In ThePrimeTime's video, the increasing reliance of junior developers on AI coding tools like GitHub Copilot is scrutinized. The video discusses the worrying trend of dependency on AI for coding tasks, leading to a decline in fundamental coding skills and understanding. As junior developers rapidly ship code using AI tools, their deep understanding and ability to solve complex problems suffer, potentially impacting future software development skills. The creator emphasizes the importance of gaining foundational knowledge and reminds us that while AI can aid speed, understanding and mastery should not be traded for convenience.
Highlights
There's a growing trend of junior developers feeling like they can't code without AI tools like Copilot 🔄
Over-reliance on AI is causing a decline in syntax knowledge and problem-solving skills 🧩
Striving for speed through AI often sacrifices deep understanding and mastery of coding 📉
Critical thinking and debugging skills are at risk if new devs continue to depend solely on AI 🤔
Engaging in real discussions and communities, like those on Discord, can provide valuable insights 🌍
Experiencing the process of coding without AI can provide significant learning benefits 🧠
Key Takeaways
Junior developers are becoming overly reliant on AI tools, hindering their skill development 🚀
The foundational knowledge needed for coding is often missing among new devs 🙈
AI tools can be beneficial, but they should complement learning, not replace it 🤖
Engaging with developer communities like Discord can enhance understanding and knowledge 🌐
Building projects from scratch is crucial for developing a deeper understanding of coding processes 👷♂️
Combining AI speed with deep understanding is key for future software development 🔧
Overview
The video highlights a concerning trend among junior developers who feel helpless without AI aids like GitHub Copilot. ThePrimeTime addresses how this dependency can erode the basic coding skills and foundational understanding essential for software development. He shares personal experiences and examples of junior developers shipping code without a deep grasp of what they create, emphasizing the importance of maintaining core skills.
The creator discusses the trade-off between speed and understanding in the context of AI-assisted coding. While using AI can accelerate the coding process, it often leaves developers with a shallow understanding of code functionality. The emphasis is placed on the necessity to remember that every line of code requires future maintenance and that comprehensive knowledge is crucial for long-term efficiency and innovation.
To combat the potential drawbacks of AI reliance, ThePrimeTime advocates for developing a learning mindset with AI, engaging with knowledgeable communities, and building projects from scratch. He argues for a balance where AI tools enhance efficiency but don’t replace the need for deep understanding and critical thinking. Ultimately, this approach aims to foster a new breed of developers who are both fast and knowledgeable.
Chapters
00:00 - 03:00: Introduction and Observations The chapter discusses a trend where individuals, after using AI-assisted programming tools like co-pilot, feel reliant on these tools. They experience increased coding speed but notice a loss in muscle memory, especially related to syntax, as they frequently use prompts instead of writing code independently.
03:00 - 06:00: Dependency on AI Tools The chapter titled "Dependency on AI Tools" discusses the increasing reliance on AI technologies, focusing particularly on programming and coding tasks. The author mentions experiences highlighting this dependency, such as challenges faced when developing a to-do app or configuring tools without AI assistance. It is noted that even helpful tools like GitHub's Co-Pilot can create a crutch, where developers may forget syntax or steps if the AI tool is not available, as exemplified by an instance of forgetting to log in to Co-Pilot, which led to issues in rewriting a configuration.
06:00 - 09:00: Learning and Understanding Code The chapter discusses the speaker's experience with code and their struggle to remember how to write a for loop in Lua, which led them to turn off co-pilot due to its impact on their coding ability. The speaker reflects on the importance of understanding and recalling code without relying on external tools.
09:00 - 12:00: Stack Overflow versus AI The chapter titled 'Stack Overflow versus AI' describes the process of rewriting a code base to make it more resilient and efficient, with a focus on features like voting. The author shares a personal experience of rewriting the entire project on long plane rides without internet, relying on integration tests to ensure everything was working correctly. This experience highlights the satisfaction of coding independently and meeting challenges without external help.
12:00 - 15:00: The Importance of Fundamental Knowledge The chapter titled 'The Importance of Fundamental Knowledge' discusses concerns about the current state of junior developers' skills in software development. It touches upon the issue that many new developers rely heavily on AI tools like CoPilot, Claud, or Jity, causing a shift in the ability to code independently. It suggests an era where fundamental knowledge is being overshadowed by an over-reliance on assistive technologies. The chapter implies a reflection on how this reliance might be creating a generation of programmers who are potentially less literate in traditional coding practices and problem-solving skills.
15:00 - 18:00: Concerns About AI Reliance The chapter discusses the concern of prioritizing speed and immediate shipping of AI products over true understanding and mastery. The narrative suggests that while companies boast about 24/7 operations and rapid code shipment, the deeper understanding of the impact or complete functionality of these technologies is lacking. It highlights a cultural shift where the goal is to impress the Silicon Valley elite and gain social media validation, particularly exemplifying Mark Andreessen's hypothetical approval, rather than focusing on the intrinsic value and long-term enjoyment of mastering technology.
18:00 - 21:00: Strategies for Better Coding Practices The chapter delves into the importance of understanding the deeper workings of code rather than relying on quick fixes. It highlights how the current trend of prioritizing speed over comprehension can have long-term negative impacts on coding projects. Emphasizing the consequences, the transcript discusses how each line of code adds to future maintenance costs, and how rewrites can both fix and introduce bugs, urging a more thoughtful approach to coding practices.
21:00 - 24:00: Final Thoughts The chapter 'Final Thoughts' reflects on the limitations of AI in debugging code, highlighting that there are instances where even AI struggles due to the size and complexity of problems and context limitations. An example is mentioned involving '03 mini' or '03' and tests (AG tests) to evaluate its effectiveness.
Jr Devs - "I Can't Code Anymore" Transcription
00:00 - 00:30 so I've been noticing kind of a crazy Trend and I've just been seeing it a lot which is just this repetitive post that keeps happening over and over and over again which looks something like this that it's like I can't code anymore ever since I started using an AI IDE like co-pilot of cursor I've become super reliant on it it feels amazing to Cod at a let's see a speed I've never experienced before but I've also noticed that I'm losing some muscle memory especially when it comes to syntax instead of just writing the code myself I often find myself promting again and again and this is just like you'll see
00:30 - 01:00 this over and over and over again so I wanted to read this kind of like point out these two that I just found these are from just two days ago seven days ago um can't code a to-do app all that kind of stuff the syntax part is really real yeah the syntax part by the way I have experienced the syntax thing uh even with co-pilot I remember when I co-pilot I uh what's it called I I rewrote my neovim config last year at the beginning of last year and I just forgot to just include or I forgot to just log into co-pilot with my new computer and so just didn't show up and
01:00 - 01:30 I was just like oh my gosh how do I write a for Loop in Lua I forgot it's you know 4kj in I pairs item if you want array or pairs if you want map right I just forgot that simple simple aspect and so I was just like what is this and it was just so odd of an experience that I I actually turned off co-pilot because it was eroding my ability to just to write code and I do want to kind of say that for one second this little rewrite that I've been doing of my of my delicious
01:30 - 02:00 delicious code base to be able to be a little bit more resilient to be able to do voting and all that and make it really really fast and snappy um I wrote I rewrote the entire thing on two 7h hour plane rides no internet just wrote the entire dang thing when I didn't know how to test I opened up the integration test and I just made like actual twitch integration test being able to actually go through and make sure that it all is coming through and all that and it felt fantastic just being able to code with no help with nothing unstoppably for hour after hour and so when I read these
02:00 - 02:30 things I just get really sad and I I I I wanted to bring up these two CU now we're going to jump into this one right here the big one okay uh we've actually read uh one of his blogs earlier and everyone seemed to really love it yeah in fact it was the it was his previous article AI is creating a generation of illiterate programmers so let's try this one out new Junior de let see new Junior developers can't actually code something's been bugging me about new devs and I need to talk about it we're at this weird inflection point in software development every Junior Dev I talk to has Co pilot or Claud or jity
02:30 - 03:00 running 24/7 their shipping code faster than ever but when I dig deeper into their understanding of what they are shipping that's where things get concerning you know this works on this works for a certain amount of time I think you know the worst part is is we've trade we've traded a lifetime of enjoyment and Mastery so that we can ship faster and maybe impress some dudes in Silicon Valley like and maybe one day Mark andreon will like your Tweet because you've managed to ship fast It's Kind
03:00 - 03:30 it's kind of sad right sure the code works but I ask why it works that way instead of another way crickets uh ask about edge cases blank stairs the foundational knowledge that uh used to come from struggling through problems is just missing we're trading deep understanding for quick fixes and while it feels great in the moment we're going to pay for this later of course I mean of course every line of codes you had just remember every line of codes you add is maintenance in the future every rewrite costs money every rewrite reintroduces bugs every rewrite introduces new bugs you've never seen
03:30 - 04:00 before it always happens every single time ask them to debug code and they'll look at you with the blanks there and debugging code sometimes like you get to points where AI can't even debug code it's just too big of a it's just too big of a problem right it's not even just too big of a problem sometimes they just the context is only there in certain amounts there's this one like uh what's it called oh remember when 03 uh 03 mini came out or 03 came out and they had these like tests the these a these AG tests to like prove like hey is this thing good or bad and and how how
04:00 - 04:30 effective are they and there's this one test where it showed like you know a bunch of crap on the screen right and one of them had these like two little lines right here and then like one right here and then the next screen had the same thing except for it drew uh lines in between these two things and then it had a couple examples of this and then the test example had the following which was like two on this side two on this side and it was unable to do it because it was drawing lines between these two now by technical input definition that
04:30 - 05:00 is true because it never actually covered that use case where it should have drawn a line across but there's a little special ability of humans to realize what the intention is regardless of the like the ambiguity of the information the information didn't show this the examples were all across whereas this technically fit in the examples and so it's kind of this painful realization that AIS are only able they they're only able to deduce it seems like it's very hard for them to induce uh uh the problem I Let's see we
05:00 - 05:30 can read that here in a moment I recently realized that there's a whole generation of new programmers who don't even know what stack Overflow is now that's fine that's just a like looking up information is not bad using lookups the more efficient a lookup is I don't think that's a problem right like I don't think it's a problem that uh I have a auto complete right here like looking up information is not an issue understand TS yes right I I the faster your lookup is I actually think the better it is
05:30 - 06:00 back when uh back when Cloud was not a chatbot but a man who invented the field of information entropy there was a different way to debug problems first you search on Google then hope some desperate Soul had posted a similar question as you had and if they did they find a detailed thoughtful and often patronizing answer from a wise graybeard on the site called stack Overflow you know I always had to question the wisdom of these graybeards of stack Overflow cuz I just want to just want to throw this out here who in their right mind would monitor a website where newbies
06:00 - 06:30 ask the same 10 sets of questions over and over and over again and then you smugly answer the question or even worse answer the question just in a comment which is kind of like the ending of stack Overflow like I think we all saw the end of stack overflow when often the questions were answered just in a comment and it was super smug it was the strangest thing they weren't teachers I know someone said oh they're teachers they're not teachers they're not teachers they're like referees I'm not
06:30 - 07:00 exactly sure what to what to call it there's something it's just a very strange strange strange World um here's one of the questions I posted 12 years ago see the top answer by Nathan wall who has earned a whopping 10,000 points and is apparently CTO since 2015 what is the difference between unary plus and number and parse oh number and unary are the same and parse float actually does a string thing well first off yeah this parses floats which also allows for some like i e cool
07:00 - 07:30 stuff um yeah what's the answer to this parse float parent is for parsing a string while number is for coercing a value to a number they behave differently yeah uh empty string he doesn't even put an empty string in here and these two are very these two are very hilarious right anyways okay good so as long as I can have a let's see so this is just the differences and the differences can you imagine that some dude wrote uh an answer with this level of detail raw without any Ai and for free my original question well it's funny cuz the AI is now trained off that my original
07:30 - 08:00 question was thoroughly answered of course but he didn't stop there I learned so much uh I learned so many new facts some of them I didn't even know existed and others I didn't even want to know but now they're etched into my brain forever I think a big thing is that again this is why I really valued my education at a university because you know what day one of programming was day one of programming was okay let's talk about INS INS how do ins work well first off you need to understand how memory Works ins are stored as ones and zeros
08:00 - 08:30 this is how binary works this is how an INT is theoretically stored obviously they didn't do twoos complement at that point which is I think fine right okay here's a value that'd be 2 6 uh 22 there you go that's 22 and you're like oh yeah okay I get this yeah int is stored in the balls and so it gives you this whole idea of how ins work and so as you understand how ins work you're then able to go okay so now I understand why these things work and it starts getting better and better and better and you build off
08:30 - 09:00 this whole you build off foundational knowledge as opposed to just like learning how to program and I think that's a big difference that people are missing they don't learn how they don't learn what like why things are they learn how things are like oh first learn about a function what's a function function is a reusable piece of uh like code that you can keep on executing from all over that's I mean yes that is a definition of a function but why does this exist this way what's actually happening underneath that you know underneath the hood it's dude bro it's a
09:00 - 09:30 go-to bro it's a go-to and a and some return crap right like you you lose some of this the reasons why things happen and so it's it feels bad it feels really really bad this is valuable no but rarely needed see I actually think it's the opposite I think it's valuable to know and it's constantly needed because then people because then people ask questions like this well what the hell's going on here why does this happen well one's a you can see one is just coer coercion of an object the other one is not and that's why you get into
09:30 - 10:00 something like this which that answer didn't cover why is an empty string zero well it needs to be coerced into some number right that's just the way it is all right let's see Junior devs have it easy uh they go to chat.com and copy and paste whatever errors they see even lazier ones don't even uh do the 30 second effort uh of toggling to a browser window they just use a tool that does it all in one place yep it's a convenient it's convenient and quick but but there's one reason that stack Overflow was Superior reading discussions by experienced developers about your topic is the best way to
10:00 - 10:30 learn yeah but then you could you could easily argue that uh AI can also do that uh but the one thing you're missing is that AI will oftenly confidently answer a way that is and that's just the way it is remember we just got done using beautiful Devon to build a sweet game and what did the sweet game get built with web pack delicious delicious webpack you know why cuz that's a very popular answer in open source so statistically speaking weback PS probably the thing you're going to be
10:30 - 11:00 using statistically speed versus knowledge gained I I actually I disagree with the fundamental premise or I disagree with the Practical premise of this document that stack Overflow was actually better versus speed of of chat jity but I actually agree with the fundamental premise of this which is that people don't strive for knowledge gain they they strive for Speed they strive for whatever makes them faster not what makes them better because here's the problem like I mean the general problem with that approach is
11:00 - 11:30 this is that as you can ship faster and you understand less your ability to ship new features and cool stuff kind of flatlines and slowly grows and maybe those that take more time to understand stuff will glow will grow faster but you're not getting any dividends dividends don't work by just putting up you know dividends work by long amounts of time right and consistent reinvestment whereas people who don't go that route maybe they they don't ship nearly as fast but as they go they get
11:30 - 12:00 that nice growth and that growth is so important because that's going to grow faster and faster and faster and faster and soon it's very very simple to be able to solve hard problems stack Overflow doesn't guarantee that you'll get any answer that's also true uh I'm already good fast is just a bonus now yeah fast is just a bonus now it clearly uh let's see it's clear what speed means but what's what's better well better is you are given a task how how well do you understand a task and
12:00 - 12:30 can you solve it right how many people have to go I don't know how to critically think stack Overflow what are the steps I need to go to do this okay or not stack Overflow chat chippity chat chippity what steps should I do to be able to solve this okay step one how do I do step one okay do these steps okay step 1.1 can you please do that for me okay awesome implement it okay step 1.2 how do I do that for me okay you you literally are just CL you're literally just being cursor cursor's going to
12:30 - 13:00 replace you budy like that's all you're doing is you're just being the manual transportation of small bits of code uh AI gives you answers but knowledge uh but knowledge you gain is shallow with stack Overflow he had to read multiple expert discussions to get the full picture it was slower but you come out understanding not just what work but why it worked I would say that there's something else that also happened a lot people who used stack Overflow more than once and kind of got the idea of how stack Overflow worked understood that when you asked a question you had to have like a lot of surrounding knowledge around it and so You' often do more
13:00 - 13:30 initial research yourself into asking a question and by doing more of that initial research kind of like asking a cooworker when you ask a cooworker after your first time of going there and go why doesn't this work and they're like what have you tried and you're like I didn't try anything and you feel super stupid for not trying anything and then you go back and that social contract of having to try stuff it makes you become like much much better because often you just go oh well that's the answer to my problem I didn't even know about this thing and then the initial research
13:30 - 14:00 solves a huge amount of that problem and you had to go and do it yourself and you like become much better you have the tribal knowledge of what you're doing think about every great developer you know did they get good uh by copying Solutions no they got there by understanding systems deeply and understanding other developers thought processes that's exactly what we're losing I'm not uh let's see I'm not trying to be the guy who complains about kids these days uh I use AI tools a daily I'm literally building one but we do need to be honest about what we're trading for this convenience I think that's I I think that's generally the
14:00 - 14:30 problem I mean I I use AIS I I probably ask a good four or five questions a week on on the AIS and you know it's hard for me to love it's hard for me to love what I get out of it because often I you know when you really press it for information what it tells you looks correct and I think a lot of people are being duped this way I feel actually genuinely bad because a lot of people have this problem they it's like this it's like an appeal to Authority problem is when you're told everything's like these multiple phds I mean these things what PhD does a does Chad chippity have
14:30 - 15:00 yes it has all of them it has all of them it has all all of the phds you just assume that it's correct and if you really take apart a lot of the information you'll often find that once Chad chippity gets into anything of any sort of depth it just starts kind of saying things and the things may look correct but they're not correct they're really filled with a lot of misconceptions and it feels really bad so I'm not an expert in biology so whatever question question I ask Chad jippy I'm going to be like oh wow yeah
15:00 - 15:30 that seems really good that doesn't mean I actually understand it and I don't I have no formal way of verifying that it's good or bad whereas when you do it with uh like when I do with programming I have a formal way of understanding it I can go oh that's bad and here's why it's bad it's really the gelman's uh gelman's Amnesia uh the Gelman uh gelan Amnesia this guy gentan Amnesia is a cognitive bias describing the tendency of individuals to critically assess media reports in a
15:30 - 16:00 domain they are knowledgeable about yet continue to trust reporting in other areas despite recognizing similar potential inaccuracies it's like it's like the perfect way to do this the gentan uh Amnesia effect it's so so amazing which is that you you read an article that you're an expert at and you completely take it apart and go a this is just trash article the next day it's about horses and you're like this article's great I know about horses and you're like no you don't like dude you just thought the article is trash yesterday and now today it's fantastic like what are you doing brother what are
16:00 - 16:30 you doing why are you slurping up stuff that you you literally know is incorrect all right I've been experiment experimenting with ways to fix this because let's face it AI isn't going anywhere here's what's actually working first use AI with learning mindset when it gives you an answer interrogate it ask why sure it takes longer but that's literally the point careful also you have to figure out ways to like double check its answers I also deleted I mean I recently deleted one of my partitions full fully trusting in AI by acccident uh next find
16:30 - 17:00 your tribe Reddit Discord Mastadon wherever the smart people hang out that's where you'll find the real discussion happening the ones that make you go huh I never thought about it that way there you go uh I mean Reddit reddit's kind of a tough one masson's kind of a tough one Discord is kind of the one you make it I think discord's probably honestly the best place to be not even mentioning X I know the best part is is that X made this article blow up and doesn't even get a mention but Discord is probably one of the best places to do good technical discussions
17:00 - 17:30 I think I think the rest of these are actually trash honestly I think the rest of these are pure trash this one's probably the best does your dis Discord server count not really sure uh I I would say that it would probably be better like if you want to learn about Zig probably joining Zig joining the zig like server is probably the best way to do that right it's probably the best way to do that honestly uh do code reviews uh differently instead of just checking if the code Works start a conversation with your team what are the approaches did they consider what did they pick
17:30 - 18:00 Let's see why did they pick this one make understanding at part of the process and and result you know this one sounds great but I guarantee you this one is going to uh that one's going to be really hard to implement you know like I love the idea can we all agree the this idea sounds great but it doesn't work because often you're like I need to get this stuff in and I'm already working on the next thing and now you keep calling me back to do something and no right but no one has
18:00 - 18:30 time for that yeah it's a major waste and so this this is a tough one build things from scratch sometimes yes all this okay so this is actually the best answer to me this is the best answer uh if you ever are looking for like an RFC to implement fully from scratch I think websockets is probably the best one to implement from scratch very straightforward one there's even a test Suite called Auto Autobon I actually had to pass Autobon uh I had the pass the Autobon websocket test suite for the my reimplementation of websockets in the Netflix product and so I had to pass all
18:30 - 19:00 these this is super good it's very very great it goes through every single one of these all the different craziness for creating web sockets and this is absolutely an amazing tool to do oh welcome back thank you Janet app oh Janet you rated hey Janet appreciate you let's go shout out uh Janet if you haven't seen Janet you're if you haven't seen Janet you're a loser okay think about that one think about that one damn it Janet damn it
19:00 - 19:30 Janet yeah I find Reinventing stuff from the like building it from scratch is such a great experience and you can you can go one level lower every single time because once you build something from scratch you can go lower right you have to first if you build a project using react just using Dom nodes then you understand then you can say hey let me try to build my own version of react and then you go oh that's how that works and then you can keep on going from there how would I go from server to client how
19:30 - 20:00 would I do like rehydration like if you're really into front end you could do that now I'm not really into front end so I'm just not going to do that but you can imagine yes AI can generate that authentication system for you but try building one yourself first that's actually why oh by the way that's actually why on boot. deev that's why uh we're going to be shortly releasing the uh HTTP from TCP is because I want you to understand how HTTP works so therefore when you you have hp3 or two or you're doing something you're debugging request response models you understand why it
20:00 - 20:30 happens which is very good it's very very good uh but try building one yourself first you'll write the you'll write worst code absolutely but you'll understand every line of it that knowledge compounds it com it dude it's great dividends looking forward here's the reality the acceleration has begun there's nothing we can do about it I'm hoping that we can do something about it that's I mean I am genuinely trying to do something about it by programming on stream having a good time showing that you can build something without AI and not only that but build something quick
20:30 - 21:00 fast and amazing and then on top of all that actually showing that AI can be really good for things and not using it as a as a crutch but using it as a tool to actually make you faster um open source models are taking over or uh and we'll have AGI running in our pockets before we know it I'm not exactly sure if we'll have AGI running in our pockets before we know it um I don't know when that will happen I'm sure someday at some point in some future that can happen uh I just don't know if it's going to happen the next 10 years because that's a lot of
21:00 - 21:30 power we will probably what we will actually end up happening is we'll have something in our pocket that you know makes a request off to a server and then that server will then have all of your information it'll determine what information you can and cannot actually have access to and that's what's going to actually end up happening uh but that doesn't mean that we have to make it uh worse for us developers the future isn't about whether we use AI it's about how we use it and maybe just maybe we can find a way to combine the speed of AI with the depth of understanding that we need to learn let let me know if you found uh other ways to balance this or let's see or let's see or tell me I'm
21:30 - 22:00 just being an old man yelling at Clouds either ways let's fix this together there we go um yeah I think we're I assume that we're on a train where we're just not getting out of programming right or getting out of uh AI I think that that is absolutely 100% going to be at all parts but what I don't want to see is people become absolutely incapable despite me not liking the stack Overflow part of
22:00 - 22:30 all of this I really do love the idea behind this which is don't just get answers you are not an answer relay right if you if you make yourself an answer relay you are in fact making yourself useless your skill and your depth will only be as deep as the llm is able to go and yes I know everyone tells you it's PhD absolute top level all this but anybody who has been in the industry long enough knows for 100% fact that if
22:30 - 23:00 you work with a PhD level programmer they are hands down the worst every single time so good answer good answer can agree yeah dude please dude s like literally there is there there is like a midwit curve that exists but instead it's like goodness of programming right you know this classic a midwit there's like a midwit version where it's just like this is education level or this is education level and this is skill and it's just like there is an education
23:00 - 23:30 level that it just like all of a sudden your skills I don't know what happens but you're an idiot I don't know what it is I have never I it's the it's the phenomena is incredible I don't know how it happens or why it exists but some odd reason it just it just happens uh too deep in theory yeah I'm not exactly sure it's it's the god syndrome I'm not sure I I don't know what the name of it is um all phds aren't engineers neither are software uh
23:30 - 24:00 developers so there you go hate to break it to you but us software us software Engineers are in fact not Engineers the name is the primagen