@@blarghblargh yeah for sure, that's why I always plan the general outline, try to rule out any obvious shortcomings or dead ends and then start coding to find all the problems I wasn't able to think of. Because at a certain depth of analysis, the tree expands so wide that there are so many possibilities on things that could go wrong that you're better off just doing it and seeing what *does* go wrong.
This was my favorite Prime video in a while. It's concise and puts a bow on what you've been speculating about AI for a while. Really enjoyed this one.
@@thekwoka4707 "decent" lmao. And he does not have "bad ideas". He has own thoughts, in opposite to people who just repeat what youtube brocoders feed them
Honestly this is the kind of content I really love and I think it's probably your biggest potential in terms of value added through youtube videos, reflecting on programming articles is fun and all, but experiencing actual programming (or career) issues while coding a project and deriving nuggets of wisdom from it is much more interesting, at least for me
like another great video related to this would be what kind of conclusions did you get from doing that autoscaling, how does it even work? I have no idea myself, I would love to hear it from you
@@slpwrm YES! parsing through 4/5 hour VODs to know how he is actually making the stuff or what his conclusions are, is a pain. I want to know more about his experience with Zig, but parsing through endless VODFs where each has only a section with Zig, and find the ones where he worked with Zig is not a fun activity
I like this way more than a lot of your react content. Not trying to knock the react stuff, but I really like how genuine you are here. Not having chat also helps the pacing a bit. I guess what I’m saying is I hope you keep making these kinds of videos. Regardless, I appreciate all the content. You’re a gem!
I guess I don't get it. I've never worked in a language that had any boilerplate code, so it's really weird for me to hear that. Like what is it producing? I can't even imagine. Just get a better language.
@@InfiniteQuest86 Either have never written much code or don't understand what boilerplate means, likely both. Even writing a shabang line or exit true is a level of boiler plate.
@@InfiniteQuest86 If you make the same programs over and over, a bootstrap is nice. I have various boilerplates for various tasks in multiple languages.
@@barongerhardt You've definitely never written much code and don't understand boiler plate. You're using AI to write a shebang??!!?! Shame on you. You shouldn't have a job. A return statement isn't boiler plate and is faster to type than letting AI do it. But fast typing speed comes with actually coding a lot. You'll learn someday. I'm more than 20 years into my career. If you're writing boilerplate code, I feel sorry for you. You're probably doing something boring like web.
Dude, this is exactly why I'm learning programming the way I am even though both IRL acquaintances and the entire internet seems to be blasting me to use AI for just about everything in order to fast-track my way into a job. I do NOT want to be an AI Andy at a company just to cause more havoc for myself and probably others in the long run. I want to deeply understand what I'm engaging with, even if it's slower, because I've suspected that if I didn't, I'd find myself in deep waters not knowing how to swim. Thanks for posting dude.
I can relate, I want nothing to do with AI at the moment and believe the people leaning heavily on it to learn will eventually come to regret that decision.
@@Jeremyakneither of you make any sense. You can leverage AI as a learning tool to tell you why it wrote what it wrote and continue prompting it until you gain a solid understanding of the design choices it made and why. It’s trained on terabytes of information that allows it to understand the structure of “good code” and “good coding practices” and it understands it quite well. Using AI to blindly develop for you vs using it as a tool to help you learn and help you when you’re stuck are two different things
@@Youtoober6947I'm was really confused with how he used it because he doesn't strike me as the type that would just blindly use code spite out by the large llms without reviewing it and comparing and contrasting what the documentation says. It reminds me of that lawyer that cited a case given by chat gpt without double checking that it was. in fact an actual case that can be referenced?
This style of video is freaking great. This and the DHH interview pair well. Feels like the reward you get from reading the footnotes. The mental game and and focusing on mindset is not talked about enough and is more important than a lot of people make it seem. A growth mindset can be more important than a finished product sometimes, especially when getting better is the goal. Makes failing feel like a case-study instead of heart-rending.
I don't comment super frequently, but I just wanted to mention I really love this video! When you switched to full-time content this was the type of stuff I was hoping you would post!
For me this is your best channel. The first one is really nice and ThePrimeTime is more of a "distraction for coders", but this is where I come to really get better.
Hey, I really appreaciate that talk. I also very frequently prioritized the speed of implementation over the deeper understanding of the problem and I find myself paying the price for that now. You channel really helps with understanding not only the basics but also those deeper more insightful ideas about programming and project development. I'm happy to be a part of this journey of yours and that I'm able to learn from you! Thanks!
This kind of project is actually a perfect match for Elixir & friends. You get vertical auto-scaling for free, transparent horizontal scaling, self-healing, crash recovery, in-mem database, a bunch of introspection utilities, etc
I've been watching that first hand this week. A senior who has been using ai for 8 months has struggled significantly during a relatively complex piece of work specific to our code base. It required good contextual knowledge of the last 3 months worth of the teams work. The junior however, flew through it. The junior had never used ai tools, and instead had put in the upfront comprehension needed which paid off in dividends. The senior was seeking 'quick wins' but was floored by a lack of understanding after that outcome.
Love these show and tell kind of video! Don't have the patience to watch people coding but just getting the overview plus problems and learnings is awesome
I'd really love to hear more about the ecology aspect of AI too. If any of us can invest the time to find answers and use tooling with less of a carbon and freshwater footprint, we need to do it. AI still has a HUGE freshwater footprint. In a world where freshwater scarcity is becoming the norm, it really takes all of us to make a conscious use of tooling, as well as to demand companies reinvest in sustainability strategies (and abstain from using their tools if they don't).
I saw a comment on that Grace Hopper talk that got released recently that quoted her saying something to the effect of "when computers learn to think then people no longer will". I don't know if that's real or not, but either way I think it's definitely something we need to be wary of. You lose what you don't use, so personally I'm shying away from most tools that "increase productivity", AI or not, just because I think it'll pay off more in the long run.
This was huge, Prime. Competence leads to true 10x engineering. More code doesnt lead to easy maintenance, especially if you didnt develop it and learn from it.
Hey Prime, I just want to say thank you so much for being you and just enjoying programming. I am in my 3rd of my CS degree and enjoy my (very noob) programming but sometimes it is just so discouraging hearing about AI boosting this or that and being faster than humans etc, or how if I don't use AI then I'll be behind my peers. But your videos are always very inspiring and encouraging to me, helping me to know that it is okay if I am perhaps slower and not as "productive" as others, as long as I consistently put in the effort/time and enjoy myself. So thank you so much. Much blessings upon you.
Funny thing is this month I decided to slow down and start putting polish on things. Instead of just writing the code and putting out a PR, I’ve been letting the branch sit for a bit and I’ll come back and rename things, move things around, try different patterns, add clarifying comments/tests for unintuitive behaviors, and just give it some extra love. I only put up a PR now when there’s nothing else I can think of improving about it. It’s a bit slower, but I’m extra confident that that code won’t break. It’s a totally different feeling of accomplishment when you put out things you’re proud of rather than just meeting the requirements.
Oh hell yeah. I've been doing the same thing to my personal projects, and it makes such a difference for how much I understand the problem and how much I like the code when I return to it later. Plus, I've caught so many bugs by simply letting it sit and maybe even sleeping on the problem, it's insane.
im kind of a coder boomer and get completely missed by the content on your main. this type of video is much more digestible for me, and is something i'd sub to (i'm not subbed to any of your channels and found this via recommendations).
One of the things I do when I am using AI is if it generates something I'm not familiar with, like in your example the db.Exec() function is I'll immediately go to the docs the first time I see it generated. My favorite use case for AI though is stuff that I struggle to figure out even with the documentation. Stuff like awscli query syntax (or jq) confuse me every time and it's great asking AI to generate it for me in natural language and just assume the result might be wrong and quickly test it. I'll usually be able to iterate over the query even if it's generated improperly faster using AI than trying multiple times myself. Once I have it working I can then ask AI to explain the syntax step by step and slowly get a better understanding of how it works.
Now I have some people under me and I'm somehow the senior dev in my company, despite being only 5 years of experience. What I feel with the people starting now or trying to rotate into dev/data positions is they rely A LOT in AI. They don't really understand what is going on and what every lane of code does and why do they need to do that or not do it. The other day, one of my devs needed to pivot a table and create some columns in a dataframe. And later in code, he needed to unpivot the table again. I asked him how he unpivotted the table and he explained me a conundrum of 100 lines of code that he did not really understood at all. I asked him to flat the resulting pivotted table and the adding it to a new column, and 10 minutes latter after doing it step by step his head almost exploded. It's going to be a problem, but not in coding, but in everything. Even conversations, where people starts using AI to argue with their friends about historical facts or things they should remember...
well, many people forget that using the AI is also a skill. Recently I wrote a program that uses Winows COM interface. It's a mess, you define an interface with some guids attrbute, put slap a marshal attribute on an object that implements that interface and now you can use it. I used AI for it, but then I asked "what are those guids?" "Find docs about them" "Docs says this method return bool, not int" etc. By the time I was done just talking to the AI, it generated a scaffolding that just worked, and now I know enough about COM to be fairly sure it is written correctly. And it worked first try. If you just copy the first result from chatgpt, you are using it wrong. It's a tool you have to learn to use
There is definitely a time and place for "progress over anything", but outside of those few instances taking the time to understand and grow is going to be a much better approach.
Ai helped me learn the basics and help me get interested and have fun in the docs. At least it helps you know what you really wanna do or a bit of a way.
I really appreciate this perspective. i started playing hackmud this year. i used chat gpt hoping that i could get some trailheads from it while trouble shooting and proofreading my scripts hoping to end up reading some good documentation or retrieved information.... NOPE learn for yourself we say. you will be proofing chat gpt anyways. I definitely doubled my play time in hackmud just deciphering chat gpt hallucinating from trying to explain things to me instead of telling me where it's guessing from.
I love this content a lot I am a junior fullstack dev trying to improve as much as I can. You are mentor even though we don't physically know each other lol.
I clicked thinking it would be some sort of flex on how competent you are on programming. And came out with a heartfelt talk about using AI to code on a project. This video is good! THANK YOU!!
This is honestly such a good take. I feel when learning a language and especially programming for the first time you just need to dive in raw. No LSP and no AI. I feel it's the best way to learn. Also I was using copilot for like 6 months and I turned it off back in July as I just noticed I was getting lazy and getting copilot to write things for me instead of actually having fun coding it my self.
Loved the video format! If you already know how to write the code yourself then AI is likely just a tool for laziness, because in many cases you can probably write the code yourself faster than you can fully understand someone or something else's code well enough to debug it and build off of it.
Hi Prime! I just wanted to say that this video format is amazing! Showing the struggles and the lessons learned is not only useful but also inspiring to new coders such as myself and many others. Keep up the great work 😄
This was my whole experience in a SWE fellowship recently, I spent the last 2 years teaching myself web dev (especially next JS and the whole ecosystem around that) My entire team was relatively newer at web dev, and used AI the entire fellowship and it drive me crazy because I was the only one who knew the inner workings and really struggled while fighting against the AI code and they’re lack of understanding the entire 7 weeks of the fellowship Thankful for the dividends
100%. AI has been great at getting toy ideas up and running quickly. Especially in a domain I may not be super familiar with. However, its way more satisfying being self sufficient.
You are right, but IMO, stumbling through the errors and fixing it to proper behavior with the docs help makes for better understanding of that framework/platform/tool
so interesting your perspective, I'm trying to go trought this in my carreer I go up faster with projects, learning new things but I'm not a especialist in nothing, right now I understood that I want to learn in a deep way. that means use less AI but still delivered at time to be efficient
Just ran into this in a way. Had AI help me with a more complicated problem of sorting and switching and resorting of multiple arrays of things and it all seemed to work great, until I found a fatal flaw and the AI was unable to understand what exactly the problem was, so I had to go in myself and realized that I didn't know half the functions it used well enough to just fix it, which basically made the code useless. So basically, it's all good, until it isn't (and that will happen with current AI), and then you're stuck with code you didn't write, using stuff you never learned.
I have found AI extremely useful for writing use once then throw away code. Something where the output is easily verifiable so you don't need to understand how it works exactly, just that it did indeed work. If it's something you're gonna live with for a long time then it gets a lot less valuable. Knowing is better than receiving. That said, AI is still helpful for learning. I needed a way to randomly select 10 entries out of a 20 entry array with no heap allocations and in at least linear time. AI turned me onto the Fischer-Yates shuffle algorithm. I didn't have it write the code. But it did function as a much better search engine that had deeper understanding of my needs and surfaced more specifically relevant information, even with example code that proved it was a valid path to a solution.
I like this type of content for the authenticity and encouragement to take the harder path to benefit growth. I like the reacts as well because it gives me exposure to developments I might miss.
Engaging in the act of indiscriminate copy-pasting without comprehension is akin to the superficial consumption of explicit media -both provide a fleeting semblance of gratification while obscuring the deeper nuances. Just as one may find themselves perplexed by the dissonance between fantasy and reality, so too does the unexamined replication of code lead to exasperation and unforeseen consequences, ultimately resulting in a painful reckoning with the limitations of mere imitation.
I've tried doing networking in go recently server-client or client to client and i wanted to learn the basics of go and AI is a great tool to get code examples to learn from. Diverse and different. And also it's great that the code sometimes has bugs and a great way to learn debugging in go.❤
Literally 23.26 o'clock i promised myself i will learn better to work with linux and terminal and i will block 30min a day at least. My goal is literally to have control of my environment and to learn neovim because first it looks amazing sexy and of course because of your influence. i strongly believe in the 'slowcoding' movement which is a term i made up loool becuase literally i have been alwasy a slow learner but from the AI am not lying that am affraid and sometimes i get anxious. hugs from berlin
So AI is a great work simulator where you need to fix a bug or add some feature to the program that someone did 5 years ago. Of course that someone also left the company 5 years ago and no one else knows anything. My favorite way of learning new stuff
I kinda feel the same about stuff like React Native - at some point, you’ll need to understand the platform the app runs on, it’s just a matter of learning that now or later. So either learn it up front, or defer it to when you run into a problem. You’ll probably end up at equal or more time spent by deferring it.
I had a similar issue in SQlite with gpt. I don’t remember what I was doing, but I asked it for help and it gave me an answer that wanted me to join like 2 more tables together and some other stuff. I knew it was wrong and there HAD to be a better way, but it was dead certain that was the best way. 15 minutes of thinking later I found a solution on my own that did the same thing in one statement with all the joins I already had. It was a simple not equals statement. Never trusting AI except for obvious errors like typos. Not worth the risk.
I think you could have had a best of both worlds approach here with better prompt engineering. You can have the AI explain its reasoning step by step and also have it explain all of the functions / methods you havent seen before. Then you can "learn" while its showing you the code. You can also prompt it to give you options / alternative solutions that are less risky or have it evaluate your code. All that to say is that "just using AI to replace the work / boiler plate" is rarely ever the solution. I use it primarily to understand why something was done the way it was OR have it write something tedious when i already know how it works. If its a brand new thing i havent seen before, i usually prompt it to explain it well and identify potential issues in the future.
The thought that there are many people watching this understanding what he's talking about, depresses me about how little I've accomplished being in this industry for more than 10 years.
I feel like AI has lowered the barrier to entry for a lot of tools or at least made them more approachable and less daunting. It can help you get a broad overview/perspective of something and then you have to go in and read the docs for the details
Can't agree with that more. And this kind of though it's not applicable only to AI and dev, but also to "quick-response" sites like StackOverflow or ServerFault. Yes, the world (expecially the work-related environments) moves fast, and often we are pushed to give a quick solution instead of taking time to understand why what we found is the solution to the problem. Taking the time to learn, to read the documentation, to understand what we do and what we use is incredible valuable, either in the short and (but even more) in the long term, it develop not only our knowledge, but also our creativity to apply the same solution, the same flow of though that drove us to that solution, also to different problems.
Using AI to point to document is already a clever idea, I never heard people do that. They mostly talk about feeding errors into the prompt and let AI solve by itself. I'm a doc guy. I read all the doc, but sacrifice ability to iterate in confined time. But, even though I was kicked out of a project, I can re-make most of the stuffs by navigating document shortcuts. And I just learned that saving offline doc actually helps more that relying on online doc. Reading docs makes you feel incompetent in business, but it's fun when you have to solve a challenge and not just circumvent problems. But, to be competent in programming, thinking about it while doing it everyday is the only key to stay in that frequency.
For me the sweet spot is to actually spend that X amount of hours on the docs, and THEN use LLMs to speed me through the process of building the thing. Because by doing this I am much more prepared to challenge the AI generated stuff, which leads to getting it right the first time around, instead of waiting to run into crashes. There will be oopsies of course, just way fewer than blindingly accepting what the lil robot is saying.
If something feels like quick and easy success, it's just life telling you that you're setting yourself up for failure. Always worked like this, always will continue working like this
In the end chatgpt is also just a tool that needs to be learned how and when to use. And you did get more competent in when it makes sense to use chatgpt and when not
It reminds me of a old saying, "Weeks of coding can save you hours of planning." The times never change.
Unfortunately it goes both ways. You often discover most of your hidden requirements when actually making the code.
@@blarghblargh yeah for sure, that's why I always plan the general outline, try to rule out any obvious shortcomings or dead ends and then start coding to find all the problems I wasn't able to think of.
Because at a certain depth of analysis, the tree expands so wide that there are so many possibilities on things that could go wrong that you're better off just doing it and seeing what *does* go wrong.
"Plans are worthless, but planning is everything" @@NihongoWakannai
@@blarghblargh even more when someone starts using the software.
I haven't learned that phrase. Wished I learned it sooner.
This was my favorite Prime video in a while. It's concise and puts a bow on what you've been speculating about AI for a while. Really enjoyed this one.
Agreed. I love the content.. but I don't have hours to watch a typical ThePrimeagen video.
The post DHH clarity hits different
the post DHH monologue xD
one of THE best interviews/podcasts I've ever listened to. Constantly funny, constant knowledge and experience and just a blast all around.
He's got bad ideas on types.
But he obviously decent at some things.
@@thekwoka4707 "decent" lmao. And he does not have "bad ideas". He has own thoughts, in opposite to people who just repeat what youtube brocoders feed them
@@thekwoka4707let the man enjoy his dynamically typed lifestyle
Honestly this is the kind of content I really love and I think it's probably your biggest potential in terms of value added through youtube videos, reflecting on programming articles is fun and all, but experiencing actual programming (or career) issues while coding a project and deriving nuggets of wisdom from it is much more interesting, at least for me
like another great video related to this would be what kind of conclusions did you get from doing that autoscaling, how does it even work? I have no idea myself, I would love to hear it from you
@@slpwrm YES! parsing through 4/5 hour VODs to know how he is actually making the stuff or what his conclusions are, is a pain. I want to know more about his experience with Zig, but parsing through endless VODFs where each has only a section with Zig, and find the ones where he worked with Zig is not a fun activity
I like this way more than a lot of your react content. Not trying to knock the react stuff, but I really like how genuine you are here. Not having chat also helps the pacing a bit.
I guess what I’m saying is I hope you keep making these kinds of videos. Regardless, I appreciate all the content. You’re a gem!
Exactly this. His videos where he is jumping into his chat every 2 minutes drives me crazy.
1000%
The Primeagem.
@@randomtroll980 that’s ADHD for you
Second this. I kinda got bored of the constant article react
People say "AI saves me from writing boilerplate code" but maybe writing boilerplate code is good for you. Like stretching before running a mile.
I guess I don't get it. I've never worked in a language that had any boilerplate code, so it's really weird for me to hear that. Like what is it producing? I can't even imagine. Just get a better language.
Writing boilerplate codes prepare the brain for structure of the program.
@@InfiniteQuest86 Either have never written much code or don't understand what boilerplate means, likely both. Even writing a shabang line or exit true is a level of boiler plate.
@@InfiniteQuest86 If you make the same programs over and over, a bootstrap is nice. I have various boilerplates for various tasks in multiple languages.
@@barongerhardt You've definitely never written much code and don't understand boiler plate. You're using AI to write a shebang??!!?! Shame on you. You shouldn't have a job. A return statement isn't boiler plate and is faster to type than letting AI do it. But fast typing speed comes with actually coding a lot. You'll learn someday. I'm more than 20 years into my career. If you're writing boilerplate code, I feel sorry for you. You're probably doing something boring like web.
Dude, this is exactly why I'm learning programming the way I am even though both IRL acquaintances and the entire internet seems to be blasting me to use AI for just about everything in order to fast-track my way into a job. I do NOT want to be an AI Andy at a company just to cause more havoc for myself and probably others in the long run. I want to deeply understand what I'm engaging with, even if it's slower, because I've suspected that if I didn't, I'd find myself in deep waters not knowing how to swim.
Thanks for posting dude.
I can relate, I want nothing to do with AI at the moment and believe the people leaning heavily on it to learn will eventually come to regret that decision.
@@Jeremyakneither of you make any sense. You can leverage AI as a learning tool to tell you why it wrote what it wrote and continue prompting it until you gain a solid understanding of the design choices it made and why.
It’s trained on terabytes of information that allows it to understand the structure of “good code” and “good coding practices” and it understands it quite well.
Using AI to blindly develop for you vs using it as a tool to help you learn and help you when you’re stuck are two different things
@@Youtoober6947thanks Andy.
@@Youtoober6947 AI doesn't "understand" anything. It's not conscious.
@@Youtoober6947I'm was really confused with how he used it because he doesn't strike me as the type that would just blindly use code spite out by the large llms without reviewing it and comparing and contrasting what the documentation says.
It reminds me of that lawyer that cited a case given by chat gpt without double checking that it was. in fact an actual case that can be referenced?
This style of video is freaking great. This and the DHH interview pair well. Feels like the reward you get from reading the footnotes.
The mental game and and focusing on mindset is not talked about enough and is more important than a lot of people make it seem.
A growth mindset can be more important than a finished product sometimes, especially when getting better is the goal. Makes failing feel like a case-study instead of heart-rending.
I don't comment super frequently, but I just wanted to mention I really love this video! When you switched to full-time content this was the type of stuff I was hoping you would post!
For me this is your best channel. The first one is really nice and ThePrimeTime is more of a "distraction for coders", but this is where I come to really get better.
Hey, I really appreaciate that talk. I also very frequently prioritized the speed of implementation over the deeper understanding of the problem and I find myself paying the price for that now. You channel really helps with understanding not only the basics but also those deeper more insightful ideas about programming and project development. I'm happy to be a part of this journey of yours and that I'm able to learn from you! Thanks!
This kind of project is actually a perfect match for Elixir & friends. You get vertical auto-scaling for free, transparent horizontal scaling, self-healing, crash recovery, in-mem database, a bunch of introspection utilities, etc
I've been watching that first hand this week. A senior who has been using ai for 8 months has struggled significantly during a relatively complex piece of work specific to our code base. It required good contextual knowledge of the last 3 months worth of the teams work. The junior however, flew through it. The junior had never used ai tools, and instead had put in the upfront comprehension needed which paid off in dividends. The senior was seeking 'quick wins' but was floored by a lack of understanding after that outcome.
Love these show and tell kind of video! Don't have the patience to watch people coding but just getting the overview plus problems and learnings is awesome
yeah these vids are the truth
Totally agree!
I'd really love to hear more about the ecology aspect of AI too.
If any of us can invest the time to find answers and use tooling with less of a carbon and freshwater footprint, we need to do it.
AI still has a HUGE freshwater footprint.
In a world where freshwater scarcity is becoming the norm, it really takes all of us to make a conscious use of tooling, as well as to demand companies reinvest in sustainability strategies (and abstain from using their tools if they don't).
I saw a comment on that Grace Hopper talk that got released recently that quoted her saying something to the effect of "when computers learn to think then people no longer will". I don't know if that's real or not, but either way I think it's definitely something we need to be wary of. You lose what you don't use, so personally I'm shying away from most tools that "increase productivity", AI or not, just because I think it'll pay off more in the long run.
being competent with spelling is even more fun
difficultee level: impossibly
Ikd what day si
And being competent in spelling and punctuation is a superpower.
TLDR; AI codegen is the payday loan of programming where you speed through the code at the cost of wisdom and tech/knowledge debt later.
This was huge, Prime. Competence leads to true 10x engineering. More code doesnt lead to easy maintenance, especially if you didnt develop it and learn from it.
this is awesome content, pure code, design choices, do this 80% of the time please. you are a legend!
Hey Prime, I just want to say thank you so much for being you and just enjoying programming. I am in my 3rd of my CS degree and enjoy my (very noob) programming but sometimes it is just so discouraging hearing about AI boosting this or that and being faster than humans etc, or how if I don't use AI then I'll be behind my peers. But your videos are always very inspiring and encouraging to me, helping me to know that it is okay if I am perhaps slower and not as "productive" as others, as long as I consistently put in the effort/time and enjoy myself. So thank you so much. Much blessings upon you.
10:18 "I'm just going to talk about code and my thoughts on coding and life of coding" YES YES YES YES YES ITS FINALLY HAPPENING
Agree, Prime. Being competent is better. Keep doing what you do.
Funny thing is this month I decided to slow down and start putting polish on things. Instead of just writing the code and putting out a PR, I’ve been letting the branch sit for a bit and I’ll come back and rename things, move things around, try different patterns, add clarifying comments/tests for unintuitive behaviors, and just give it some extra love. I only put up a PR now when there’s nothing else I can think of improving about it. It’s a bit slower, but I’m extra confident that that code won’t break.
It’s a totally different feeling of accomplishment when you put out things you’re proud of rather than just meeting the requirements.
Oh hell yeah. I've been doing the same thing to my personal projects, and it makes such a difference for how much I understand the problem and how much I like the code when I return to it later. Plus, I've caught so many bugs by simply letting it sit and maybe even sleeping on the problem, it's insane.
im kind of a coder boomer and get completely missed by the content on your main. this type of video is much more digestible for me, and is something i'd sub to (i'm not subbed to any of your channels and found this via recommendations).
I much prefer this format! Your opinions tend to resonate with me. This one especially.
One of the things I do when I am using AI is if it generates something I'm not familiar with, like in your example the db.Exec() function is I'll immediately go to the docs the first time I see it generated. My favorite use case for AI though is stuff that I struggle to figure out even with the documentation. Stuff like awscli query syntax (or jq) confuse me every time and it's great asking AI to generate it for me in natural language and just assume the result might be wrong and quickly test it. I'll usually be able to iterate over the query even if it's generated improperly faster using AI than trying multiple times myself. Once I have it working I can then ask AI to explain the syntax step by step and slowly get a better understanding of how it works.
The DHH interview was really great. thanks for the great questions and the opportunity to listen to DHH for 2h straight.
Now I have some people under me and I'm somehow the senior dev in my company, despite being only 5 years of experience. What I feel with the people starting now or trying to rotate into dev/data positions is they rely A LOT in AI. They don't really understand what is going on and what every lane of code does and why do they need to do that or not do it. The other day, one of my devs needed to pivot a table and create some columns in a dataframe. And later in code, he needed to unpivot the table again. I asked him how he unpivotted the table and he explained me a conundrum of 100 lines of code that he did not really understood at all. I asked him to flat the resulting pivotted table and the adding it to a new column, and 10 minutes latter after doing it step by step his head almost exploded.
It's going to be a problem, but not in coding, but in everything. Even conversations, where people starts using AI to argue with their friends about historical facts or things they should remember...
well, many people forget that using the AI is also a skill. Recently I wrote a program that uses Winows COM interface. It's a mess, you define an interface with some guids attrbute, put slap a marshal attribute on an object that implements that interface and now you can use it.
I used AI for it, but then I asked "what are those guids?" "Find docs about them" "Docs says this method return bool, not int" etc. By the time I was done just talking to the AI, it generated a scaffolding that just worked, and now I know enough about COM to be fairly sure it is written correctly. And it worked first try.
If you just copy the first result from chatgpt, you are using it wrong. It's a tool you have to learn to use
I'm a big fan of this type of content. Very informative in a way that I can actually apply it in my daily work.
commenting to let you know I like this content. I appreciate hearing things you are passionate about and your perspective. Thanks for making this!
There is definitely a time and place for "progress over anything", but outside of those few instances taking the time to understand and grow is going to be a much better approach.
Ai helped me learn the basics and help me get interested and have fun in the docs. At least it helps you know what you really wanna do or a bit of a way.
I like this style of video and all the things you said at the end.
I really appreciate this perspective. i started playing hackmud this year. i used chat gpt hoping that i could get some trailheads from it while trouble shooting and proofreading my scripts hoping to end up reading some good documentation or retrieved information.... NOPE learn for yourself we say. you will be proofing chat gpt anyways. I definitely doubled my play time in hackmud just deciphering chat gpt hallucinating from trying to explain things to me instead of telling me where it's guessing from.
I love this content a lot I am a junior fullstack dev trying to improve as much as I can. You are mentor even though we don't physically know each other lol.
I love that kind of content. The focus, the tone, the story. That's really good.
I clicked thinking it would be some sort of flex on how competent you are on programming.
And came out with a heartfelt talk about using AI to code on a project.
This video is good! THANK YOU!!
Arguably one of the best videos on the side effects of gen AI in coding right now, great content 👏
This is honestly such a good take. I feel when learning a language and especially programming for the first time you just need to dive in raw. No LSP and no AI. I feel it's the best way to learn. Also I was using copilot for like 6 months and I turned it off back in July as I just noticed I was getting lazy and getting copilot to write things for me instead of actually having fun coding it my self.
Loved the video format!
If you already know how to write the code yourself then AI is likely just a tool for laziness, because in many cases you can probably write the code yourself faster than you can fully understand someone or something else's code well enough to debug it and build off of it.
Love the format; great way to place what DHH spoke about into a context that clearly shows the merit of what he said.
Awesome format, keep going! Love to see both active discussion and thinking on stream and such more dry and structured conclusions here
Thank you. Not being lazy is actually ... easier and more fun?? Took me a while to learn that.
“The Hard Way is Easier” - Zed Shaw, LPTHW
Hi Prime! I just wanted to say that this video format is amazing! Showing the struggles and the lessons learned is not only useful but also inspiring to new coders such as myself and many others. Keep up the great work 😄
Love to see some content on this channel. It's my favorite out of all of yours.
Love this kind of content. Would love to watch more of it.
Probably one of the most interesting advice I have ever heard. Thank you !
2:44 Is that how you “ping”?
This was my whole experience in a SWE fellowship recently, I spent the last 2 years teaching myself web dev (especially next JS and the whole ecosystem around that)
My entire team was relatively newer at web dev, and used AI the entire fellowship and it drive me crazy because I was the only one who knew the inner workings and really struggled while fighting against the AI code and they’re lack of understanding the entire 7 weeks of the fellowship
Thankful for the dividends
Here here. You’re one of my hero’s dude thank you for being you.
100%. AI has been great at getting toy ideas up and running quickly. Especially in a domain I may not be super familiar with. However, its way more satisfying being self sufficient.
Love this content format! Please do more.
You are right, but IMO, stumbling through the errors and fixing it to proper behavior with the docs help makes for better understanding of that framework/platform/tool
But so does writing your own code that doesn't work
so interesting your perspective, I'm trying to go trought this in my carreer I go up faster with projects, learning new things but I'm not a especialist in nothing, right now I understood that I want to learn in a deep way. that means use less AI but still delivered at time to be efficient
Absolutely love it. More of this please
Fave vid I've seen of yours in some time Prime (not that they've not been good, but this is a great format)
I really love this format! Please continue❤
Just ran into this in a way. Had AI help me with a more complicated problem of sorting and switching and resorting of multiple arrays of things and it all seemed to work great, until I found a fatal flaw and the AI was unable to understand what exactly the problem was, so I had to go in myself and realized that I didn't know half the functions it used well enough to just fix it, which basically made the code useless.
So basically, it's all good, until it isn't (and that will happen with current AI), and then you're stuck with code you didn't write, using stuff you never learned.
I have found AI extremely useful for writing use once then throw away code. Something where the output is easily verifiable so you don't need to understand how it works exactly, just that it did indeed work. If it's something you're gonna live with for a long time then it gets a lot less valuable. Knowing is better than receiving. That said, AI is still helpful for learning. I needed a way to randomly select 10 entries out of a 20 entry array with no heap allocations and in at least linear time. AI turned me onto the Fischer-Yates shuffle algorithm. I didn't have it write the code. But it did function as a much better search engine that had deeper understanding of my needs and surfaced more specifically relevant information, even with example code that proved it was a valid path to a solution.
If you think AI is only good for writing “use once then throw away” code you haven’t really utilized it’s full capabilities yet
Love the content brother. You got me excited about coding again haven’t felt excited about it in a long time
this came at the right moment for me, thank you prime.
I like this type of content for the authenticity and encouragement to take the harder path to benefit growth. I like the reacts as well because it gives me exposure to developments I might miss.
Love this format, keep it with the great work!
Engaging in the act of indiscriminate copy-pasting without comprehension is akin to the superficial consumption of explicit media -both provide a fleeting semblance of gratification while obscuring the deeper nuances. Just as one may find themselves perplexed by the dissonance between fantasy and reality, so too does the unexamined replication of code lead to exasperation and unforeseen consequences, ultimately resulting in a painful reckoning with the limitations of mere imitation.
I needed this today, thank you!
Yes, this content is better.
It would be great to know how you think of the implementation and explain some cool feature you built it.
I've tried doing networking in go recently server-client or client to client and i wanted to learn the basics of go and AI is a great tool to get code examples to learn from. Diverse and different. And also it's great that the code sometimes has bugs and a great way to learn debugging in go.❤
Love this content! Definitely down for more of this
Literally 23.26 o'clock i promised myself i will learn better to work with linux and terminal and i will block 30min a day at least. My goal is literally to have control of my environment and to learn neovim because first it looks amazing sexy and of course because of your influence. i strongly believe in the 'slowcoding' movement which is a term i made up loool becuase literally i have been alwasy a slow learner but from the AI am not lying that am affraid and sometimes i get anxious. hugs from berlin
making mistakes and debugging is great for learning. Reverse learning indeed.
So AI is a great work simulator where you need to fix a bug or add some feature to the program that someone did 5 years ago. Of course that someone also left the company 5 years ago and no one else knows anything. My favorite way of learning new stuff
I kinda feel the same about stuff like React Native - at some point, you’ll need to understand the platform the app runs on, it’s just a matter of learning that now or later. So either learn it up front, or defer it to when you run into a problem. You’ll probably end up at equal or more time spent by deferring it.
Mega loving this kinda content my guy
Love videos like this, super interesting and insightful
I adore this type of format! 💕
I had a similar issue in SQlite with gpt. I don’t remember what I was doing, but I asked it for help and it gave me an answer that wanted me to join like 2 more tables together and some other stuff.
I knew it was wrong and there HAD to be a better way, but it was dead certain that was the best way.
15 minutes of thinking later I found a solution on my own that did the same thing in one statement with all the joins I already had. It was a simple not equals statement. Never trusting AI except for obvious errors like typos. Not worth the risk.
I think you could have had a best of both worlds approach here with better prompt engineering. You can have the AI explain its reasoning step by step and also have it explain all of the functions / methods you havent seen before. Then you can "learn" while its showing you the code. You can also prompt it to give you options / alternative solutions that are less risky or have it evaluate your code.
All that to say is that "just using AI to replace the work / boiler plate" is rarely ever the solution. I use it primarily to understand why something was done the way it was OR have it write something tedious when i already know how it works. If its a brand new thing i havent seen before, i usually prompt it to explain it well and identify potential issues in the future.
This is incredible content, thank you!
The thought that there are many people watching this understanding what he's talking about, depresses me about how little I've accomplished being in this industry for more than 10 years.
I like this kind of videos.
But I also like the stream VODs of you actually making stuff.
I feel like AI has lowered the barrier to entry for a lot of tools or at least made them more approachable and less daunting. It can help you get a broad overview/perspective of something and then you have to go in and read the docs for the details
Can't agree with that more. And this kind of though it's not applicable only to AI and dev, but also to "quick-response" sites like StackOverflow or ServerFault. Yes, the world (expecially the work-related environments) moves fast, and often we are pushed to give a quick solution instead of taking time to understand why what we found is the solution to the problem. Taking the time to learn, to read the documentation, to understand what we do and what we use is incredible valuable, either in the short and (but even more) in the long term, it develop not only our knowledge, but also our creativity to apply the same solution, the same flow of though that drove us to that solution, also to different problems.
This came at a great time for me, thanks for sharing. I need more dividend investments in my work
Love this format, keep em coming Prime.
This is the reason I did away with copilot when Im starting a project I don’t know about.
Using AI to point to document is already a clever idea, I never heard people do that. They mostly talk about feeding errors into the prompt and let AI solve by itself.
I'm a doc guy. I read all the doc, but sacrifice ability to iterate in confined time. But, even though I was kicked out of a project, I can re-make most of the stuffs by navigating document shortcuts. And I just learned that saving offline doc actually helps more that relying on online doc.
Reading docs makes you feel incompetent in business, but it's fun when you have to solve a challenge and not just circumvent problems. But, to be competent in programming, thinking about it while doing it everyday is the only key to stay in that frequency.
sqlite doesn't have "for update" and locks the entire database during transactions. you'll have to account for that when you adapt to a different db.
I love this kind of yapping. It's the yapping of a subject Prime understands and is passionate about.
For me the sweet spot is to actually spend that X amount of hours on the docs, and THEN use LLMs to speed me through the process of building the thing. Because by doing this I am much more prepared to challenge the AI generated stuff, which leads to getting it right the first time around, instead of waiting to run into crashes. There will be oopsies of course, just way fewer than blindingly accepting what the lil robot is saying.
I love this these types of videos! keep them up and esp the stuff by DHH
I like this format, keep it up
Also I love this format, please make more
This is such a great video. This is the most important take on AI I've seen!
u can use mysql docker container for unit testing
I love this format.
edit: I think "The Yappeagen" is going to become a reality haha.
this video just means a lot! thank you
If something feels like quick and easy success, it's just life telling you that you're setting yourself up for failure. Always worked like this, always will continue working like this
In the end chatgpt is also just a tool that needs to be learned how and when to use. And you did get more competent in when it makes sense to use chatgpt and when not
great talk, like to see more video in this format.