request gets scattered into a chaotic request gets scattered into a chaotic firehose of disconnect events.

firehose of disconnect events. firehose of disconnect events. Distributed tracing solves this by Distributed tracing solves this by Distributed tracing solves this by linking them back together, allowing you linking them back together, allowing you linking them back together, allowing you to follow a user action like clicking to follow a user action like clicking to follow a user action like clicking login as it jumps from the API gateway login as it jumps from the API gateway login as it jumps from the API gateway to the auth service to the database and to the auth service to the database and to the auth service to the database and back. So in other words, if you just back. So in other words, if you just back. So in other words, if you just have a simple application, have a simple application, have a simple application, you can just look at the log files if you can just look at the log files if you can just look at the log files if something goes wrong. But if you have something goes wrong. But if you have something goes wrong. But if you have tons of microservices, you need some way tons of microservices, you need some way tons of microservices, you need some way to track those requests across these to track those requests across these to track those requests across these services. And in order to do this, you services. And in order to do this, you services. And in order to do this, you need something called instrumentation need something called instrumentation need something called instrumentation and the de facto standard for this is and the de facto standard for this is and the de facto standard for this is OpenTelemetry. So what they did was they OpenTelemetry. So what they did was they OpenTelemetry. So what they did was they created a benchmark for OpenTelemetry created a benchmark for OpenTelemetry created a benchmark for OpenTelemetry instrumentation. So basically, the instrumentation. So basically, the instrumentation. So basically, the implementation of this distributed implementation of this distributed implementation of this distributed tracing. And they tested 14 frontier tracing. And they tested 14 frontier tracing. And they tested 14 frontier models on these tasks. And the first models on these tasks. And the first models on these tasks. And the first task that it that they tried was as an task that it that they tried was as an task that it that they tried was as an example uh start with a basic task by example uh start with a basic task by example uh start with a basic task by which is adding instrumentation to a which is adding instrumentation to a which is adding instrumentation to a single microservice in a single single microservice in a single single microservice in a single language, okay? The AI agents get a language, okay? The AI agents get a language, okay? The AI agents get a small microservice with around 300 lines small microservice with around 300 lines small microservice with around 300 lines of code from a realistic application and of code from a realistic application and of code from a realistic application and work in a Linux terminal and work in a Linux terminal and work in a Linux terminal and get the task of implementing tracing for get the task of implementing tracing for get the task of implementing tracing for this. And this makes me sleep a lot this. And this makes me sleep a lot this. And this makes me sleep a lot better because I am actually not afraid better because I am actually not afraid better because I am actually not afraid at all of AI replacing my job. As they at all of AI replacing my job. As they at all of AI replacing my job. As they say in the article here, they were say in the article here, they were say in the article here, they were surprised that even the top models of surprised that even the top models of surprised that even the top models of 2026 struggle. The tasks were we 2026 struggle. The tasks were we 2026 struggle. The tasks were we proposed were trivial compared to proposed were trivial compared to proposed were trivial compared to real-world scenarios, okay? In a typical

real-world scenarios, okay? In a typical real-world scenarios, okay? In a typical SRE job, services are massive, legacy SRE job, services are massive, legacy SRE job, services are massive, legacy written, and poorly documented. If written, and poorly documented. If written, and poorly documented. If models fail on 300 lines of clean Go models fail on 300 lines of clean Go models fail on 300 lines of clean Go code, they cannot handle production. code, they cannot handle production. code, they cannot handle production. It's not me saying this, it's in the It's not me saying this, it's in the It's not me saying this, it's in the article, okay? So, in other words, article, okay? So, in other words, article, okay? So, in other words, they're basically saying the AI models they're basically saying the AI models they're basically saying the AI models cannot handle production yet. Now, why cannot handle production yet. Now, why cannot handle production yet. Now, why is that? In order to understand this is that? In order to understand this is that? In order to understand this properly, you need to know a little bit properly, you need to know a little bit properly, you need to know a little bit of AI theory here. And I've been of AI theory here. And I've been of AI theory here. And I've been experimenting with this over the past experimenting with this over the past experimenting with this over the past few months a lot, and the core concept few months a lot, and the core concept few months a lot, and the core concept you need to know here is what is called you need to know here is what is called you need to know here is what is called a context window. So, when you send a a context window. So, when you send a a context window. So, when you send a prompt to an LLM, which is running in prompt to an LLM, which is running in prompt to an LLM, which is running in the cloud somewhere you you only can get the cloud somewhere you you only can get the cloud somewhere you you only can get to send a certain amount of tokens. to send a certain amount of tokens. to send a certain amount of tokens. There's only a certain amount of data There's only a certain amount of data There's only a certain amount of data that you can give to an LLM, which it that you can give to an LLM, which it that you can give to an LLM, which it then can process. And this is called a then can process. And this is called a then can process. And this is called a context window. Now, Claude Opus 4.5 has context window. Now, Claude Opus 4.5 has context window. Now, Claude Opus 4.5 has a context window of 200,000 tokens. So, a context window of 200,000 tokens. So, a context window of 200,000 tokens. So, in other words, you can only give it in other words, you can only give it in other words, you can only give it 200,000 tokens. But, if you have been 200,000 tokens. But, if you have been 200,000 tokens. But, if you have been doing any experimentation with AI doing any experimentation with AI doing any experimentation with AI yourself yourself yourself and if you have been keeping track of and if you have been keeping track of and if you have been keeping track of your context window, your context window, your context window, then you also may have noticed that when then you also may have noticed that when then you also may have noticed that when the context window is about 40% full, the context window is about 40% full, the context window is about 40% full, the performance of the AI model the performance of the AI model the performance of the AI model basically breaks down. Even though the basically breaks down. Even though the basically breaks down. Even though the window is 200,000, window is 200,000, window is 200,000, you can only really give it about

you can only really give it about you can only really give it about 100,000 and have the model work 100,000 and have the model work 100,000 and have the model work efficiently with that. So, what I have efficiently with that. So, what I have efficiently with that. So, what I have noticed in my own experiments is I will, noticed in my own experiments is I will, noticed in my own experiments is I will, for example, give my AI assistant a job for example, give my AI assistant a job for example, give my AI assistant a job on my home lab, on my home lab, on my home lab, and I will tell it specifically that and I will tell it specifically that and I will tell it specifically that it's not allowed to apply manifests to it's not allowed to apply manifests to it's not allowed to apply manifests to my Kubernetes cluster directly. It must my Kubernetes cluster directly. It must my Kubernetes cluster directly. It must always use GitOps. And when I give it a always use GitOps. And when I give it a always use GitOps. And when I give it a very complicated task, it will take very complicated task, it will take very complicated task, it will take those instructions, and it will start those instructions, and it will start those instructions, and it will start exploring my cluster, exploring my cluster, exploring my cluster, and it will get a lot of context and a a and it will get a lot of context and a a and it will get a lot of context and a a lot of information that it needs to lot of information that it needs to lot of information that it needs to process. process. process. And then, because the context gets full, And then, because the context gets full, And then, because the context gets full, I have noticed that I have noticed that I have noticed that it will start ignoring instructions. It it will start ignoring instructions. It it will start ignoring instructions. It will literally say, "Applying to GitOps will literally say, "Applying to GitOps will literally say, "Applying to GitOps doesn't work. Let me just apply it doesn't work. Let me just apply it doesn't work. Let me just apply it directly to the cluster." It will directly to the cluster." It will directly to the cluster." It will literally say that. And then I have to literally say that. And then I have to literally say that. And then I have to like control C and stop it because it is like control C and stop it because it is like control C and stop it because it is basically making these decisions on its basically making these decisions on its basically making these decisions on its own, which can do a lot of dam- damage own, which can do a lot of dam- damage own, which can do a lot of dam- damage potentially. And I've seen that happen potentially. And I've seen that happen potentially. And I've seen that happen with my own eyes so many times. Real with my own eyes so many times. Real with my own eyes so many times. Real quick, 73% of you guys are not quick, 73% of you guys are not quick, 73% of you guys are not subscribed to my channel, so please subscribed to my channel, so please subscribed to my channel, so please subscribe right now. It's a free way to subscribe right now. It's a free way to subscribe right now. It's a free way to support the channel, and it allows me to support the channel, and it allows me to support the channel, and it allows me to put out free content like this every put out free content like this every put out free content like this every single week. So, the context window is single week. So, the context window is single week. So, the context window is basically everything that the model can basically everything that the model can basically everything that the model can hold in its in its awareness at one hold in its in its awareness at one hold in its in its awareness at one point in time. And humans are the same.

point in time. And humans are the same. point in time. And humans are the same. I believe we humans can only remember I believe we humans can only remember I believe we humans can only remember seven things at the same time, and seven things at the same time, and seven things at the same time, and that's the limit of working memory, that's the limit of working memory, that's the limit of working memory, uh or at least average people can only uh or at least average people can only uh or at least average people can only memorize seven things at the same time. memorize seven things at the same time. memorize seven things at the same time. There is a certain limit to what you can There is a certain limit to what you can There is a certain limit to what you can hold in your in your attention span in hold in your in your attention span in hold in your in your attention span in in one instance. And this is what is the in one instance. And this is what is the in one instance. And this is what is the limiting factor of AI models currently. limiting factor of AI models currently. limiting factor of AI models currently. The second problem with AI agents is The second problem with AI agents is The second problem with AI agents is that their output is non-deterministic. that their output is non-deterministic. that their output is non-deterministic. What that means is that their output is What that means is that their output is What that means is that their output is not predictable. not predictable. not predictable. Even if you give it a prompt to output Even if you give it a prompt to output Even if you give it a prompt to output things as a JSON object with a certain things as a JSON object with a certain things as a JSON object with a certain specification, specification, specification, in my experience, about in my experience, about in my experience, about si- 40% of the time, it will not adhere si- 40% of the time, it will not adhere si- 40% of the time, it will not adhere to those instructions. You will get to those instructions. You will get to those instructions. You will get output that maybe includes a few lines output that maybe includes a few lines output that maybe includes a few lines like, "Oh, I understand. Here is your like, "Oh, I understand. Here is your like, "Oh, I understand. Here is your adjacent object." For example, it will adjacent object." For example, it will adjacent object." For example, it will include those lines. But, if you if you include those lines. But, if you if you include those lines. But, if you if you start building applications based on start building applications based on start building applications based on that output, you can fail, right? And that output, you can fail, right? And that output, you can fail, right? And that is the biggest challenge that AI that is the biggest challenge that AI that is the biggest challenge that AI agents have at this point. That it's agents have at this point. That it's agents have at this point. That it's unpredictable. Now, I've been doing this unpredictable. Now, I've been doing this unpredictable. Now, I've been doing this for some years. I've been in this game for some years. I've been in this game for some years. I've been in this game for a while, and I can tell you that no for a while, and I can tell you that no for a while, and I can tell you that no company that runs any critical company that runs any critical company that runs any critical infrastructure is going to let any AI infrastructure is going to let any AI infrastructure is going to let any AI agent loose on their production agent loose on their production agent loose on their production environment, okay? There always needs to environment, okay? There always needs to environment, okay? There always needs to be a human in the loop there.

be a human in the loop there. be a human in the loop there. Now, it doesn't mean that AI is useless. Now, it doesn't mean that AI is useless. Now, it doesn't mean that AI is useless. Like, it fails, but it did it did Like, it fails, but it did it did Like, it fails, but it did it did succeed 30% of the time, okay? So, I do succeed 30% of the time, okay? So, I do succeed 30% of the time, okay? So, I do think that these AI agents and systems think that these AI agents and systems think that these AI agents and systems are going to have a a effect. So, now are going to have a a effect. So, now are going to have a a effect. So, now that we know that the context window is that we know that the context window is that we know that the context window is a big limiting factor, the article also a big limiting factor, the article also a big limiting factor, the article also gives a few more reasons why AI agents gives a few more reasons why AI agents gives a few more reasons why AI agents struggle. And one of them being that it struggle. And one of them being that it struggle. And one of them being that it requires polyglot back-end development requires polyglot back-end development requires polyglot back-end development skills. In other words, large skills. In other words, large skills. In other words, large distributed systems often are built from distributed systems often are built from distributed systems often are built from multiple components in multiple multiple components in multiple multiple components in multiple programming languages. AI models are programming languages. AI models are programming languages. AI models are really good at Python and TypeScript really good at Python and TypeScript really good at Python and TypeScript because those are the most common ones. because those are the most common ones. because those are the most common ones. But when you then start mixing in other But when you then start mixing in other But when you then start mixing in other languages here, it they really start to languages here, it they really start to languages here, it they really start to struggle. struggle. struggle. And another reason is that the there is And another reason is that the there is And another reason is that the there is less training data. So, although adding less training data. So, although adding less training data. So, although adding instrumentation is a standard instrumentation is a standard instrumentation is a standard engineering task, it is not common engineering task, it is not common engineering task, it is not common practice in open source.

practice in open source. practice in open source. The most popular applications where The most popular applications where The most popular applications where reliability matters the most are in reliability matters the most are in reliability matters the most are in private repositories of big tech private repositories of big tech private repositories of big tech companies such as Apple, Airbnb, or companies such as Apple, Airbnb, or companies such as Apple, Airbnb, or Netflix. In other words, Netflix. In other words, Netflix. In other words, these AI models, they're trained on these AI models, they're trained on these AI models, they're trained on Stack Overflow, GitHub, and publicly Stack Overflow, GitHub, and publicly Stack Overflow, GitHub, and publicly available code. And these big companies available code. And these big companies available code. And these big companies that will have advanced s- uh that will have advanced s- uh that will have advanced s- uh reliability engineering practices like reliability engineering practices like reliability engineering practices like this and implementations of open this and implementations of open this and implementations of open telemetry, telemetry, telemetry, they don't expose their code. they don't expose their code. they don't expose their code. So, the AI models don't have anything to So, the AI models don't have anything to So, the AI models don't have anything to base their work on, and that's why base their work on, and that's why base their work on, and that's why they're less efficient at this task. So, they're less efficient at this task. So, they're less efficient at this task. So, their conclusions are that the best their conclusions are that the best their conclusions are that the best models struggle. The state of the art models struggle. The state of the art models struggle. The state of the art Clau- Claude Opus only solved 29% of Clau- Claude Opus only solved 29% of Clau- Claude Opus only solved 29% of tasks. There are language gaps, and tasks. There are language gaps, and tasks. There are language gaps, and there is are silent failures, and cost there is are silent failures, and cost there is are silent failures, and cost efficiency. Okay? So, it's very efficiency. Okay? So, it's very efficiency. Okay? So, it's very expensive to let AI do these tasks, even expensive to let AI do these tasks, even expensive to let AI do these tasks, even if they fail all the time. if they fail all the time. if they fail all the time. So, as they say, AI SRE is still mostly So, as they say, AI SRE is still mostly So, as they say, AI SRE is still mostly hype, but there is hope, they say. So, hype, but there is hope, they say. So, hype, but there is hope, they say. So, they are going to There are they are going to There are they are going to There are constantly developments being made in constantly developments being made in constantly developments being made in this field. So, now to the question, this field. So, now to the question, this field. So, now to the question, will AI replace DevOps jobs? And I will AI replace DevOps jobs? And I will AI replace DevOps jobs? And I recently made a video about the RELF recently made a video about the RELF recently made a video about the RELF technique, for example. You can check it technique, for example. You can check it technique, for example. You can check it out over here.

out over here. out over here. And in that video, I explain how And in that video, I explain how And in that video, I explain how software engineering is basically um software engineering is basically um software engineering is basically um reduced to the cost of $10 an hour, and reduced to the cost of $10 an hour, and reduced to the cost of $10 an hour, and software development, so not engineering software development, so not engineering software development, so not engineering but development, the act of writing code but development, the act of writing code but development, the act of writing code is almost completely outsourced to AI is almost completely outsourced to AI is almost completely outsourced to AI models right now. models right now. models right now. That is a concerning trend for software That is a concerning trend for software That is a concerning trend for software engineers and they are lying awake at engineers and they are lying awake at engineers and they are lying awake at night right now. At least the ones who night right now. At least the ones who night right now. At least the ones who are not asleep at the wheel. For DevOps, are not asleep at the wheel. For DevOps, are not asleep at the wheel. For DevOps, I think the timeline is longer. The I think the timeline is longer. The I think the timeline is longer. The truth is nobody knows for sure and truth is nobody knows for sure and truth is nobody knows for sure and nobody can know for sure. I can give you nobody can know for sure. I can give you nobody can know for sure. I can give you my thoughts but I don't have a crystal my thoughts but I don't have a crystal my thoughts but I don't have a crystal ball that that knows the future, right? ball that that knows the future, right? ball that that knows the future, right? What I can give you are my thoughts and What I can give you are my thoughts and What I can give you are my thoughts and my expectations for this. So reading my expectations for this. So reading my expectations for this. So reading this article this article this article it gives me hope it gives me hope it gives me hope because as we see the frontier models because as we see the frontier models because as we see the frontier models are failing at entry-level SRE tasks at are failing at entry-level SRE tasks at are failing at entry-level SRE tasks at this point in time. Because of the this point in time. Because of the this point in time. Because of the context window limitations context window limitations context window limitations these AI models they cannot hold these these AI models they cannot hold these these AI models they cannot hold these large distributed systems in their their large distributed systems in their their large distributed systems in their their focus in their in their brain right now.

focus in their in their brain right now. focus in their in their brain right now. So that gives me hope. But then the So that gives me hope. But then the So that gives me hope. But then the question is basically okay, we're just question is basically okay, we're just question is basically okay, we're just waiting for these context windows to waiting for these context windows to waiting for these context windows to become bigger. And in fact become bigger. And in fact become bigger. And in fact Cloud Opus 4.6 came out recently Cloud Opus 4.6 came out recently Cloud Opus 4.6 came out recently and that context window was increased and that context window was increased and that context window was increased from 200,000 to 1 million tokens from 200,000 to 1 million tokens from 200,000 to 1 million tokens already. already. already. So there's already developments being So there's already developments being So there's already developments being made over there. So on a long enough made over there. So on a long enough made over there. So on a long enough time horizon, these models are going to time horizon, these models are going to time horizon, these models are going to be more efficient at these kinds of be more efficient at these kinds of be more efficient at these kinds of tasks. Now all of that being said, I am tasks. Now all of that being said, I am tasks. Now all of that being said, I am not worried about my job or my career as not worried about my job or my career as not worried about my job or my career as a DevOps engineer at all. a DevOps engineer at all. a DevOps engineer at all. I I have no no expectations that DevOps I I have no no expectations that DevOps I I have no no expectations that DevOps engineers are going to be replaced by AI engineers are going to be replaced by AI engineers are going to be replaced by AI especially in the sense of software especially in the sense of software especially in the sense of software development being completely outsourced development being completely outsourced development being completely outsourced to LLMs. to LLMs. to LLMs. They don't have enough context to They don't have enough context to They don't have enough context to understand these distributed systems. understand these distributed systems. understand these distributed systems. The thing is, writing code is one thing, The thing is, writing code is one thing, The thing is, writing code is one thing, but if you have infrastructure that runs but if you have infrastructure that runs but if you have infrastructure that runs important applications or important applications or important applications or banks or stock exchanges, banks or stock exchanges, banks or stock exchanges, you there's no room for error. There's you there's no room for error. There's you there's no room for error. There's no room for error on these large no room for error on these large no room for error on these large production systems.

production systems. production systems. And any system, any AI system that has And any system, any AI system that has And any system, any AI system that has even a minimum percentage of error even a minimum percentage of error even a minimum percentage of error is not going to be managing any of these is not going to be managing any of these is not going to be managing any of these production systems. So, production systems. So, production systems. So, yes, I think that these AI agents are yes, I think that these AI agents are yes, I think that these AI agents are going to have an effect on DevOps work going to have an effect on DevOps work going to have an effect on DevOps work and DevOps engineering. In fact, I'm and DevOps engineering. In fact, I'm and DevOps engineering. In fact, I'm just going to think it's going to have a just going to think it's going to have a just going to think it's going to have a positive effect because it's going to positive effect because it's going to positive effect because it's going to eliminate a lot of the toil in site eliminate a lot of the toil in site eliminate a lot of the toil in site reliability engineering and DevOps work. reliability engineering and DevOps work. reliability engineering and DevOps work. So, all of the boring stuff, it can So, all of the boring stuff, it can So, all of the boring stuff, it can handle that perfectly. But, that means handle that perfectly. But, that means handle that perfectly. But, that means that see, DevOps engineers are going to that see, DevOps engineers are going to that see, DevOps engineers are going to have more time to focus on the important have more time to focus on the important have more time to focus on the important stuff, to make things secure, to get stuff, to make things secure, to get stuff, to make things secure, to get that extra nine of uptime and that extra nine of uptime and that extra nine of uptime and reliability. But, I don't see AI reliability. But, I don't see AI reliability. But, I don't see AI replacing us in the near future. And replacing us in the near future. And replacing us in the near future. And finally, the best part about all of this finally, the best part about all of this finally, the best part about all of this is that AI is obviously increasing. is that AI is obviously increasing. is that AI is obviously increasing. There's $7 trillion There's $7 trillion There's $7 trillion being invested in data centers worldwide being invested in data centers worldwide being invested in data centers worldwide right now.

right now. right now. And the AI boom is only going to grow. And the AI boom is only going to grow. And the AI boom is only going to grow. But, all of that AI needs to run But, all of that AI needs to run But, all of that AI needs to run somewhere, right? And according to the somewhere, right? And according to the somewhere, right? And according to the CNCF, Kubernetes is the de facto AI CNCF, Kubernetes is the de facto AI CNCF, Kubernetes is the de facto AI platform right now. So, what that means platform right now. So, what that means platform right now. So, what that means is that if AI grows, Kubernetes also is that if AI grows, Kubernetes also is that if AI grows, Kubernetes also grows. So, the DevOps engineers who have grows. So, the DevOps engineers who have grows. So, the DevOps engineers who have been smart and who have been learning been smart and who have been learning been smart and who have been learning Kubernetes over the past few years, we Kubernetes over the past few years, we Kubernetes over the past few years, we are experiencing golden times right now. are experiencing golden times right now. are experiencing golden times right now. There's so much work available for us. There's so much work available for us. There's so much work available for us. There's like 130,000 There's like 130,000 There's like 130,000 jobs available for Kubernetes engineers jobs available for Kubernetes engineers jobs available for Kubernetes engineers right now. It's it's crazy. So, if AI right now. It's it's crazy. So, if AI right now. It's it's crazy. So, if AI grows, our work also grows. Uh if you've grows, our work also grows. Uh if you've grows, our work also grows. Uh if you've not learned Kubernetes yet, don't worry. not learned Kubernetes yet, don't worry. not learned Kubernetes yet, don't worry. I got you covered. There's a free 7-day I got you covered. There's a free 7-day I got you covered. There's a free 7-day course I created. Link's in the course I created. Link's in the course I created. Link's in the description. So, this is my current description. So, this is my current description. So, this is my current stance on this topic. I sleep really stance on this topic. I sleep really stance on this topic. I sleep really good at night knowing that there are good at night knowing that there are good at night knowing that there are articles out here that show that the articles out here that show that the articles out here that show that the models cannot do these types of tasks models cannot do these types of tasks models cannot do these types of tasks yet. I am very hopeful for the future of yet. I am very hopeful for the future of yet. I am very hopeful for the future of DevOps with the increase of AI and DevOps with the increase of AI and DevOps with the increase of AI and therefore the increased need of therefore the increased need of therefore the increased need of Kubernetes uh cluster management that I Kubernetes uh cluster management that I Kubernetes uh cluster management that I am very skilled at.

am very skilled at. am very skilled at. And in general, I think, especially in And in general, I think, especially in And in general, I think, especially in the DevOps infrastructure niche, there's the DevOps infrastructure niche, there's the DevOps infrastructure niche, there's always going to be work for people who always going to be work for people who always going to be work for people who have high skills, okay? So, have high skills, okay? So, have high skills, okay? So, if you are watching this and you have if you are watching this and you have if you are watching this and you have this you identify with my approach and this you identify with my approach and this you identify with my approach and the craftsman's mindset that I always the craftsman's mindset that I always the craftsman's mindset that I always talk about, talk about, talk about, like I I think craftsmen are not going like I I think craftsmen are not going like I I think craftsmen are not going to be out of a job anytime soon. to be out of a job anytime soon. to be out of a job anytime soon. Especially in these high-stakes Especially in these high-stakes Especially in these high-stakes production environments, production environments, production environments, they are not going to let AI near that. they are not going to let AI near that. they are not going to let AI near that. They need people who understand. So, if They need people who understand. So, if They need people who understand. So, if you have deep infrastructure skills, you you have deep infrastructure skills, you you have deep infrastructure skills, you don't need to worry. But, let me know in don't need to worry. But, let me know in don't need to worry. But, let me know in the comments. Like, maybe some of you the comments. Like, maybe some of you the comments. Like, maybe some of you have been using AI agents at work. I have been using AI agents at work. I have been using AI agents at work. I would love to know more about this. I am would love to know more about this. I am would love to know more about this. I am also learning as I go. So, let me know. also learning as I go. So, let me know. also learning as I go. So, let me know. Thank you so much for watching, and I'll Thank you so much for watching, and I'll Thank you so much for watching, and I'll see you in the next video.