I Coached 1,000+ DevOps Engineers, Stop Using AI

If you're in your first year of landing If you're in your first year of landing a DevOps or Kubernetes job, you should a DevOps or Kubernetes job, you should a DevOps or Kubernetes job, you should not be using AI. In this video, I'm not be using AI. In this video, I'm not be using AI. In this video, I'm going to explain exactly why I think going to explain exactly why I think going to explain exactly why I think this. My name is Mischa van den Berg. this. My name is Mischa van den Berg. this. My name is Mischa van den Berg. I'm a DevOps engineer and I've helped I'm a DevOps engineer and I've helped I'm a DevOps engineer and I've helped over 1,000 people land six-figure over 1,000 people land six-figure over 1,000 people land six-figure Kubernetes and DevOps jobs. This is what Kubernetes and DevOps jobs. This is what Kubernetes and DevOps jobs. This is what I do for a living and it's very I do for a living and it's very I do for a living and it's very important to me that I give you and my important to me that I give you and my important to me that I give you and my students the correct advice. So, I've students the correct advice. So, I've students the correct advice. So, I've spent several weeks thinking about this, spent several weeks thinking about this, spent several weeks thinking about this, tearing this argument apart, and really tearing this argument apart, and really tearing this argument apart, and really coming with the best conclusion to this, coming with the best conclusion to this, coming with the best conclusion to this, which is you should not be using AI in which is you should not be using AI in which is you should not be using AI in your first year. So, let's dive into it. your first year. So, let's dive into it. your first year. So, let's dive into it. My company and the place where I teach My company and the place where I teach My company and the place where I teach is called CubeCraft and I chose that is called CubeCraft and I chose that is called CubeCraft and I chose that name very deliberately. It is cube for name very deliberately. It is cube for name very deliberately. It is cube for Kubernetes and craft for the Kubernetes and craft for the Kubernetes and craft for the craftsmanship because to run Kubernetes craftsmanship because to run Kubernetes craftsmanship because to run Kubernetes clusters and to provide modern clusters and to provide modern clusters and to provide modern infrastructure to me is a craft. It's infrastructure to me is a craft. It's infrastructure to me is a craft. It's something that requires deep study and a something that requires deep study and a something that requires deep study and a high skill level in order to do it high skill level in order to do it high skill level in order to do it efficiently. And I think that this is efficiently. And I think that this is efficiently. And I think that this is not easily replicated by AI. Not with not easily replicated by AI. Not with not easily replicated by AI. Not with the current state, not with the current the current state, not with the current the current state, not with the current models that we see. But for software models that we see. But for software models that we see. But for software engineering, the story is very engineering, the story is very engineering, the story is very different. If you are on tech Twitter different. If you are on tech Twitter different. If you are on tech Twitter recently, you have seen the Claude code recently, you have seen the Claude code recently, you have seen the Claude code hype. You have seen Ralph. And if you hype. You have seen Ralph. And if you hype. You have seen Ralph. And if you don't know what Ralph is, you can check don't know what Ralph is, you can check don't know what Ralph is, you can check out this video. I made a video about out this video. I made a video about out this video. I made a video about explaining what what it is. But explaining what what it is. But explaining what what it is. But fundamentally, software development is fundamentally, software development is fundamentally, software development is now costs $10 now costs $10 now costs $10 an hour. Generating code is now cheaper an hour. Generating code is now cheaper an hour. Generating code is now cheaper than someone flipping burgers at than someone flipping burgers at than someone flipping burgers at McDonald's. Software development is now McDonald's. Software development is now McDonald's. Software development is now outsourced to AI agents, mostly.

outsourced to AI agents, mostly. outsourced to AI agents, mostly. Software engineering and steering the Software engineering and steering the Software engineering and steering the agents is a whole different story. But agents is a whole different story. But agents is a whole different story. But the thing is, software engineers now the thing is, software engineers now the thing is, software engineers now need to become AI native. They need to need to become AI native. They need to need to become AI native. They need to start working with AI in order to have a start working with AI in order to have a start working with AI in order to have a competitive advantage. And this is why I competitive advantage. And this is why I competitive advantage. And this is why I was thinking, well, shouldn't DevOps was thinking, well, shouldn't DevOps was thinking, well, shouldn't DevOps engineers also become AI native? But the engineers also become AI native? But the engineers also become AI native? But the answer is no. The answer that I arrived answer is no. The answer that I arrived answer is no. The answer that I arrived at is no. For the first year, you should at is no. For the first year, you should at is no. For the first year, you should learn how to wield the tools yourself learn how to wield the tools yourself learn how to wield the tools yourself properly. I recommend my students and properly. I recommend my students and properly. I recommend my students and I'm recommending you to learn Linux the I'm recommending you to learn Linux the I'm recommending you to learn Linux the hard way, by hand, using Vim, using only hard way, by hand, using Vim, using only hard way, by hand, using Vim, using only bash for at least 6 months to a year. bash for at least 6 months to a year. bash for at least 6 months to a year. And the reason I say this is because in And the reason I say this is because in And the reason I say this is because in order to build these skills properly and order to build these skills properly and order to build these skills properly and to get the sense of technical to get the sense of technical to get the sense of technical understanding and the ability to debug understanding and the ability to debug understanding and the ability to debug things without just using man pages, things without just using man pages, things without just using man pages, you're not going to you learn these you're not going to you learn these you're not going to you learn these skills. Every time that you run into a skills. Every time that you run into a skills. Every time that you run into a problem, you're going to be asking Chat problem, you're going to be asking Chat problem, you're going to be asking Chat GPT about it. And yes, there are ways of GPT about it. And yes, there are ways of GPT about it. And yes, there are ways of using Chat GPT and asking it to be your using Chat GPT and asking it to be your using Chat GPT and asking it to be your mentor and that kind of thing, but the mentor and that kind of thing, but the mentor and that kind of thing, but the reality is that average people do not reality is that average people do not reality is that average people do not have the discipline to actually use Chat have the discipline to actually use Chat have the discipline to actually use Chat GPT in such a way. Most people are going GPT in such a way. Most people are going GPT in such a way. Most people are going to ask it for an answer, blindly copying to ask it for an answer, blindly copying to ask it for an answer, blindly copying that answer, and then using that to that answer, and then using that to that answer, and then using that to solve their problem. And now you might solve their problem. And now you might solve their problem. And now you might be thinking, well, what's the issue? You be thinking, well, what's the issue? You be thinking, well, what's the issue? You have Chat GPT available to you all the have Chat GPT available to you all the have Chat GPT available to you all the time, right? You can just use AI as a time, right? You can just use AI as a time, right? You can just use AI as a DevOps engineer. That is true to some DevOps engineer. That is true to some DevOps engineer. That is true to some degree, but what you have to think about

degree, but what you have to think about degree, but what you have to think about here is reverse engineering your here is reverse engineering your here is reverse engineering your interview situation. And this is what interview situation. And this is what interview situation. And this is what most people don't think about. Most most people don't think about. Most most people don't think about. Most people won't think this far ahead. So people won't think this far ahead. So people won't think this far ahead. So right now, put yourself in that right now, put yourself in that right now, put yourself in that situation. You're sitting at your DevOps situation. You're sitting at your DevOps situation. You're sitting at your DevOps interview. You have the hiring manager interview. You have the hiring manager interview. You have the hiring manager across you and he's asking you about across you and he's asking you about across you and he's asking you about stories. He's asking you about things stories. He's asking you about things stories. He's asking you about things that you have implemented in production. that you have implemented in production. that you have implemented in production. For example, let's say that you have a a For example, let's say that you have a a For example, let's say that you have a a few projects on your CV and then he asks few projects on your CV and then he asks few projects on your CV and then he asks about the implementation of a technical about the implementation of a technical about the implementation of a technical detail. He asks you how that went. And detail. He asks you how that went. And detail. He asks you how that went. And then you're going to talk about a few then you're going to talk about a few then you're going to talk about a few problems that you had and then he asks problems that you had and then he asks problems that you had and then he asks you, "How did you solve them?" If your you, "How did you solve them?" If your you, "How did you solve them?" If your answer in that situation is, "I asked answer in that situation is, "I asked answer in that situation is, "I asked Chat GPT," then you are not going to be Chat GPT," then you are not going to be Chat GPT," then you are not going to be hired for that position. Asking Chat GPT hired for that position. Asking Chat GPT hired for that position. Asking Chat GPT makes you helpless in the case where makes you helpless in the case where makes you helpless in the case where Chat GPT does not have the answer. When Chat GPT does not have the answer. When Chat GPT does not have the answer. When I interview DevOps candidates, I look I interview DevOps candidates, I look I interview DevOps candidates, I look for people who know how to search the for people who know how to search the for people who know how to search the man pages for information. I'm looking man pages for information. I'm looking man pages for information. I'm looking for people who are fast on the command for people who are fast on the command for people who are fast on the command line and who can get information by, for line and who can get information by, for line and who can get information by, for example, kubectl get events, wrapping example, kubectl get events, wrapping example, kubectl get events, wrapping that or taking out a few columns and that or taking out a few columns and that or taking out a few columns and sorting the information in order of sorting the information in order of sorting the information in order of recency, that kind of thing. If only recency, that kind of thing. If only recency, that kind of thing. If only thing you do is just copying the logs, thing you do is just copying the logs, thing you do is just copying the logs, pasting it into chat GPT and waiting for pasting it into chat GPT and waiting for pasting it into chat GPT and waiting for chat GPT to answer it, you are at the chat GPT to answer it, you are at the chat GPT to answer it, you are at the mercy of chat GPT. And that person I do mercy of chat GPT. And that person I do mercy of chat GPT. And that person I do not want to let loose on my not want to let loose on my not want to let loose on my infrastructure. I do not have the infrastructure. I do not have the infrastructure. I do not have the confidence that this person actually confidence that this person actually confidence that this person actually understands what he's doing and that he

understands what he's doing and that he understands what he's doing and that he truly understands the commands that he truly understands the commands that he truly understands the commands that he types into the command line. So, that is types into the command line. So, that is types into the command line. So, that is what you how you have to think about what you how you have to think about what you how you have to think about this. You have to reverse engineer the this. You have to reverse engineer the this. You have to reverse engineer the interview situation. And if for every interview situation. And if for every interview situation. And if for every question, your answer is I asked chat question, your answer is I asked chat question, your answer is I asked chat GPT and it gave me the answer, you are GPT and it gave me the answer, you are GPT and it gave me the answer, you are on the wrong track. Now, this is very on the wrong track. Now, this is very on the wrong track. Now, this is very different for someone who is going for a different for someone who is going for a different for someone who is going for a software engineering interview where software engineering interview where software engineering interview where modern software engineering expects you modern software engineering expects you modern software engineering expects you to be AI native and to have AI skills. to be AI native and to have AI skills. to be AI native and to have AI skills. Those people would benefit from saying I Those people would benefit from saying I Those people would benefit from saying I use cloud code to generate 10x the use cloud code to generate 10x the use cloud code to generate 10x the amount of code in 1 hour, that kind of amount of code in 1 hour, that kind of amount of code in 1 hour, that kind of thing. There it's an asset to have these thing. There it's an asset to have these thing. There it's an asset to have these AI tools down in as a junior. And for AI tools down in as a junior. And for AI tools down in as a junior. And for infrastructure, the bar is higher and infrastructure, the bar is higher and infrastructure, the bar is higher and the responsibility is a lot higher. the responsibility is a lot higher. the responsibility is a lot higher. Someone who enters commands on the Someone who enters commands on the Someone who enters commands on the command line needs to completely command line needs to completely command line needs to completely understand what he is doing at all times understand what he is doing at all times understand what he is doing at all times because if he doesn't, he can do a lot because if he doesn't, he can do a lot because if he doesn't, he can do a lot of damage. And I when I'm hiring a of damage. And I when I'm hiring a of damage. And I when I'm hiring a DevOps engineer, I need to trust this DevOps engineer, I need to trust this DevOps engineer, I need to trust this person that he knows exactly what he's person that he knows exactly what he's person that he knows exactly what he's doing with every command that he types doing with every command that he types doing with every command that he types into the command line. And then the into the command line. And then the into the command line. And then the second point is that AI works as an second point is that AI works as an second point is that AI works as an amplifier, okay? To prompt AI amplifier, okay? To prompt AI amplifier, okay? To prompt AI efficiently, you need to have a certain efficiently, you need to have a certain efficiently, you need to have a certain baseline level of understanding of the baseline level of understanding of the baseline level of understanding of the problem that you're trying to solve with problem that you're trying to solve with problem that you're trying to solve with AI. And I think that like for example, AI. And I think that like for example, AI. And I think that like for example, AI agents, they are still not able to AI agents, they are still not able to AI agents, they are still not able to work properly and and autonomously on work properly and and autonomously on work properly and and autonomously on infrastructure. Like you cannot let infrastructure. Like you cannot let infrastructure. Like you cannot let loose an AI agent on Kubernetes clusters loose an AI agent on Kubernetes clusters loose an AI agent on Kubernetes clusters and trust that it will do the right

and trust that it will do the right and trust that it will do the right thing. Like AI agents, they are limited thing. Like AI agents, they are limited thing. Like AI agents, they are limited to their context window. And if you to their context window. And if you to their context window. And if you don't know what that is, I strongly don't know what that is, I strongly don't know what that is, I strongly recommend you do some more research into recommend you do some more research into recommend you do some more research into this. But an AI agent can only hold so this. But an AI agent can only hold so this. But an AI agent can only hold so much in its current context. And much in its current context. And much in its current context. And managing a Kubernetes context a cluster managing a Kubernetes context a cluster managing a Kubernetes context a cluster is already too much context. So imagine is already too much context. So imagine is already too much context. So imagine not just one cluster, but an entire not just one cluster, but an entire not just one cluster, but an entire system, a distributed system of multiple system, a distributed system of multiple system, a distributed system of multiple clusters with multiple layers of clusters with multiple layers of clusters with multiple layers of infrastructure. The AI agents are not infrastructure. The AI agents are not infrastructure. The AI agents are not able to hold all of that context in able to hold all of that context in able to hold all of that context in their brains yet. their brains yet. their brains yet. Not yet. And that's why I think DevOps Not yet. And that's why I think DevOps Not yet. And that's why I think DevOps is still relatively safe from AI is still relatively safe from AI is still relatively safe from AI replacement. So a junior DevOps replacement. So a junior DevOps replacement. So a junior DevOps engineer, a junior infrastructure engineer, a junior infrastructure engineer, a junior infrastructure infrastructure engineer already needs a infrastructure engineer already needs a infrastructure engineer already needs a fairly high level of systems thinking, fairly high level of systems thinking, fairly high level of systems thinking, okay? And you're only going to build okay? And you're only going to build okay? And you're only going to build these skills by doing the work yourself these skills by doing the work yourself these skills by doing the work yourself and going the long route, having these and going the long route, having these and going the long route, having these long debugging sessions, fighting with long debugging sessions, fighting with long debugging sessions, fighting with YAML indentation, working on the file YAML indentation, working on the file YAML indentation, working on the file system yourself without using cloud system yourself without using cloud system yourself without using cloud code. That is how you develop these code. That is how you develop these code. That is how you develop these debugging skills and the debugging debugging skills and the debugging debugging skills and the debugging mindset. Going back to the point of AI mindset. Going back to the point of AI mindset. Going back to the point of AI being an amplifier, if you have zero being an amplifier, if you have zero being an amplifier, if you have zero skills, you cannot amplify them. So your skills, you cannot amplify them. So your skills, you cannot amplify them. So your baseline level of skill needs to be baseline level of skill needs to be baseline level of skill needs to be fairly high. And when you have a high fairly high. And when you have a high fairly high. And when you have a high level of skill, that's when AI agents level of skill, that's when AI agents level of skill, that's when AI agents become very useful. Like I'm a senior become very useful. Like I'm a senior become very useful. Like I'm a senior DevOps engineer and I am now able to use DevOps engineer and I am now able to use DevOps engineer and I am now able to use AI agents to automate a lot of my work,

AI agents to automate a lot of my work, AI agents to automate a lot of my work, but they still need a ton of coaxing. but they still need a ton of coaxing. but they still need a ton of coaxing. They need a lot of direction and I can They need a lot of direction and I can They need a lot of direction and I can only give them very short, only give them very short, only give them very short, well-specified tasks and I still have to well-specified tasks and I still have to well-specified tasks and I still have to kind of watch what they're doing because kind of watch what they're doing because kind of watch what they're doing because very often they will just lose track and very often they will just lose track and very often they will just lose track and start hallucinating and frankly do start hallucinating and frankly do start hallucinating and frankly do damaging things in production. Like I've damaging things in production. Like I've damaging things in production. Like I've had to fix a couple of issues that my AI had to fix a couple of issues that my AI had to fix a couple of issues that my AI agents have done on my home lab already, agents have done on my home lab already, agents have done on my home lab already, for example. for example. for example. So, So, So, yeah. yeah. yeah. You It's not possible to just let them You It's not possible to just let them You It's not possible to just let them loose and they need a very, very loose and they need a very, very loose and they need a very, very specific way of prompting and a high specific way of prompting and a high specific way of prompting and a high skill level for prompting them. And I'm skill level for prompting them. And I'm skill level for prompting them. And I'm not the only one who's saying this. I not the only one who's saying this. I not the only one who's saying this. I wrote a tweet recently about this where wrote a tweet recently about this where wrote a tweet recently about this where I basically explain what I just said and I basically explain what I just said and I basically explain what I just said and thinking about junior DevOps roles and thinking about junior DevOps roles and thinking about junior DevOps roles and AI. And the creator of Ralph, again AI. And the creator of Ralph, again AI. And the creator of Ralph, again there's a video up here if you don't there's a video up here if you don't there's a video up here if you don't know what Ralph is, but the creator of know what Ralph is, but the creator of know what Ralph is, but the creator of Ralph, he is also very strong in Ralph, he is also very strong in Ralph, he is also very strong in infrastructure and he actually replied infrastructure and he actually replied infrastructure and he actually replied to my tweet saying here that there's to my tweet saying here that there's to my tweet saying here that there's systems level thinking with systems level thinking with systems level thinking with infrastructure. I've been doing it since infrastructure. I've been doing it since infrastructure. I've been doing it since Sun OS, so it's all natural and AI makes Sun OS, so it's all natural and AI makes Sun OS, so it's all natural and AI makes me an absolute weapon with it. Look what me an absolute weapon with it. Look what me an absolute weapon with it. Look what he says here. I'm not sure about he says here. I'm not sure about he says here. I'm not sure about juniors, however. There's so many juniors, however. There's so many juniors, however. There's so many concepts that need to be known about concepts that need to be known about concepts that need to be known about before you can even prompt for it. In before you can even prompt for it. In before you can even prompt for it. In other words, you need to know a high other words, you need to know a high other words, you need to know a high amount of concepts. You need to have a amount of concepts. You need to have a amount of concepts. You need to have a high level of skill in order to high level of skill in order to high level of skill in order to efficiently prompt these models and AI

efficiently prompt these models and AI efficiently prompt these models and AI agents in infrastructure situations. So, agents in infrastructure situations. So, agents in infrastructure situations. So, he says, "I don't know regarding he says, "I don't know regarding he says, "I don't know regarding juniors. I definitely would not hire a juniors. I definitely would not hire a juniors. I definitely would not hire a junior who doesn't know the AI junior who doesn't know the AI junior who doesn't know the AI inferencing loop and hasn't built an inferencing loop and hasn't built an inferencing loop and hasn't built an agent yet." So, what he's saying here is agent yet." So, what he's saying here is agent yet." So, what he's saying here is this is for software engineers, but here this is for software engineers, but here this is for software engineers, but here he says, "Yeah, perhaps I would take a he says, "Yeah, perhaps I would take a he says, "Yeah, perhaps I would take a very high heavy approach of mentoring very high heavy approach of mentoring very high heavy approach of mentoring concepts and systems thinking, concepts and systems thinking, concepts and systems thinking, developing the human." And I that's very developing the human." And I that's very developing the human." And I that's very good advice. So, DevOps engineering is good advice. So, DevOps engineering is good advice. So, DevOps engineering is not just about infrastructure and coding not just about infrastructure and coding not just about infrastructure and coding and that kind of thing. DevOps is also and that kind of thing. DevOps is also and that kind of thing. DevOps is also about understanding business needs and about understanding business needs and about understanding business needs and communicating with dev teams and mapping communicating with dev teams and mapping communicating with dev teams and mapping out requirements and that kind of thing. out requirements and that kind of thing. out requirements and that kind of thing. So, there is a strong human side as So, there is a strong human side as So, there is a strong human side as well, which simply cannot be replaced by well, which simply cannot be replaced by well, which simply cannot be replaced by AI because by its definition, it's they AI because by its definition, it's they AI because by its definition, it's they they are people skills, they're human they are people skills, they're human they are people skills, they're human skills. What he's saying here about um skills. What he's saying here about um skills. What he's saying here about um not knowing the AI inferencing loop and not knowing the AI inferencing loop and not knowing the AI inferencing loop and hasn't built an agent yet, I do agree, hasn't built an agent yet, I do agree, hasn't built an agent yet, I do agree, but I would say do this after your first but I would say do this after your first but I would say do this after your first year. Like, again, I'm not saying you year. Like, again, I'm not saying you year. Like, again, I'm not saying you should never use AI as a DevOps should never use AI as a DevOps should never use AI as a DevOps engineer. I'm just saying you should engineer. I'm just saying you should engineer. I'm just saying you should have a strong foundation before you have a strong foundation before you have a strong foundation before you start throwing AI in the mix here. You start throwing AI in the mix here. You start throwing AI in the mix here. You need to have a relatively high baseline need to have a relatively high baseline need to have a relatively high baseline skill level in order to be able to skill level in order to be able to skill level in order to be able to prompt these AI models efficiently. So, prompt these AI models efficiently. So, prompt these AI models efficiently. So, that is my current stance on this. In that is my current stance on this. In that is my current stance on this. In your first year, when you're learning your first year, when you're learning your first year, when you're learning Kubernetes, when you're learning Linux, Kubernetes, when you're learning Linux, Kubernetes, when you're learning Linux, I would steer clear of AI. I would not I would steer clear of AI. I would not I would steer clear of AI. I would not use ChatGPT. In my first 2 years as a use ChatGPT. In my first 2 years as a use ChatGPT. In my first 2 years as a DevOps engineer, ChatGPT didn't exist DevOps engineer, ChatGPT didn't exist DevOps engineer, ChatGPT didn't exist yet, and I am very grateful for that. I

yet, and I am very grateful for that. I yet, and I am very grateful for that. I had mentors who forced me to use Vim, had mentors who forced me to use Vim, had mentors who forced me to use Vim, who forced me to stay on Linux for 2 who forced me to stay on Linux for 2 who forced me to stay on Linux for 2 years before moving on to anything else, years before moving on to anything else, years before moving on to anything else, who discouraged me from using Neovim, who discouraged me from using Neovim, who discouraged me from using Neovim, and I'm very, very glad that I had these and I'm very, very glad that I had these and I'm very, very glad that I had these mentors and that I learned it yeah, the mentors and that I learned it yeah, the mentors and that I learned it yeah, the hard way. I took the long route, but now hard way. I took the long route, but now hard way. I took the long route, but now I have a skill set which is very safe I have a skill set which is very safe I have a skill set which is very safe for the next decade or so, I think. I for the next decade or so, I think. I for the next decade or so, I think. I don't think I am in danger of being don't think I am in danger of being don't think I am in danger of being replaced by AI just yet. There are many replaced by AI just yet. There are many replaced by AI just yet. There are many software engineers who are lying awake software engineers who are lying awake software engineers who are lying awake right now because their job is in right now because their job is in right now because their job is in danger, basically. So, let me know what danger, basically. So, let me know what danger, basically. So, let me know what you think. If you're a senior watching you think. If you're a senior watching you think. If you're a senior watching this and you think this guy is full of this and you think this guy is full of this and you think this guy is full of please let me know because the please let me know because the please let me know because the most important thing to me is that I most important thing to me is that I most important thing to me is that I give the right advice to people. So, give the right advice to people. So, give the right advice to people. So, let's enter the discussion. If you're let's enter the discussion. If you're let's enter the discussion. If you're interested in learning infrastructure interested in learning infrastructure interested in learning infrastructure skills directly from me, there will be a skills directly from me, there will be a skills directly from me, there will be a link down below where you can get more link down below where you can get more link down below where you can get more information about that. It's what I do information about that. It's what I do information about that. It's what I do for a living and people that I coach for a living and people that I coach for a living and people that I coach land DevOps jobs every single week. So, land DevOps jobs every single week. So, land DevOps jobs every single week. So, I'm pretty good at it. So, check it out I'm pretty good at it. So, check it out I'm pretty good at it. So, check it out if you want to. And for now, thank you if you want to. And for now, thank you if you want to. And for now, thank you so much for watching and I'll see you in so much for watching and I'll see you in so much for watching and I'll see you in the next video.