Please, and I say this as a machine learning/MLOps and backend engineer, don't delve into AI-related content. We're already getting bombarded by that everywhere else. I very much agree with your take in this video about the recent hype and possible future of things. So please keep your content on python/software-design concepts. These are the true gems. Thanks again !
I would say keep doing what you do. Focus more on creating high quality content on python designs, best practices, principles and so on. There are many sources to know about AI and its too much. I would say its best to be in a place which is less crowded and unique rather than in a place which is jam packed. You could do occasionally some videos like this about AI. The content was amazing for this video. The idea of being expected for more is interesting for me. I am a junior developer myself and I have learned a lot about code design principles, patterns and many more from you. You have actually influenced the way I code. Thank you.
I agree with you. As developer, from the old school...more than 25 years developing... I just use IA for the advantages of the language to communicate, I mean...I like to ask something basic, "but please, explain me WHY some thing happens, or it's needed"...and IA is for me a great tool for that. Not to make the whole work for me. Also chatgpt it's great for simply chat about things, it's an interactive way to google some info, and perform great. Greetings and thanks for the video.
I read an article about Gemini and refactoring. It was about updating a library in a large project that offered a new interface. I use DeepSeek at a layman's level to migrate existing code from one language to another (in my case C# to Python). Of course, the resulting code (in both cases) must be reviewed by someone - someone who knows the code. But the many hard-working hands that were necessary yesterday or the impossibility that was there yesterday due to the costs - that's no longer true today. All of this can be solved with significantly less manpower. And that means that orders for junior developers are also decreasing. On the other hand, projects can now be started (as I said) that were previously simply not financially viable. So, the market for junior developers is also growing.
Title says "Won't Matter" but what you describe seems the opposite. Companies "upping their expectations" can be a dangerous thing, leading to assured burnout (and/or worse: selection bias towards pseudo-engineers who are better at "selling" their crappy solutions written mostly by AI, which they don't really understand deeply). To me that was always the scariest thing of the whole AI hype. Also you seem to talk about software design / architecture as separate from coding, but IMHO it's not the best way to think about it. Those two things can't go one without another. I'm pretty sure you know this, but I'm mentioning it because it makes your take kind of ambiguous, it all depends on the nuance of where you draw the line. But I still appreciate you sharing your thoughts, thanks! :)
So further fulfilling the self-made prophecy of sales folks only hiring other sales folk as software engineers only to deny the reality that some of the best software engineers are not necessarily the most social or sales-y
Code-generating LLMs will change Software Engineering but obviously in a bad way since people will not spend enough time trying to understand a code. I already see this unfortunately. We are going to see more over complicated software. Previously it was copy-paste from stackoverflow, now it will be copy-paste from the generated code.
@@deathrives4252 I am not blaming tools i am just describing what I am already seeing. Stackoverflow at least does not provide you with so much code, you at least have to do something by yourself
Yes, agree that you have to understand what has been producdd or generated and be in the loop controlling the final outcome.. In some way. Experienced is important in understanding how to put code together and those principals and practices continue in the code automation workflow.
Great take! It is definitely different for people who are just getting started in development. SW tend to be pessimistic about AI but I think it is already disrupting the industry on many levels, so better to adapt and learn thanks for the advice Arjan. Also, as a request, it will be great if you start making video series about advanced Python concept (like very advanced). Many follow you since years, and approaching seniority already. Thanks!
I'm a self-taught python learner. Before I got into coding i didn't know that there were entire teams working on various portions of a project. I assumed that all engineers were "full-stack" developers. Now that I've been using Copilot to help develop my own personal projects, I have a feeling that companies will attempt to use AI to reduce these teams and lean more on individual developers to do more coding with the help of AI.
Short term: it won’t replace existing developers Long-term: once every software developer alive today ages out of the workforce, there will be very few people with coding skills because people will have become over-reliant on the easy “give me an answer” fix that the AI Tech Broligarchs are pushing
Won't matter is not true at all. If you are a Software Engineer AI will increase productivity, not just for the company you work for but if you have ambition to work on own projects/business then take advantage of the tools available. Software engineering is the thought process that goes into the code you write not writing the actual code.
One thing people forget as well is we’re ultimately dealing with tools. In the past 3 hours I’ve worked with three separate tools intended to help with AI coding, all of which had bugs I’ve just finished troubleshooting and working around. This stuff will take time to get to the corporate utopia where all that’s left is the robots and the board of directors. It’s going to be a while.
I must first have the distinct concepts expressed using clear language in order to get out a high quality product from a model. That is where people need to focus their education rather than being concerned about what concrete details will become automated. These models are the greatest tutors ever but people do sometimes need to be shown where to begin with them and how to hold a critical lens to their outputs.
For me personally the scariest part of that whole AI mess is that even now there is a shortage of Senior level developers all around the world. And where do they come from? From junior developers! And when those junior developers become unresistant or even obsolete because of AI, where will we get a single person who will understand how this all works under the hood?
thanks for your view as an experienced developer. I appreciate your channel because it is focused on coding, not how to get rid of coding by means of almighty AI, please stay the way you are ❤
I think that "using AI as a developer" would be a great vid topic (or course, etc.) ... I believe that would be valuable. (i.e. effective prompts for developing x or y, specific best practice, pitfalls, lessons leared) The "generalized" AI content on UA-cam is a bit overwhelming at the moment.
5:18 Arjan, but the reasonable thing to do, is to enjoy the increased productivity and to spend more time with your kids, wife, and family. We should actually work less and enjoy the increased productivity. Instead, we are being forced to work the same amount, implement more features in less time for the same or even less money. First, that is an extremely irrational way to go about life and second, if it’s not us, who is benefiting from the increased efficiency?
I totally agree. Unfortunately, that’s not going to happen automatically, because the ones who are benefitting are company shareholders and they’ll push for more output for the same amount of money…
Technology is here to make us work more and more. Everyone dreams about robots working for us, but that will never happen. Welcome to the industrial revolution.
Since Christmas I thought I would dive in a get an AI Assistant to genrate some code for me in Rust. On the one hand I am amazed at how well it can do sometimes. On the other hand, even when it does well the generated code can have errors that that require manual fixing. Which requires that I know Rust pretty well. Often I found trying to get the AI to fix compilation errors in the cadet produced just makes it worse and worse until I give up. In all cases I would want somebody who understands the problem and the language to review the generated code and create meaningful tests of it. I conclude that those junior devs you were talking about would be wise to become as proficient in their language a programming skills as ever. Even if getting AI to generate code is saving time.
Google AIStudio answers this question correctly, DeepSeek R1 doesn't (its code doesn't compile): In Rust, I have a struct like this: struct MyStruct(v: Vec) How can I implement the trait IntoIterator for &MyStruct?
LLMs are a slightly better google/snack-overflow or a slightly weirder pip. I don't think they are having a meaningful effect on our productivity at all.
My major worry is what managers think. I have seen a lot of managers in the last few months who thought that AI can do everything perfectly. This is really bad for the industry.
I havent listened yet but o3 just came out and i just payed for the $200 sub again its that good (as an engineer). when the r2 comes out it most certainly make a difference. ive only been using o3 a couple hours damn! this thing is WAAAAY better at coding than o1 or claude (even me lol)
I'm interested in exploring the emerging landscape of AI coding assistants, particularly tools like Cursor and Cline. As these systems become increasingly integral to software development, I'd value your perspective on their capabilities, optimal use cases, and potential impact on engineering workflows. How do you see these tools reshaping our development practices?
Please don't talk too much about AI. As other comments said, there is already a ton of that everywhere, and your channel has always been very unique with sw design, architecture,bbest practices etc... so to me this is where you stand out!
My thoughts exactly. I have been playing around with ollama today and you can use several pretty good llm's with it for free. So I certainly see AI becoming a free tool for everyone and the race to the bottom is happening. In the foreseeable future AI, or LLM's in particular, are not going to analyse or write big software projects. So I agree with your view and am not particularly interested in a lot of AI content. Getting better at writing good software is where it has been at and where it will stay.
I'm thinking that these AI input will be more or less saturated. The information AI models use its data from internet, like StackOverFlow(SOF). But since people rely on copilot or chatgpt no one will ask questions to SOF hence the models will not get any more inputs therefore they will not learn.
But also as more code is llm generated will it just end up learning from itself? I wonder how these models behave when they are trained on their own output, it feels as if it would get worse since the error in their predictions accumulate, but I have no idea
It's a tool, not more, not less.. I have installed it locally and it helps with math and short functions, that's great for me but I never ever want to loose the ability to do real problem solving or just be a copy-paste 'developer'!
I am expecting that, at first, many (most?) tech companies will stop hiring (which they've mostly done) and maybe start laying off (which they're somewhat doing) because of AIs replacing labor. But, since tech companies usually follow the pendulum of hiring, firing, hiring, etc, they will up their expectations of what their employees will produce, what quantity, and what quality. And eventually, they will want to hire more labor, again, to use the AI tools for yet more and more quantity (their first priority) and quality (their second priority, or maybe even further down the list). So I'm expecting we will have a few years of low hiring of new, junior developers. But soon, companies will want them again.
Really don't need AI for development. I would've thought it might be better for testing and error checking though. Problem is c-suite lot think it's the next big money saver for reducing staff and making themselves look more efficient. What they miss is that it can replace them just as easily on the surface. So the thought should be assisting not replacing. I'm happy that you haven't delved into it for actual development. Keep it real please ❤
It is a race to the bottom but the infrastructure costs ,esp with regards to power is going to be a huge cost. These two things are conflicting. Right now AI is a loss leader across the board. Even the paid applications are not really making money so at some point these companies will have to find out how to turn a profit. I am not sure anyone will be in the space to self-host inference let alone training. We have companies that are reactiving Nuclear plants from the 70s like Three Mile to just pay for it. This is expensive and someone will eventually have to pay for this and so the "race to the bottom" is going to have some huge implications that don't include the "IP" or propetartyness of the model. Now can we get to a generalized model that doesn't need training that is good enough to spend less on training and get better hardware to generalize inference? Maybe? But the costs still have to be passed on. But this is a side point and not really the point of your video... I agree with everything else you said.
this is how I feel about job interview take home tests. Stop telling my not to use AI.. Just make the tests alot more complex and let me show what I can do in a real world situation (using AI)
i don't mean to sound mean, but as i heard at the beginning of the vid AI will be a feature integrated on Os. well, then apps will become irrelevant, maybe system and network architecture not. but apps/software will just be created and destroyed on the fly. as needed, as a load balancer. ❤
No, DeepSeek is changing your life as a developer: * You won't have to go to and work in BigTec to be in the AI game or to professionalize. THIS IS HUGE. * Now it is not about AI (which LLM/ecosystem/'algorithm'/...). It is about APPLICATION of AI. This is where the money is at. This is a HUGE shift. Much, much, much more people are part of the game now. And big tech will have a hard, hard, hard time to centralize it.
DeepSeek matter Ultimately, AI will replace the software that humans can use. It will establish interfaces with hardware, networks, objects, and other humans to accomplish tasks. Now, if you use AI to process large datasets consisting of human-created content-such as written statements, images, videos, and other forms of big data-it requires significant funding and is quite costly. DeepSeek & Co. make ideas real for researchers without funds. Programmers as researchers is one of possible ways to be not replaced by AI
Think about tasks to be done, not about programming. AlphaGo can play better than humans, and large language models (LLMs) are anymore just static weights-they are dynamic tools that can learn at inference time.
Except for government contracts where they get paid extra for the body count, the "code pig" jobs will definitely be affected. For people that do actual software design, I think AI is just another tool.
DeepSeek models are open source. How on earth powerful open-source models are not going to affect developers? I'm building these models into my systems right now, this very moment.
AI makes diagnonys much faster and better than human, but doctors did not disappear. Now doctors may use AI to be more productive and correct in diagnostic. This is partially true for software industry. Why partially, because moving down the road AI will tend to become AGI which means that it will now automate human thinking processes. And engineering solutions which now we provide as an engineers will complete by AGI. From the other hands, history is a permanent development of industries and human always meet the situations where particular skills become obsolete. ANd so far we always were able to shift towards changing reality. This time it might be faster, but new generation will learn to manage it I am sure. Also we will find solution for becoming obsolete as an engineers (or any other industry workers). No one needs revolutionary change in job market.
@sam witteveen, check his channel out he has a lot of good content on AI frameworks and AI. Also Langchain has a youtube channel they also produce good content.
It's kinda funny. Senior engineers think junior positions will be less needed. Junior engineers think they can basically do the work of senior engineers now with the help of AI, equaling the playing field.
I'm kind of a slow pessimist, in that I think that AI will obviously write software. The LLMs are already getting pretty good at a lot of the mundane stuff, and they are able to do some basic thinking if you pay enough to the AI companies. I believe getting them to think better than (very clever) humans may take longer than they two/three years timeline being discussed. But it will happen, when the AI is smarter than the developer and not just quicker and immune to tiredness, then why will companies need software engineers? There are some interesting questions about how software engineering develops at that point, as we'll be basically dependant on the AIs, if very few humans are thinking about AI. But there are whole hosts of things we don't do in software engineering enough, because it is expensive, and it is expensive because you have to pay smart people to do it slowly. From formal verification, to extensive testing. Why doesn't your code have 100% coverage, it isn't because it is objectively difficult, it is because the rewards are diminishing and the fiddly bits, like creating test harnesses for the bits you can't test readily such as for fault conditions that are hard to trigger, take increasing amounts of time and effort to write manually.
It would be good to see you build a non trivial app using as much AI in all phases of the SDLC as possible, perhaps in a series broken into the phase from idea to prod ready deployment, that is useful. Also the same for an existing code base. Incorporate the advice you have imparted in this video.
do you think a computer can play chess? (1975: a master chess player, smiling at you with compassion) answer: it can't! it BRUT_FORCES its way through millions of possibilities, which don't work and finds the one that does! do you think, an AI can code? (2025_ software engineer smiles at you with compassion) ... HANG IN THERE, MY FRIEND!😥
I hope, that Arjan will respond me. What the difference between software developer and software engineer? I asked ChatGpt about it and got some answers. But it's really interesting to know, what's your thoughts about this? 😊
😂😂😂 another anti AI struggler sentiment It is too simple to use, focus on logic, convert it to pseudo code, focus on a small surface area of the system, feed it to multiple LLMs, use your knowledge and experience to select/combine the best As an experienced developer I don't need it to be perfect, I just need it to be correct 80% of the time, the rest... it is up to me
@@ArjanCodes nobody try to explain how to leverage this wonderful tool, I hope you do 1. Focus on a small portion then write a detailed pseudo code (90s term for today's prompt engineering) 2. Use multiple LLMs, many are free and each trained differently, giving different results - this is potent because it provides you different perspective to the same problem 3. Select or combine results 4. Test for errors, adjust as needed Knowledge and experience is still a must, let us face the fact that you will never become a medical doctor using AI 🙂
@@commonsensedev I agree, too. Most of the time it generate examples or code that doesn't work. and believe me, i don't wanna fight with the prompt when I can write that code on my own. It has its use-cases, but copy/pasting is not.
@@vitalyl1327 if you have deep experience in deploying reliable, maintainable, upgradable enterprise grade system and used the current AI tools, I will definitely say NO, not today If you will say, it can reduce the number of developers by 20 to 30 percent in a typical project without sacrificing quality, today, I can say YES to that All the tools I have seen still requires a developer to effectively deliver a robust system But what I see there soo many jobs today out there that will be under threat from AI agents, jobs that can be addressed by AI and traditional automation, many offshore virtual assistants have already lost their job Coding might be totally automated in the future, but due to inherent complexity it is one of the jobs that will be last to go
Please, and I say this as a machine learning/MLOps and backend engineer, don't delve into AI-related content. We're already getting bombarded by that everywhere else. I very much agree with your take in this video about the recent hype and possible future of things. So please keep your content on python/software-design concepts. These are the true gems. Thanks again !
Agreed. AI fatigue is real
I would say keep doing what you do. Focus more on creating high quality content on python designs, best practices, principles and so on. There are many sources to know about AI and its too much. I would say its best to be in a place which is less crowded and unique rather than in a place which is jam packed. You could do occasionally some videos like this about AI. The content was amazing for this video. The idea of being expected for more is interesting for me. I am a junior developer myself and I have learned a lot about code design principles, patterns and many more from you. You have actually influenced the way I code. Thank you.
I agree with you. As developer, from the old school...more than 25 years developing... I just use IA for the advantages of the language to communicate, I mean...I like to ask something basic, "but please, explain me WHY some thing happens, or it's needed"...and IA is for me a great tool for that. Not to make the whole work for me. Also chatgpt it's great for simply chat about things, it's an interactive way to google some info, and perform great.
Greetings and thanks for the video.
I read an article about Gemini and refactoring. It was about updating a library in a large project that offered a new interface. I use DeepSeek at a layman's level to migrate existing code from one language to another (in my case C# to Python). Of course, the resulting code (in both cases) must be reviewed by someone - someone who knows the code. But the many hard-working hands that were necessary yesterday or the impossibility that was there yesterday due to the costs - that's no longer true today. All of this can be solved with significantly less manpower. And that means that orders for junior developers are also decreasing. On the other hand, projects can now be started (as I said) that were previously simply not financially viable. So, the market for junior developers is also growing.
Title says "Won't Matter" but what you describe seems the opposite. Companies "upping their expectations" can be a dangerous thing, leading to assured burnout (and/or worse: selection bias towards pseudo-engineers who are better at "selling" their crappy solutions written mostly by AI, which they don't really understand deeply). To me that was always the scariest thing of the whole AI hype.
Also you seem to talk about software design / architecture as separate from coding, but IMHO it's not the best way to think about it. Those two things can't go one without another. I'm pretty sure you know this, but I'm mentioning it because it makes your take kind of ambiguous, it all depends on the nuance of where you draw the line.
But I still appreciate you sharing your thoughts, thanks! :)
So further fulfilling the self-made prophecy of sales folks only hiring other sales folk as software engineers only to deny the reality that some of the best software engineers are not necessarily the most social or sales-y
Agreed. Great explanation. I think it’s simple math:
AI doing things < Human + AI doing things
Humans will just be expected to output more.
Code-generating LLMs will change Software Engineering but obviously in a bad way since people will not spend enough time trying to understand a code.
I already see this unfortunately. We are going to see more over complicated software.
Previously it was copy-paste from stackoverflow, now it will be copy-paste from the generated code.
I've seen this happen with some coworkers already. They stumble when presenting their generated code bc they don't understand it
Well, both cases have a common denominator: people using tools incorrectly. Thus, I guess it's people's fault and not the tools'.
@@deathrives4252
I am not blaming tools i am just describing what I am already seeing. Stackoverflow at least does not provide you with so much code, you at least have to do something by yourself
Yes, agree that you have to understand what has been producdd or generated and be in the loop controlling the final outcome.. In some way. Experienced is important in understanding how to put code together and those principals and practices continue in the code automation workflow.
The main raison I follow your chanel, it's because of good python content.
Thank you for the contents!!
Great take! It is definitely different for people who are just getting started in development. SW tend to be pessimistic about AI but I think it is already disrupting the industry on many levels, so better to adapt and learn thanks for the advice Arjan.
Also, as a request, it will be great if you start making video series about advanced Python concept (like very advanced). Many follow you since years, and approaching seniority already. Thanks!
Thanks for sharing your thoughts Arjan
I'm a self-taught python learner. Before I got into coding i didn't know that there were entire teams working on various portions of a project. I assumed that all engineers were "full-stack" developers. Now that I've been using Copilot to help develop my own personal projects, I have a feeling that companies will attempt to use AI to reduce these teams and lean more on individual developers to do more coding with the help of AI.
Short term: it won’t replace existing developers
Long-term: once every software developer alive today ages out of the workforce, there will be very few people with coding skills because people will have become over-reliant on the easy “give me an answer” fix that the AI Tech Broligarchs are pushing
Won't matter is not true at all. If you are a Software Engineer AI will increase productivity, not just for the company you work for but if you have ambition to work on own projects/business then take advantage of the tools available. Software engineering is the thought process that goes into the code you write not writing the actual code.
" This is how we do it " : DeepSeek and offcourse your RIGHT bro.
One thing people forget as well is we’re ultimately dealing with tools. In the past 3 hours I’ve worked with three separate tools intended to help with AI coding, all of which had bugs I’ve just finished troubleshooting and working around. This stuff will take time to get to the corporate utopia where all that’s left is the robots and the board of directors. It’s going to be a while.
I must first have the distinct concepts expressed using clear language in order to get out a high quality product from a model. That is where people need to focus their education rather than being concerned about what concrete details will become automated. These models are the greatest tutors ever but people do sometimes need to be shown where to begin with them and how to hold a critical lens to their outputs.
For me personally the scariest part of that whole AI mess is that even now there is a shortage of Senior level developers all around the world. And where do they come from? From junior developers! And when those junior developers become unresistant or even obsolete because of AI, where will we get a single person who will understand how this all works under the hood?
thanks for your view as an experienced developer. I appreciate your channel because it is focused on coding, not how to get rid of coding by means of almighty AI, please stay the way you are ❤
I think that "using AI as a developer" would be a great vid topic (or course, etc.) ... I believe that would be valuable. (i.e. effective prompts for developing x or y, specific best practice, pitfalls, lessons leared) The "generalized" AI content on UA-cam is a bit overwhelming at the moment.
5:18 Arjan, but the reasonable thing to do, is to enjoy the increased productivity and to spend more time with your kids, wife, and family.
We should actually work less and enjoy the increased productivity. Instead, we are being forced to work the same amount, implement more features in less time for the same or even less money.
First, that is an extremely irrational way to go about life and second, if it’s not us, who is benefiting from the increased efficiency?
I totally agree. Unfortunately, that’s not going to happen automatically, because the ones who are benefitting are company shareholders and they’ll push for more output for the same amount of money…
Technology is here to make us work more and more.
Everyone dreams about robots working for us, but that will never happen.
Welcome to the industrial revolution.
@@ArjanCodes So maybe we should become the shareholders?
@@Firstname_SurnameYes, but again there will not be any worker bees and consumers :)))
@ Why not? You can be both a worker and a shareholder :)
Great content, Arjan
Since Christmas I thought I would dive in a get an AI Assistant to genrate some code for me in Rust. On the one hand I am amazed at how well it can do sometimes. On the other hand, even when it does well the generated code can have errors that that require manual fixing. Which requires that I know Rust pretty well. Often I found trying to get the AI to fix compilation errors in the cadet produced just makes it worse and worse until I give up. In all cases I would want somebody who understands the problem and the language to review the generated code and create meaningful tests of it. I conclude that those junior devs you were talking about would be wise to become as proficient in their language a programming skills as ever. Even if getting AI to generate code is saving time.
Google AIStudio answers this question correctly, DeepSeek R1 doesn't (its code doesn't compile):
In Rust, I have a struct like this:
struct MyStruct(v: Vec)
How can I implement the trait IntoIterator for &MyStruct?
I'm tired of reading and hearing about AI and AI slop. I say keep up the quality content to help us improve our Python craft!
AI generated code can be educational.
Great video!
Got to keep up with the Commodore..
I totally agree with your predictions.
LLMs are a slightly better google/snack-overflow or a slightly weirder pip. I don't think they are having a meaningful effect on our productivity at all.
My major worry is what managers think. I have seen a lot of managers in the last few months who thought that AI can do everything perfectly. This is really bad for the industry.
I havent listened yet but o3 just came out and i just payed for the $200 sub again its that good (as an engineer). when the r2 comes out it most certainly make a difference. ive only been using o3 a couple hours damn! this thing is WAAAAY better at coding than o1 or claude (even me lol)
This video is so relevant to me right now
I'm interested in exploring the emerging landscape of AI coding assistants, particularly tools like Cursor and Cline. As these systems become increasingly integral to software development, I'd value your perspective on their capabilities, optimal use cases, and potential impact on engineering workflows. How do you see these tools reshaping our development practices?
Please don't talk too much about AI. As other comments said, there is already a ton of that everywhere, and your channel has always been very unique with sw design, architecture,bbest practices etc... so to me this is where you stand out!
My thoughts exactly. I have been playing around with ollama today and you can use several pretty good llm's with it for free. So I certainly see AI becoming a free tool for everyone and the race to the bottom is happening. In the foreseeable future AI, or LLM's in particular, are not going to analyse or write big software projects. So I agree with your view and am not particularly interested in a lot of AI content. Getting better at writing good software is where it has been at and where it will stay.
AI-industry spends billions for a mediocre coder in expectable slow progress..? You hear my doubts
I'm thinking that these AI input will be more or less saturated. The information AI models use its data from internet, like StackOverFlow(SOF). But since people rely on copilot or chatgpt no one will ask questions to SOF hence the models will not get any more inputs therefore they will not learn.
But also as more code is llm generated will it just end up learning from itself? I wonder how these models behave when they are trained on their own output, it feels as if it would get worse since the error in their predictions accumulate, but I have no idea
It's a tool, not more, not less.. I have installed it locally and it helps with math and short functions, that's great for me but I never ever want to loose the ability to do real problem solving or just be a copy-paste 'developer'!
I think you have a point but think about it. If Juniors go away, which you didn’t say, who does the job when the current generation retires?
I am expecting that, at first, many (most?) tech companies will stop hiring (which they've mostly done) and maybe start laying off (which they're somewhat doing) because of AIs replacing labor. But, since tech companies usually follow the pendulum of hiring, firing, hiring, etc, they will up their expectations of what their employees will produce, what quantity, and what quality. And eventually, they will want to hire more labor, again, to use the AI tools for yet more and more quantity (their first priority) and quality (their second priority, or maybe even further down the list). So I'm expecting we will have a few years of low hiring of new, junior developers. But soon, companies will want them again.
Is it terrible not to use AI? I don't think i need to.
i think it will make juniors more valuable.
Really don't need AI for development. I would've thought it might be better for testing and error checking though.
Problem is c-suite lot think it's the next big money saver for reducing staff and making themselves look more efficient. What they miss is that it can replace them just as easily on the surface. So the thought should be assisting not replacing.
I'm happy that you haven't delved into it for actual development. Keep it real please ❤
It is a race to the bottom but the infrastructure costs ,esp with regards to power is going to be a huge cost.
These two things are conflicting. Right now AI is a loss leader across the board. Even the paid applications are not really making money so at some point these companies will have to find out how to turn a profit. I am not sure anyone will be in the space to self-host inference let alone training. We have companies that are reactiving Nuclear plants from the 70s like Three Mile to just pay for it. This is expensive and someone will eventually have to pay for this and so the "race to the bottom" is going to have some huge implications that don't include the "IP" or propetartyness of the model.
Now can we get to a generalized model that doesn't need training that is good enough to spend less on training and get better hardware to generalize inference? Maybe? But the costs still have to be passed on.
But this is a side point and not really the point of your video... I agree with everything else you said.
this is how I feel about job interview take home tests. Stop telling my not to use AI.. Just make the tests alot more complex and let me show what I can do in a real world situation (using AI)
Why is the title in Germany and the video in English?
UA-cam automatically translates titles nowadays.
i don't mean to sound mean, but as i heard at the beginning of the vid AI will be a feature integrated on Os. well, then apps will become irrelevant, maybe system and network architecture not. but apps/software will just be created and destroyed on the fly. as needed, as a load balancer. ❤
No, DeepSeek is changing your life as a developer:
* You won't have to go to and work in BigTec to be in the AI game or to professionalize. THIS IS HUGE.
* Now it is not about AI (which LLM/ecosystem/'algorithm'/...). It is about APPLICATION of AI. This is where the money is at. This is a HUGE shift.
Much, much, much more people are part of the game now. And big tech will have a hard, hard, hard time to centralize it.
I don't get what you are trying to say.
DeepSeek matter
Ultimately, AI will replace the software that humans can use. It will establish interfaces with hardware, networks, objects, and other humans to accomplish tasks.
Now, if you use AI to process large datasets consisting of human-created content-such as written statements, images, videos, and other forms of big data-it requires significant funding and is quite costly. DeepSeek & Co. make ideas real for researchers without funds. Programmers as researchers is one of possible ways to be not replaced by AI
Think about tasks to be done, not about programming. AlphaGo can play better than humans, and large language models (LLMs) are anymore just static weights-they are dynamic tools that can learn at inference time.
Except for government contracts where they get paid extra for the body count, the "code pig" jobs will definitely be affected. For people that do actual software design, I think AI is just another tool.
DeepSeek models are open source. How on earth powerful open-source models are not going to affect developers? I'm building these models into my systems right now, this very moment.
Do what you do best - that's what we come here for. Integrate AI if the supports what you are doing - like how to do what you do in the era of AI.
Please a video about pedantic AI
Uhhhh 10x cheaper AI assistant sure matters to me
10x developer == 1x developer after AInflation
AI makes diagnonys much faster and better than human, but doctors did not disappear. Now doctors may use AI to be more productive and correct in diagnostic. This is partially true for software industry. Why partially, because moving down the road AI will tend to become AGI which means that it will now automate human thinking processes. And engineering solutions which now we provide as an engineers will complete by AGI. From the other hands, history is a permanent development of industries and human always meet the situations where particular skills become obsolete. ANd so far we always were able to shift towards changing reality. This time it might be faster, but new generation will learn to manage it I am sure. Also we will find solution for becoming obsolete as an engineers (or any other industry workers). No one needs revolutionary change in job market.
I like the AI content and your views so far from this video
100% agree with this.
Yeah ... LLMs won't matter for developers ... The same way computers did not matter at all for accountants ... Correct ?
Are you for real man ?
100% on point.
AI will replace 90% of software engineers the rest of the functions will be basically proofreading and correcting small mistakes keep being in denial😊
Let's do some videos on how to implement design patterns with AI frameworks like LangChain or PydanticAI 🙏
Great suggestion!
@sam witteveen, check his channel out he has a lot of good content on AI frameworks and AI. Also Langchain has a youtube channel they also produce good content.
Yessssss
Yes, it would be great that you make videos about LangChain, but not just "hello world" examples, but much advanced that you definitely can 😊
No AI content please
AI makes you faster so spend some of the time you gained making whatever you did better
I'm in the IA area, I'm not an authority by any means. I agree 100% with you. Also I want to agree, hahaha
It's kinda funny. Senior engineers think junior positions will be less needed. Junior engineers think they can basically do the work of senior engineers now with the help of AI, equaling the playing field.
I'm kind of a slow pessimist, in that I think that AI will obviously write software. The LLMs are already getting pretty good at a lot of the mundane stuff, and they are able to do some basic thinking if you pay enough to the AI companies.
I believe getting them to think better than (very clever) humans may take longer than they two/three years timeline being discussed. But it will happen, when the AI is smarter than the developer and not just quicker and immune to tiredness, then why will companies need software engineers? There are some interesting questions about how software engineering develops at that point, as we'll be basically dependant on the AIs, if very few humans are thinking about AI.
But there are whole hosts of things we don't do in software engineering enough, because it is expensive, and it is expensive because you have to pay smart people to do it slowly. From formal verification, to extensive testing. Why doesn't your code have 100% coverage, it isn't because it is objectively difficult, it is because the rewards are diminishing and the fiddly bits, like creating test harnesses for the bits you can't test readily such as for fault conditions that are hard to trigger, take increasing amounts of time and effort to write manually.
Please no AI content :)
It would be good to see you build a non trivial app using as much AI in all phases of the SDLC as possible, perhaps in a series broken into the phase from idea to prod ready deployment, that is useful. Also the same for an existing code base. Incorporate the advice you have imparted in this video.
how to integrate AI (locally) to python projects to increase productivity?
do you think a computer can play chess?
(1975: a master chess player, smiling at you with compassion)
answer: it can't! it BRUT_FORCES its way through millions of possibilities, which don't work and finds the one that does!
do you think, an AI can code?
(2025_ software engineer smiles at you with compassion)
...
HANG IN THERE, MY FRIEND!😥
I hope, that Arjan will respond me.
What the difference between software developer and software engineer?
I asked ChatGpt about it and got some answers. But it's really interesting to know, what's your thoughts about this? 😊
😂😂😂 another anti AI struggler sentiment
It is too simple to use, focus on logic, convert it to pseudo code, focus on a small surface area of the system, feed it to multiple LLMs, use your knowledge and experience to select/combine the best
As an experienced developer I don't need it to be perfect, I just need it to be correct 80% of the time, the rest... it is up to me
Well said!
@@ArjanCodes nobody try to explain how to leverage this wonderful tool, I hope you do
1. Focus on a small portion then write a detailed pseudo code (90s term for today's prompt engineering)
2. Use multiple LLMs, many are free and each trained differently, giving different results - this is potent because it provides you different perspective to the same problem
3. Select or combine results
4. Test for errors, adjust as needed
Knowledge and experience is still a must, let us face the fact that you will never become a medical doctor using AI 🙂
@@commonsensedev I agree, too. Most of the time it generate examples or code that doesn't work. and believe me, i don't wanna fight with the prompt when I can write that code on my own. It has its use-cases, but copy/pasting is not.
The remaining 20% can be ironed out in a simple test-and-debug feedback loop (also driven by an LLM). Developers are obsolete.
@@vitalyl1327 if you have deep experience in deploying reliable, maintainable, upgradable enterprise grade system and used the current AI tools, I will definitely say NO, not today
If you will say, it can reduce the number of developers by 20 to 30 percent in a typical project without sacrificing quality, today, I can say YES to that
All the tools I have seen still requires a developer to effectively deliver a robust system
But what I see there soo many jobs today out there that will be under threat from AI agents, jobs that can be addressed by AI and traditional automation, many offshore virtual assistants have already lost their job
Coding might be totally automated in the future, but due to inherent complexity it is one of the jobs that will be last to go
Capo. Crack