Software Engineers and IT Leaders are Dead Wrong about AI

[@SergeyNeskhodovskiy]

This one is the most balanced and reasonable angle to take when looking at AI. On one hand, it has made me (as an engineer) 8x more productive. On the other, in the hands of a non-developer it's still gonna fail to produce software that goes a notch more complicated above a simple to-do list demo app. It quickly turns into a cargo cult if you don't understand databases, networking, architecture, and all the complexity lying under the hood of a modern web app. I was satisfied to learn that even the prompting methods, levels, and prompts themselves that you use are similar to mine when I work with cursor.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @SergeyNeskhodovskiy! I love this kind of feedback! It's really validating to hear that other people like you are doing similar things and getting similar results. It means that, if others will learn this too, they can also be successful. As you know, it does take time and effort. I haven't discovered any way to get around that 😆

[@bentos117]

how do you measure 8x more productive? like, 8x more code written?

[@SergeyNeskhodovskiy]

@@bentos117 I was able to complete a piece of software (a WordPress plugin, to be specific) in an hour, that earlier would take me a day or so, with all the UI and CRUD stuff

[@flanderstruck3751]

​@@SergeyNeskhodovskiyfrom a day to an hour I'm guessing you're going with code you don't fully understand

[@SergeyNeskhodovskiy]

@@flanderstruck3751 Initial task definition generation took 2 minutes, manual corrections to task definition took 8 minutes, base code generation = around 5 minutes (multiple files), then I reviewed all the code (10 minutes), identified problems and improvement opportunities and asked ClaudeDev to implement them bit by bit for the remaining 35 minutes, so no, I did review and understand everything it wrote. I never release the code I did not review or understand. I'm the company CTO and irresponsible approach is something I just can't afford.

[@dennissdigitaldump8619]

It's like an artificial tutor, it won't give you the "right answer", just an answer. If you don't know code, it won't actually help. But if you do know code, it's a good bootstrap for something that works.

[@CodingtheFuture-jg1he]

Yep, I can see that @dennissdigitaldump8619! Anyone that tries just blindly having AI generate code without ever trying to understand (and review) it is begging for trouble.

[@tonnytrumpet734]

Maintenance, Maintenance, Maintenance, Technical Debt. Extremely important part of every bigger project, AI makes this worse :)

[@NoCodeFilmmaker]

One thing I think a lot of people underestimate is what powerful educational tool AI is. Even in college I struggled to grasp software engineering. However, working with AI in the past month over two projects, I've finally started to truly grasp and enhance my development skillset and abilities. Just like Premier is a tool that enhances a filmmakers ability to craft a story, I feel Ai coding tool will serve to enhance a developer's ability to deliver their vision. I don't see Ai eliminating the need to understand core fundamentals, but rather enhance a users ability to grasp them.

[@reyreyalldayday5708]

Yeah I've always used ai as a tutor or just another dev next to me to fill gaps in my knowledge.

[@CodingtheFuture-jg1he]

Thanks for your thoughts on this @NoCodeFilmmaker! You make outstanding points! So far, this channel has focused mostly on coding, but learning is IMHO possibly an even stronger use case for AI. I'm going to create some content on this theme soon, as I think that even folks who aren't sold on using AI for coding (and maybe some are just trying to use it to learn to code) should get benefit from AI in this area. I'd love to know more about how you're using AI to accelerate your learning.

[@RaZziaN1]

No you don't, if u use AI you learn how to use ai. But AI assisted learning is not teaching you pitfalls you would learn while struggling by yourself. Same as watching course how to build some webpage, after watching it you think you know, but while doing it by yourself your mind is blank.. there is no information there.

[@user-ze9tj9yj4t]

@@RaZziaN1It depends how you use the ai. you can give detailed prompts during your learning which will both teach you the intricacies you’re talking about and save you the time of struggling through trivial things for 20+ minutes.

[@HarpreetSingh-xg2zm]

@@user-ze9tj9yj4tyou learn the pitfalls after trying to solve the problem. The ai provides an answer more than it teaches you

[@SL3DApps]

It’s kinda offensive to think that SWE only code and the Ai will be able to replace them. SWE are solutions specialists that translate problems to solutions in many different ways outside of coding for example in the idea space. Good luck asking an Ai why X solution is better than Y solution for YOUR business based on client growth for the next 10 years.

[@CodingtheFuture-jg1he]

I agree with you @SL3DApps! We are MUCH more than code-slingers. The best SWEs are "critical thinkers" and, as you say, solutions specialists. I do have one thing to add to your last statement though... I routinely use AI to brainstorm, do market research, do major technology studies and comparisons (in my role as a software architect)... all that take into account specifics of the business, its environment, goals, constraints, etc. If you were to provide really solid context, both ChatGPT and Claude.ai are superb at evaluating and proposing solutions for your specific business profile and projections. It's all about the human getting clear, providing the necessary context (which we'd better be able to provide, whether human or AI) and being willing to work iteratively. It doesn't obviate my need to think, but it does remove a ton of cognitive load for doing the boring, but necessary parts. I'm now generating really high-quality technology evaluations in probably 1/20 or less the time it used to take me - and in a format that senior leadership and customers can easily consume.

[@draghicistefan1984]

Exactly!

[@daviddickey9832]

Its funny how LLM's basically stepped in as the new semantic search as google search kept getting worse.

[@chrisschneider1746]

AI is best at bite sized chunks. Either because the whole project is just a few hundred lines, or because a good architecture was built up first (by a human). It's really hard to drop an AI into a large pre-existing system that has its own idioms and style, and a large internal world of data and business concepts. I've had success: * Frameworking a new react page * Writing unit tests for an under tested class (wrote ~80% of the test, I verified and finished) * tiny refactors of confusing snippets. * Writing shell scripts on the edge of the app. little helpers, CI tooling and such. No luck: * any task that touched a handful of files, where they all had to agree. Like wiring up a bit of data flowing through the app. * dealing with our non-standard approach to using postgres. Nothing crazy, just not what 99% of the training would have supplied. I've given aider, and cursor a shot, and supermaven is a solid autocomplete engine too. Nothing is groundbreaking, but it's another nifty tool in my toolkit.

[@petaflop3606]

i agree with pretty much all of this. Very helpful for things like 'give me a shell script to automate doing x' and for introducing new concepts by asking 'explain xyz concept using examples in ' etc. But try to introduce it to a sprawling codebase where there is a lot of nuance and it really just misses so much of whats important that you can't trust it to actually write anything for you. Anything that goes off piste enough from its training data is a nightmare as well - so novel/creative solutions that you quite often need to produce in the real world of engineering, like patching interfaces between systems that don't quite play ball nicely, it really fucking struggles there too

[@chrisschneider1746]

@@petaflop3606 A lot of what I do is just encoding business rules. We have meetings, talk a problem to death from every angle, then implement. Very little of my programming day job is algorithms, fancy architecture, and so on. We spent the time to do our architecture right up front, database patterns, code organization and so on. But day to day, we're so far away from leetcode algorithm problems that huge portions of the coding corpus just don't really help.

[@petaflop3606]

@@chrisschneider1746 That's interesting. I'm at a very different place, I work at a SME that's very much growing and building and improving interop (we do medical image transfer primarily so its a lot of integrations work between proprietary systems, some of which we wrote, some we didn't)

[@CodingtheFuture-jg1he]

I really appreciate your feedback @chrisschneider1746! I agree with what you and many of the others who responded to your comment are saying. Really large codebases are a problem for current models, depending on factors, such as what you're trying to have the AI do for you, how well the code is designed/modularized/commented, whether you have a great design doc to feed into the assistant and lots of other stuff. This is why the aider team added a feature to let you cd into a specific sub-directory of your codebase and have aider ignore the rest of the code while you work on a task. That doesn't solve all problems for sure, but it certainly helps. But if the code isn't well organized, well documented or is just huge... yeah, we're still not there yet. We will be soon.

[@CodingtheFuture-jg1he]

Your experience - not really needing to develop complex algorithms and fancy architecture is by far the most common scenario. I've been doing this for almost 30 years - coding, architecting, hiring devs, etc. I typically develop enterprise apps and sometimes frameworks and SDKs. The number of times I've actually needed to fall back on any semi-complex CS kinda algorithm and really think about "Big O" type issues... I can probably count them on my 10 fingers. And I've worked in about 10 different industries on all sorts of systems. Those skills are necessary for certain types software dev - say you're a developer on the Cassandra DB project. Yeah, you're gonna need to rely on some complex algorithms or things like "quadratic time complexity". When we're interviewing a dev for our projects and I hear one of my fellow interviewers ask a candidate "What's the time complexity of this algorithm?", I chuckle. How about asking them something that they'll actually USE on our projects? For most software, simple, clean, easy-to-understand is FAR more important.

[@CtjogibdCecil]

enterprise-ai AI fixes this (Code complete projects in PHP or Python). sconceptions about AI revealed.

[@kapsxd]

I usually just discuss with it, so I bring it a domain problem, it generates a solution, I analyze the response and make some suggestions, iterate and I reach a conclusion and get to work, however this can only be done if you know when the AI is wrong

[@yassirman1]

I am a RPA Citizen dév, i had to learn power automate desktop during the past 4 years and I got a very good level. That was a life changing for me to be able to automate my tedious everyday tasks. And also that helped me to compensate my bad coding skills that I Always blamed my self for. It Felt like a cheat code in my life's video game. Now with the LLM's, I had another life breakthrough, as I dont open anymore PAD, unless I had to automate a GUI task. God level cheat mode 😂

[@CodingtheFuture-jg1he]

Thanks for sharing your experience @yassirman1! That just shows that there's not just one way to use AI and automation isn't reserved for hard-core software devs. Sure, if you're pushing out mission-critical apps that a business depends on, you're going to still need years of dev experience. But it's not like it's an all or nothing proposition. Some people will only use ChatGPT (for now), some will use lots of AI tools to do complex software work. But there's a TON of degrees between those extremes. Props to you for making it work for YOU!

[@brettgarnier107]

I've been a dev for over 25 years. Currently building my own product using a lot of Semantic Kernel. This is a new nuget package from microsoft for working with llms. The documentation changes quickly, there are lessons, there are juypter notebooks, there is source code, its hard to get through all of it and find the most "right" way to implement. I dump all that info into a .txt file and upload it to claude projects, now i have a semantic kernel architect to ask questions. I use a smaller version of that file to add into Cursor context when implementing Plugin, Prompt auto function calling and its right. This type of workflow will be adopted by other devs who also want to make smart applications, and we'll start seeing so many new ones soon. UX changes when you can just tell your software what you want to do,

[@CodingtheFuture-jg1he]

I really appreciate you sharing that @brettgarnier107! You're a great example of a developer who understands that it's about experimenting with new technologies, taking the time to really learn them AND explore realistic ways of integrating them into your workflows. You're not looking at the LLM as some "one stop shop" to do it all for you. You're still using your obvious solid engineering skills to work WITH the technology. I find way too many folks hyper focus on the LLM - then evaluate AI based on what they can get from the LLM in one or two shots - those are just unreasonable expectations, at least for the foreseeable future.

[@Tarkusine]

After almost two years of using AI I still have a hard time pinning down whether AI is smart or stupid, or fully describing what it's actually good for. I use it all the time. I format data with it, generate short programs, generate all kinds of little things. It helps me understand old code and is generally an aid for all kinds of things. It is however, prone to errors. I never trust anything it makes fully. If it's important, I always review what it does. You have to know what you're doing both with the task you're getting it to do and with the limits and functionality of AI. AI, as it is with LLM's, is not going to replace programmers but it is going to enhance them. Things that would've taken me an hour to finish, especially something throwaway, can be cut down to taking 5 minutes. I think there's bull on both sides of this issue.

[@CodingtheFuture-jg1he]

Thank you for that @Tarkusine! I couldn't agree more on all counts!

[@niamhleeson3522]

It's stupid but it can process a lot of data.

[@greblus]

I'm often surpriced by how good current big AI models are in explaining rust's borrow checker's subtleties 😊. It's a great help with unlimited patience and much more, considering how huge context windows they have.

[@CodingtheFuture-jg1he]

Thank you for sharing that @greblus! A great example of another use case for leveraging technology to accelerate our learning and our work! These days, I review a lot of code that's brand new to me. With an AI assistant or chatbot sitting right next to the new code, I'm not kidding when I say I can figure out what's going on at least 10x faster. Previously, if I came across code that was using some framework, language or language feature I wasn't familiar with I'd have to go Google it, sift through page after page reading lots of stuff I didn't need to know... I'm thinking of making a video soon that addresses all the other numerous use cases like this for AI and ignore the "code generation" (which is everyone's hot button, as if pumping out hundreds of lines of code every day is what the typical dev does😆).

[@jonchappell2]

You have given lots of great perspective and advice here, both the hype cycle observations and the practical examples of successful AI assisted code. Good!

[@grodesby3422]

So does the prompt eventually become the source of truth instead of the codebase? In which case we have a new, higher level type of programming language, with even more layers between code and execution

[@CodingtheFuture-jg1he]

Great question @grodesby3422! I'm not sure about the prompt as the source of truth, but I guess that depends on your definition of that concept 😉 But I am convinced that prompting (call it prompt engineering if you'd like) IS becoming the new universal skill. It's the one skill (aside from critical thinking) that's going to come into play in almost every aspect of software engineering - from brainstorming ideas to evaluating design approaches to generating your docs and helping you lay out your plan all the way through coding, testing and eventually deployment. We talk mostly about "coding", but that's not even half the picture. I agree with where I think you're heading with your observation about the new abstraction... we've been constantly moving up the abstraction ladder since the beginning of coding. Just think about what it would be like to still be writing assembly - probably not fun and certainly not productive. This is a next natural step in that evolution. It really shouldn't be a surprise to anyone.

[@equious8413]

Sota models have output token limits less than 9k. Fully functional repos for relatively small projects have token counts that often exceeded 15-20k This is all anyone needs to know to understand that an AI isn't going to be a simple one stop shop to create something for a little while at least.

[@patrickdegenaar9495]

Over the past year I was expecting huge productivity improvements in the projects developed by my EEE MSc students. It turns out the main limitation is that they still need to understand the code. If they don't understand it, I don't accept it. So, net improvements appear to be around 50%. With this year's ai models, which derive the maths behind the code, I would expect 100% (i.e., double) improvement.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @patrickdegenaar9495! And good job making sure your students understand and can explain what's being created - regardless of who's/what's creating it, human or machine. Although I'm a huge proponent of learning to leverage AI to automate a bunch of stuff, I am a bit concerned that some of us will defer too much to the machine and much needed skills may atrophy (or never be learned in the first place). The human has to remain in control and we can't do that without fully understanding what's happening.

[@atoms.channel]

If I could double thumbs up this video I would; you covered this ground so well and so much 🔥🔥, and I've already gone through it twice now since it's so thorough. Oh, I also signed up for your Coding the Future with AI, Skool Community Course.

[@intrestingness]

Great point. Your job is to deliver business functionality. That you do it with code is secondary

[@CodingtheFuture-jg1he]

Thank you for that @intrestingness! You get it! Doesn't minimize coding, but puts it in perspective I think.

[@AaronBlox-h2t]

Great video....I"m going to go through your archive and hope to see more in future. Worth a sub, for sure.

[@rquintino4313]

absolutely top work, super insightful, packed with solid examples, balanced view and exploring both sides with open mind easily my top video on the topic in the latest week, subscribed!

[@CodingtheFuture-jg1he]

Thank you so much for your kind words @rquintino4313! And I appreciate your sub! I'm hoping to see more people talking about how we need to stop listening to the hyperbole and instead use our own critical thinking skills more.

[@DangRenBo]

I love that Aider is dogfooding. 1. It's probably the best way to surface papers and needed functionality. 2. It reminds me of compiler development where a major milestone is self compiling. So once Aider is developing almost all their new code with Aider, you know that it's ready for at least complicated Python development.

[@CodingtheFuture-jg1he]

Bingo @DangRenBo! You called it. aider is by NO means an example of "simple" coding. It's doing quite complex stuff. And I've taken a peek here and there at the code and it's damned good, IMHO. And it's a REALLY small team and they pump out impressive releases very frequently. So, what does it tell us that the aider devs are using AI to do all this and do it fast and do it well? Says that AI itself may not be the limiting factor - at least, not to the level some would have us believe. More likely that most of us just haven't yet made the effort to master the skills required to leverage AI like they have 😀

[@justdoityourself7134]

The uncertainty of if the AI is hallucinating, even for one single character of code will prevent AI from writing mission critical code, ever. That being said, the remaining 90% of code that is bug ridden anyway will be migrated to AI in the next decades.

[@snorman1911]

You are entirely correct, and I apologize! - AI

[@BruceWayne15325]

AI isn't going to replace software dev's for a long time. That doesn't mean it won't have a significantly negative effect on our salaries though. AI is going to have the effect of universally averaging salaries across all industries as it grows in capability. Why? Because in the not too distant future, AI will enable a junior developer, fresh out of school to easily code as well as a senior level developer with decades of experience. Now there's more to a senior level developer than working with complicated code bases. But it will mean that companies will prefer hiring junior level developers over senior. Ultimately, this will result in many senior devs being forced to take a junior dev salary if they want to keep working. As AI continues to become more capable, the bar will continue to lower (across all industries). Eventually, just about anyone could do just about any job. Salaries would be normalized across the board. This means that the income gap between the rich and the middle class will peel away like a bad band-aid. Lots of upper-middle class Americans will lose their homes when they aren't able to pay their mortgages due to the decreased salary, this could cause a collapse like 2008. Tough times ahead all around, but I still think the future will ultimately be brighter once we get through this terrible transition.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @BruceWayne15325! Man, you thought that all out SO well! It definitely makes sense. Now, I'm not really sure all the details will work out just that way, but the gist is in line with what I and one of my friends were talking about just a few days ago. One concern I have related to software engineering (IT in general) is how we ensure that we don't end up in a situation where you HAVE to have senior engineer-level expertise (which we absolutely will) and it's super scarce - because we became overly dependent on AI, went too far with the "AI + Junior = Senior" thinking, and other misguided thinking. One way for senior devs, who already have years of experience in software engineering, to ensure they're in high demand and compensated well is to add AI expertise to their skillset. We're on the cusp of a period in which engineers with deep expertise in the combo are going to be highly sought after, IMHO. Not quite there yet, but it's coming soon (my guess - by end of 2025 latest). I think very soon most companies are going to consider at least basic AI proficiency to be table stakes for engineers - as in, "you won't use AI in your workflow? Then you won't work here". I know that's controversial, but I fully believe within the next 2-3 years, an engineer who won't use AI will be thought of like a developer who doesn't believe in testing their code. Those folks simply won't be able to compete in the market. On the socio-economic front, oh man, we're likely in for a tough period as the world adjusts to the new reality. But I do believe, if most of us remain calm and level-headed and start preparing ourselves now, the other side can be a better world than we live in today!

[@africandynamite9630]

@Bruce wow interesting take. My grandfather had a weird ruler thing called a slide rule. Apparently in the 70s people would get hired just because they knew how to compute complex calculations with it. When calculators were created no one needed a slide rule expert anymore.

[@BruceWayne15325]

@@africandynamite9630 I'm probably dating myself, but when I was in elementary school I had a slide rule :P. I could actually work it faster than a calculator in many cases. I don't know that AI will ever replace ALL jobs, but I do think it will reduce the number of jobs to the point that very few people actually have them. It would be a privilege rather than a requirement. People would do the job because they wanted to.

[@00peterv00]

Thank you for the video. Can you explain how you created the sequence diagram.

[@engcurious]

I don't anyone should ever pay attention to demos, AI or not. A correctly skeptical view comes when you have a competent engineer attempt to use it for things, and a much more frightening view comes when you consider the risk this poses to us growing the next generation of engineers. I've got one video already on this, and will be doing more, my focus is on growing engineers.

[@equious8413]

It's kind of ironic that SWE historically don't like middle management. Mark my words, the SWE of today will be the PMs of AI teams tomorrow

[@madeniran]

In a nutshell it’s a better version of autocomplete/ copy+paste. Since we all agree that it won’t give the best results in 1 prompt. It will be iterative prompting until satisfaction.

[@CodingtheFuture-jg1he]

Thank you for that feedback @madeniran! I'll say yes, I agree with your last statement. But IF I'm interpreting your first statement, I think you may still be underestimating the power or value of AI here. I can see why it may seem like the value AI and coding assistants add is similar to when we've just added, say a really great IDE extension in the past that sped up aspects of our work. If I communicated this in a way that makes it seem like I'm saying "AI's just another incremental step in the natural progression of minor productivity improvements", I apologize. I believe no such thing. AI is and will be transformational. There's no doubt at all. The fact that we still have to prompt AI multiple times to build an app doesn't change this fact. That's a major oversimplification.

[@vitalyl1327]

But, it can prompt itself iteratively. With a feedback from tools.

[@alexrafter]

Really interesting take. Subbed! 👍

[@j01237]

Isn't replacing all drivers by last decade?

[@CodingtheFuture-jg1he]

Did you mean to say "developers" vs "drivers" @j01237? Just making sure, because this channel is about AI for software. We don't get into non-IT applications of AI. But if you did mean "developers"... I really cannot see AI replacing all or the majority of software developers by 2030. I mean, even if AI does reach the point of being able to take in a bunch of very detailed context that describes exactly the requirements and constraints of a very sophisticated software system/product and then work on it (as a team of AI agents) until it gets to "working" code, there are just too many other variables that will prevent it from being so widely deployed to make us obsolete. And I still see roles for lots and lots of skilled software engineers even in that world. I think we'd just shift our focus away from hand-crafting so much code and tests and to other higher-level functions that humans will likely continue to excel at for quite some time. And... none of us can predict anything that's going to happen next month at this point, much less 6 years down the road.

[@amotriuc]

When you say ready, ready to do what? For me it works only as replacement for stack overflow and not a really good one yet since it tends to hallucinate on things that you can't find on stack overflow or gives similar solution as you get on stack overflow.

[@marcelbricman]

lol ‚CORS‘ is kinda like a catchphrase at our office. when ever a dev gets a CORS problem they scream ‚CORS!‘ and everybody has a laugh knowing thats another wasted hour down the drain

[@CodingtheFuture-jg1he]

That's funny @marcelbricman! I only run into CORS very occasionally these days, since the software systems I oversee are mostly backend distributed systems comprised mostly of microservices. I knew it was common, but didn't realize it was so much so that it warranted its own catchphrase 😆

[@artscience9981]

Paraphrase: “I estimate it would have take five to ten times longer to write this code by myself than I was able to produce the app with AI assistance.” Does that mean that after the transients have died out, we will need 10 to 20 percent of the human coders we have now?

[@CodingtheFuture-jg1he]

Great question @artscience9981! I can't predict with certainty. But I don't think that will be the case, at least not for a very long time yet. I think we'll just raise our expectations around our level of output and productivity and a lot of the tasks we software engineers (and frankly almost everyone) will focus our energy on, while AI handles a bunch of stuff for us. So, I'm thinking yeah, maybe humans only need to WRITE 10 percent of new code soon, but that doesn't mean they expertise won't be required. Having said that, over time, it's going to be challenging to make sure human devs keep up their technology and problem solving chops - and that means we can't become complacent and just start entrusting everything to AI.

[@davidboreham]

No

[@twerner5496]

Brilliant explanation, thank you

[@mohdluai1808]

The thing is companies would need to have bot AI and developers to build the products which wouldn't be sufficient enough for them .

[@CodingtheFuture-jg1he]

Hey @mohdluai1808! Yep, it's going to be more like a shift in how developers build products vs "AI instead of developers". At least in the next few years - who knows beyond that. Even if you had an army of AI agents building the software and all you needed to worry about was prompting it with all the requirements, no sane company is going to not have knowledgeable software engineers reviewing/guiding/correcting that process.

[@AbhijitKrJha]

One request or suggestion: your way of explaining and approach is just awesome, so why not create a series on tool like aider for real world project development and put it on platform like udemy so that anybody can find it. Also, if possible, do some videos or entire series with cheaper models like deepseek or gemini flash etc.

[@CodingtheFuture-jg1he]

Thank you so much for your kind words @AbhijitKrJha! And for your great suggestions! I'm planning to launch a Skool.com community very soon. I'll be offering more in-depth content such as what you suggested there. I agree that what's needed is more details on exactly how to use these tools on real projects. Please keep the suggestions coming. It really helps me when I get such concrete requests to know what you would find helpful.

[@slamislife74]

What's the product you used to build these apps?

[@CodingtheFuture-jg1he]

Thank you for your question @slamislife74! I used the aider AI coding assistant mainly (used Continue for a couple of parts) and the Claude 3.5 Sonnet LLM.

[@slamislife74]

@@CodingtheFuture-jg1he Very cool, thanks! :)

[@aj-uo3uh]

With copilot I have the same speedup as with the Google search field which I use already 25 years. Sometimes it's better to do the Google search.

[@CodingtheFuture-jg1he]

Sorry, not following @aj-uo3uh. You're developing code using Google search? We are talking about software development here.

[@nickbarton3191]

Good advice, had a go last year and wasn't successful so gave up. Will try again.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @nickbarton3191! Good luck with round 2 😉 I recommend you keep at least 3 things in mind as you give another whirl: 1. At this time, stick to Claude 3.5 Sonnet 2. Be sure to provide AI with the context you'd need to provide any human developer; and 3. Always work iteratively when using AI - make a fairly narrow update, test it, then next incremental update, test it, etc.

[@mohammadsszai3019]

I'm learning software development javascript language is bit difficult, is there any future with that please assist on below.

[@CodingtheFuture-jg1he]

Thank you for that @mohammadsszai3019! Although I don't pretend to predict the future, I can say with a fair amount of certainty this: Not only is there still a future in learning to code, but I also see this as THE ideal time to learn to code. Why? Because when you augment your learning with AI you learn far more rapidly. Instead of a "study for 6 months before you can actually build anything" approach, you can now use AI to help you learn faster and to build some things along the way. Having said that, you should still review what you build - but, again, you can ask AI to explain various parts of what's being built. BTW, I do plan to publish some videos (soon?) to talk about some ways you could use AI to help you learn faster. Now, a year from now, my answer could change. My crystal ball is in the shop right now 😉

[@draghicistefan1984]

This is probably the most unbiased, clear headed and realistic video about AI code assistants. Great job!

[@CodingtheFuture-jg1he]

Thank you so much for your kind words @draghicistefan1984!

[@captaingabi]

Ai is just a faster stackoverflow search. It is very usefull indeed. But only engineers, and among them seniors and experts can prompt it efficiently. In fact, when I saw juniors prompt it and using its output, I was facepalming. Then mentoring them what they did wrong.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @captaingabi! Yep, I agree that unless you're just building something like a simple chatbot for personal or "friends & family" use, you need engineering skills. Glad you're mentoring the junior folks before they get those bad habits ingrained 😃 But I personally see AI as far more than a faster search.

[@jeffg4686]

was this before or after the latest OpenAI models? They'll be there soon enough that we just type in specs (for pure software) and specs plus some prior AI generated content for a game.

[@CodingtheFuture-jg1he]

Hi @jeffg4686! This video was produced before OpenAI released their latest o1 models. I haven't had a chance to really exercise those yet. My analysis is mostly based on Claude 3.5 Sonnet. Now, you can bet that Anthropic is also busy working on an even better Claude that 3.5, which is already (no matter what they nays gotta say) really good. Having said that, I think if the kinds of games you're talking about are the "snake" level games, sure... LLMs will soon be nailing those really easily. But, if you're talking about really complex software, including rich video games... I'm still in the "not so fast" crowd. At some point, just probably not very soon.

[@jeffg4686]

@@CodingtheFuture-jg1he - Who knows... The pace is nuts. It might be here tomorrow.

[@ajcics]

I'm creating a very sophisticated booking app with little coding experience. It's definitely not overhyped

[@KS-tj6fc]

Great video as always, I love your mixture of business and economic realities with the technical aspects of coding and software development. Could you please put out a video specifically detailing your prompt engineering, iteration, and specific examples/use cases of how you prompt the AI via aider or LLM’s in general as automated executioners or coding assistants? I think a critical point for this new video idea is the question you posed in this video - in which your task is for advancing the business objectives and mission not just the act of producing code. As such, prompting must guide the AI in whatever role towards those ends and not just the endless pursuit of “perfect” code.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @KS-tj6fc! Yes, that's a great and much needed tutorial. It's one thing to watch someone develop a specific app using AI. But we need more broadly applicable strategies that help folks really understand "why" things need to be done a certain way and how to approach various use cases. If not my next video, it's going towards the top for sure!

[@KS-tj6fc]

@@CodingtheFuture-jg1he First of all - Thank you for replying! Secondly - I so look forward to this video, notification turned on!

[@stefano94103]

I really appreciate your balance view on this topic. I live in San Francisco and it’s talked about a lot here. From talking to C-suite employees they are mostly Ai zealots. IT and engineers are the ones saying it won’t happen anytime soon. I think it will continue to get better faster than most people think. More than 50% of all code is now Ai generated. No one should be shocked that in five years 100% will be generated by hyper intelligent Ai.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @stefano94103! I love hearing from folks who are talking to actual C-levels and also engineers, especially when it's in the Silicon Valley area. I agree with you that many are underestimating how fast AI's going to improve. It's not really the "technology" they're underestimating - it's the super smart and ambitious humans behind it. We can't predict the actual trajectory or when AI will be able to achieve "elite coder" level, but I find it odd when people express the opinion that "we've reached the limit with AI". That view just boggles my mind - and from some really intelligent people at that. Seems to me that, folks at both ends of the spectrum are basing their views on their hopes or fears for their own financial future (big shock, right?). The C's and investors see big dollar signs - but for that, they have to believe that AI is a rocket ship to glory. On the other side, engineers fear for their financial future. Instead of taking the wise approach, which would be learn all you can so you're super valuable to the market either way, they seem to think just going on about how AI won't go anywhere is going to make it go away. I know from talking with some engineers that ego is also driving their AI bashing. Probably the same set of devs who during code reviews reject code because "it's not the way THEY would have written that". Fear will get us nowhere. What's going to happen will happen. All we can do is accept, learn and prepare.

[@m12652]

All good stuff but, so far no mention of the processing cost to your pocket or the environment. Even if it's reliable in a few years, the fact is it's going to be massively more expensive than it is now and just another external dependency for most. Cutting costs rarely results in a better value in the long term, instead the tools used to save money become more expensive over time and invariably even more expensive to migrate away from when something better and cheaper comes along.

[@hristoistoyanov]

Very well thought out opinion piece, coming from obviously an experienced software engineer!

[@CodingtheFuture-jg1he]

Thank you so much for your kind words @hristoistoyanov!

[@northofvalhalla5087]

Very solid and pragmatic overview. Subscribed.

[@CodingtheFuture-jg1he]

Thank you so much @northofvalhalla5087!

[@helloworldcsofficial]

Great vid! Please go in detail on what new devs should focus on. A video on it will greatly help. Thanks!

[@CodingtheFuture-jg1he]

Thank you for that @helloworldcsofficial! I will be creating some videos focused on new devs very soon. In fact, I'll be starting a Skool.com community as well. So, be on the lookout for that. Curious - where are you in your dev journey and what kind of content would help you?

[@ameerbadri]

Thank you for your sensible and pragmatic view on coding AIs.

[@CodingtheFuture-jg1he]

Thank you so much @ameerbadri!

[@iverbrnstad791]

It's sad, but I think you're right. Overseeing AI creating code will eventually be more efficient than writing it yourself, even if the latter is far more fulfilling work in my opinion...

[@023_shahriarkabir7]

As software engineers, we should first focus on solving the issues that AI tools currently struggle with. Our goal should be to enhance LLMs (Large Language Models) so that they continuously learn from future data, improving their performance and eventually surpassing human capability in these tasks. Data is the most powerful resource, and while humans are limited by our lifespan (rarely exceeding 100 years) AI can learn from thousands of years' worth of data. This gives AI a significant advantage over humans. Ultimately, our role as software developers is to solve problems, whether the solutions come from humans or AI. Isn't it?

[@dranon0o]

> we can use AI to code Python and Javascript with already built framework made by humans!!! Sure it's fine to make few api calls, docs, and few web page rendering Now ask to do the full implementation of a rope algorithm or an reasonable optimized graph library in C lmao Nah Truth is it's useful for repetitive tasks with more intelligence than scripting it yourself > here lot of code, extract the structs and make a schema based on the properties > ... > done There we go, it useful to replace lot of small tasks that we have to do manually sometimes Of course both sides are kinda wrong, it's a good tool. Alas the hardware and technological limits of our times won't allow the creation of ultra advanced AI. We will dream of it, we will try! Current type of AI will be as important as the invention of the compiler by Hopper, we will make the creation of software less frictional for many!

[@boonkiathan]

let's take AI code generation with a nice helping of salt the code we develop aren't exactly really great algorithms, bug free or productive we do what we need to ship our shit AI will get better and better, let's find our place as humans in control of these new paradigms

[@CodingtheFuture-jg1he]

I really appreciate your feedback @dranon0o! I mostly agree with you here. No doubt - AI is currently limited to small, focused tasks that are pretty easy to explain to it. Probably won't do a great job with the C graph library you mentioned. On the other hand, you have to ask yourself: "How likely is it that someone would really need to implement such a thing?" If it's like 20 people on planet Earth, then really, who cares that AI can't do it? I think we often overestimate the NECESSARY complexity of the vast majority of apps that drive our daily lives and businesses. I've been building enterprise systems for a long time and almost never find any of the devs I work with needing to do that sort of "edge case" work. AI (Claude 3.5 anyway) is really good for the types of microservices that drive the majority of medium to large businesses these days. I'm honestly able to use AI coding assistants to build out a typical microservice, including adding JWT auth, full unit test suite, etc. almost without ever typing any code. Not that I recommend that, but I've done it several times now.

[@TechnoMageCreator]

As someone that has no professional software developer experience (just skills I developed as a hobby in eastern Europe 90's - late 2000 we were born broke so we all became pirates and hackers over there), I had a path in engineering/technician. When gpt 3 came out it was obvious for me what was coming. In recent months I had to learn a lot though. AI can make anyone that puts the time and attention to build enterprise level software. The advantage is as long as your open to learn many idea people will become MVP's without no previous software knowledge as long as they will put the work into it. Now with o1 and what is coming is becoming easier and easier. I don't think it will replace all software developers, is gonna replace the lazy ones and is gonna take the ones that do it out of passion to levels they couldn't even dream of.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @TechnoMageCreator! Exactly! It's not that AI is going to replace software developers any time soon. But now is THE time for anyone who has passion and goals to learn to leverage AI to remove a lot of the friction and increase their odds of success (and accelerate their journey). Here's an example not related to software development from my own life... Recently, I stood up my company. Meant getting clear on my vision, creating a business plan, researching and brainstorming ideas, creating my website, researching corporate laws, filing for my LLC and lots of other stuff, registering with the IRS and the EU for tax purposes and on and on. I used AI in every single step, being careful of course not to plop any PII into any chats in the process. When I'd run into a form that had some inputs that were unclear, I snapped a screenshot and gave it to ChatGPT to guide me. When I was creating my WordPress website and wasn't sure how to get the behavior I was after - again, screenshot, ChatGPT and "how do I do X?". Although I don't have an exact figure on the amount of time it saved me overall, I do know it was very significant (not as in 20-30% - possibly an order of magnitude). And it's not just the time savings. My frustration levels were eased, as I didn't have to Google and read a bunch of docs and try this and that - in that experience, I'd say AI was correct nearly every time.

[@MrMiniPilote]

Glad to see another video. I think you nailed it. AI is a force-multiplier for developers, prompting is just another tool now needed by today's developers. The documentation and check-point uses are huge for AI. Thanks for the content.

[@CodingtheFuture-jg1he]

Thank you so much @MrMiniPilote! Totally agree with your points. I understand that people don't want to learn prompting strategies. But to be honest, it's not rocket science - at least not anymore. And prompting soon becomes a pretty natural and intuitive part of your workflow after spending the time to experiment and learn.

[@roo2592]

I have always, and will always, maintain that AI is a tool. Whether you are using it for creative writing, coding, or whatever. It is a tool. It is not meant to replace human involvement, it is meant to help. That's why Bing and Github call their AIs "copilot", they are not meant to be the captain.

[@CodingtheFuture-jg1he]

Thank you for that @roo2592! Yep, it's just a very powerful tool. Now, will AI eventually reach the point where we can totally automate to the point we can't really yet imagine - absolutely, it will. But we'd better be sure we remain in that captain's chair. The best way to do that is by learning all we can about AI starting now - its strengths, its weaknesses, how to use and control it...

[@SiimKoger]

I am always confused when people say "AI makes mistakes" as if that means it's useless. Humans make WAY more mistakes throughout the day. The only difference is that we adjust ourselves (oftentimes we do it automatically without even consciously thinking of our mistake as mistakes). The solution to this AI problem is to just ask it more questions and try to approach a task from a different angle just as you would if you didn't use AI. Of course one has to know how to program overall to know what is valid and what is bs. My workflow is easily 3 times faster than a few years ago.

[@equious8413]

Assistants like copilot are gamechangers. I'm self taught and have a hard time recalling syntax and remembering implementations. I've coded for years though, so I have a sense of what's possible and how to accomplish it already. Assistants like copilot let me see a proposed implementation then tweak it for my use case. It's easily 10xd what I'm capable of.

[@CodingtheFuture-jg1he]

Thanks for sharing your experience @equious8413! Yep, as long as we have enough of a handle on the programming language (which exist mainly for the benefits of us humans) and its basic syntax that we can review and understand and course-correct what the AI generates, not memorizing every little piece of syntax and being able to recall it as quickly as we do our "native" human language, we'll be good with using AI to assist us.

[@bart2019]

Could AI help converting simple code (like, in Python) to Rust? Dealing with all the problems (borrow checker) that introduces?

[@CodingtheFuture-jg1he]

Hi @bart2019! I try to resist speaking authoritatively on things I don't have direct knowledge of, so take what I'm about to say with a huge grain of salt... If we are in fact talking "simple code", as in a Python function that's maybe a dozen lines of well-written code (especially if it's accurately commented - you could use AI to FIRST add accurate comments), I'd think that Claude 3.5 Sonnet would have no issue converting that to Rust. Large blocks of code - yeah, it'd likely struggle. But most human developers aren't proficient enough in multiple languages to do that translation, much less in the 5 seconds it would take Claude to do it (again, assuming it would do it correctly). The only way to know for sure is to try it though. Find a Python function of code block that you know for sure works correctly, use Claude to first add comments to it (if there are none), THEN feed it to Claude and ask it to convert it to Rust. Run it, test it and see. If you try it, would you mind letting me know how it goes?

[@terminally_lazy]

I use Devin on the regular and it's great. I'd say a person who hasn't coded or is still quite new would/will struggle with Devin. But, I can certainly say, it is not a fluke or hoax. The media blew the original video, etc way out of proportion essentially making claims or putting words in the mouths of the folks at Cognition Labs. I got access about 3 weeks ago, not exactly sure why I was picked to have access, of course I signed up for the wait-list, but never expected to receive an invite.

[@imflyingoverclouds]

They marketed it too agressively.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @Username_Jones! Man, I wish the Devin team would give me access. I submitted a request right after the original demo, then again last week. I'd love to hear first-hand more details of what impact it's having on your work. For now, I can only rely on my own imagination to decide what Devin really might be 😆

[@CodingtheFuture-jg1he]

I think you're absolutely right!

[@taragnor]

What do you actively use it for? I'd love to see an actual AI-enabled workflow working on an actual large project and not just "Write me a basic game of snake from scratch." Is it any good at tracking down bugs? Can it utilize your codebase to tie things together using the appropriate functions or does it just do standalone function snippets?

[@blaisepascal3905]

Thank you for the informative video. I just doubt a bit the learning speed part. It is easy for a beginner to let the Generative AI write the whole code and think, wow look at what I can do... but in reality they learned almost nothing. I teach data science and I see that a lot

[@CodingtheFuture-jg1he]

I really appreciate your feedback @blaisepascal3905! That's why you see me emphasizing over and over in nearly every video: always review the code (or any critical LLM output) AND "do not become overly dependent on AI - you must continue to write code too".

[@JoseBasilio]

"It is unlikely the VCs made that kind of investment without significant due diligence" ... remember Theranos?

[@onyangomwangi3495]

Awesome Viewpoints

[@CodingtheFuture-jg1he]

Thank you so much @onyangomwangi3495!

[@drxyd]

The examples you provided look like a 3-6 hours of work to me. Like you said code generation isn't the hard part so I'm not expecting an LLM to perform much better than a human.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @drxyd! Yep, we devs do tend to hyper focus on writing code. Obviously, working code is necessary. But there's certainly a lot more to software development than that.

[@NaderTheExpert]

I like and agree with video and comments. But for me, Electrical Engineer and garage/mancave hardware project hobbyist, who does not have slightest interest to become software engineer or software developer, AI has been a life saver and game changer. Only god knows how many robotic/automation projects I did not complete because of not being able to have software work. Now, I take care of software by a prompt and 3 clicks... Click 1. Copy ChatGPTgenerated code click 2. Paste the code in IDE, and click 3. Run button.

[@CodingtheFuture-jg1he]

Thanks your sharing that @golestaan2259! Wait - you don't want to become a software developer? I thought that was everyone's dream - kinda like becoming a rock star 😆 Really glad to hear AI's helping you with the stuff you care about!

[@FilterChain]

What do you ask it to do that you can copy paste and it works just like that

[@gp390]

Come and convince my boss after he has seen all that marketing nonsense

[@intrestingness]

Unfortunately, the decision as to whether AI will replace devs wont be up to Devs. It will be up to who is paying them. Secondly, AI doesnt have to be better than a dev. When the dev sitting beside you can do their job and about 65 percent of yours with AI then your job is at risk. A team of 25 devs with 65 percent in productivity boost across multiple kpis has now 10 'extra AI devs'. Its just a question of which and how many real devs to fire. Like i said, whoever write the checks will make an economic decision.

[@Beloved_Digital]

Absolutely useful 😅 sir. Thank you for sharing this guide

[@CodingtheFuture-jg1he]

Thank you so much @Beloved_Digital!

[@paulodomingues4179]

Wonderful analysis. ❤

[@CodingtheFuture-jg1he]

Thank you so much for your kind words @paulodomingues4179!

[@bjorn1761]

Dont see the unittests, only see 1 with very limited scope, besides that it is design-wise over-complex

[@CodingtheFuture-jg1he]

Hello @bjorn1761! Direct quote from the video transcript: "Now I’m not claiming that these tutorial apps are ready for production. They were intended to be used in tutorials and so are a bit incomplete. I’d add more unit tests, for example, but they are highly indicative of many types of apps professional developers build every day. They’re very useful in understanding the capabilities of current AI for code-related tasks." As for the complexity of the design... I've been building these very types of apps for a very long time and can assure you that the controller-service-repository pattern you see used for that Java app is THE pattern most experienced enterprise developers still choose for this scenario - at least in the Java world. This gentleman does a great job of explaining a few benefits to this pattern: www.linkedin.com/pulse/controller-service-repository-pattern-comprehensive-guide-shikder-azicc/ And the Spring framework team (who I consider among the most knowledgeable on this very topic) also promotes its use, as you can see here: spring.io/guides/tutorials/rest Although not necessarily obvious to the inexperienced, there are very solid reasons seasoned architects/engineers choose these layered and decoupled patterns.

[@joelvalim]

you re just to the point. It is not just kind of "bro we have this assistant for code generation and it just kills". Put your feet on the ground. It is just not the code, but about on wich environment this code will run on and idiossincrassies and particularities of the very underlying environment code runs over and how each part of this hole interact which each other makes up the whole figure of an enterprise implementation... It is such more deep that just generate an app... There are years and far levels to dig under and understand and conect and make them cooperate efitiently, each organization being unique, kkkk, till one system can be able to deal with such complexity. Just an AGI with incredible learning and generalization skills to catch the overall stuff able to automate everything, everywhere, any business, anyhow do stuff, whatever be the doing... just uncomensurable and unachievable.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @joelvalim! You got it! I mean, even if AI were twice as good as it is today at generating code, there's a whole lot more to it than just the code. And I can see no way we humans are removed from this loop anytime soon.

[@ericwarncke]

Very interesting!!!

[@dontmindmejustwatching]

9:35 dammmm, that's actually very precise

[@CodingtheFuture-jg1he]

Hey @dontmindmejustwatching! Yeah, my intention is not to trivialize the (mostly) art of writing elegant code, but to hopefully remind us that no customer or company pays us for that. If they did, they'd want to see it, right? They just want to get business done and make money... period.

[@davidboreham]

It'll come for the lawyers first..

[@furqantarique3484]

I am interested in software engineering

[@CodingtheFuture-jg1he]

Hey @furqantarique3484! A little more info. Have you started your journey? What are your goals? How do you see AI fitting into those goals?

[@PtolemySoter]

Surely, each produced code is not fit for customisation required, how much time do you spend understanding the code you receive to analyze ans see if and how you can customize it? Does it worth the time of reading given code from anywhere ? Is it better to do it yourself and keep the AI samples as suggestions in small niches ? I tested 9 AI coding platforms to produce a CUDA C script to present prime numbers between two integers and ALL of them on all versions FAILED to present a working CUDA C script, some models were presenting multiple pages essays to make believe 98 is a prime number . It was total wast of time. All AI coding is based on open source low or bad quality code, do not expect AI coding to solve serious projects for you. The cost of wasted time outweights the possible benefit. Maybe it is good for novices to learn how to waste their time. Real software engineers will be advancing in value, after the job market distraction C-level AI believers will bring to the real world.

[@debunker-dw3mn]

Devin demo is the new Project Milo

[@cubanlincoln1767]

And that's why you become a youtuber

[@CodingtheFuture-jg1he]

Thank you so much @cubanlincoln1767!

[@CaimAstraea]

This code example is dogsh1t simple. For me it gets in the way or leads me down bad paths constantly when dealing with very complex logic. I have to remind it that what they say is wrong and give examples and re-review and review what they output and it's like oops sowie my bad oopsie.

[@CodingtheFuture-jg1he]

Thank you for sharing @CaimAstraea! Sorry to hear that you're having a different experience than some of us. But hang in there - it's going to get more reliable and we'll all learn to break down complex tasks when using AI, just the way we should without AI.

[@jefferystartm9442]

Wow brother man, I really couldn't understand how everyone is urgent at the matter @ hand - bless you sir😂

[@CodingtheFuture-jg1he]

Thank you so much for your kind words @jefferystartm9442! Even if it's not urgent today, because it's such a new way of thinking for us all and takes so much time to master and integrate - AND because this tech is moving at lightning speed - waiting to get started on this "wave" is really risky.

[@emanuelec2704]

o1-preview entered the chat

[@CodingtheFuture-jg1he]

Thank you for that @emanuelec2704! Got an experience with 01 you'd like to share? Haven't had time to really put it through its paces.

[@snorman1911]

It's OK.

[@dennissdigitaldump8619]

Yeah, I've tried it. It can build a good program in a single domain. Now then I pushed it, "a gui based program on 3 OS's, with a single codebase, minimal changes". Yeah it has the same problem a human does, aka QT, GTK, Tkinter, etc is not truly universal.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @dennissdigitaldump8619! Yep, QT and Tkinter may well be a blind spot for many LLMs. When you tried that, did you use Claude 3.5? And did you have the AI assistant (or Claude.ai) build the app up iteratively, just the way you would? Or did you try having it generate all the code in just 1-2 shots? If it was the latter, I'd encourage you to go back and try having it first scaffold the project, then just add feature (better, partial feature) X, test, the Y, etc.

[@afzalhussain605]

When I see AI and the hype, it reminds me of VB6 bubble and burst leaving developers reeling from Microsoft's decision to move onto .NET platform. Microsoft had a legit reason, so the AI trained on public data is not as well trained as production code hidden behind firewalls. AI is as good as the source it's trained from. I see AI is going to be part of coding practices as it's going to aid but not replace software engineers and sorts.

[@machine1st]

I use AI to document code which I write.

[@CodingtheFuture-jg1he]

Hey @machine1st827! Yep, that's one thing AI's really good at consistently. In fact, in almost every experience I've had when I've asked Claude 3.5 to add doc strings, etc. to an existing function or method, it does a far better job than almost any developer, including myself, would take the time to do.

[@Rami_Zaki-k2b]

I think you need to review your video. AI is already main stream in coding.

[@CodingtheFuture-jg1he]

Interesting perspective @ramielkady938! What's your standard for "main stream"? And your source for this analysis? Since we don't talk about AI in scenarios not related to software engineering on this channel, I must assume you're saying that AI has already become mainstream for software dev? We all know that AI has long been embedded in many products and services we use, such as Amazon (e.g. recommendation engine), social platforms, etc. So, in a way, AI in the broader sense is mainstream, but both industry reports, regular interactions with other senior engineers at various companies and feedback from places such as this channel all tell me the same story: there are many times more software engineers out there who've never even touched an AI coding assistant. Maybe they're toyed with ChatGPT to craft a bit of code, but that wouldn't meet my personal definition of mainstream adoption for software dev.

[@Rami_Zaki-k2b]

@@CodingtheFuture-jg1he Junior developers as well as seniors are eating Cursor AI raw, this ship has sailed ...

[@AbhijitKrJha]

Every point made in this video is brilliant. What people fail to realize is AI does not need to be perfect, it just needs to assist 4 people do the task of 5 people and 20% jobs are over till new requirements or areas of development open up. I am being very conservative here in estimates as my personal instincts are more like 1 or 2 people doing job of 5 taking into account just current job requirements. Only silver lining is with new tech, client requirements will also get enhanced needing the replaced people back in workforce, only problem will be the transition phase. At least this is what i have come to conclude with my own 15+ years of software industry experience.

[@JoshMarom-zx9mv]

How dare you! Bringing all this neuance and common sense to this discussion...

[@CodingtheFuture-jg1he]

LOL @JoshMarom-zx9mv! I know, crazy right!? I mean, the real world demands extremism, not this boring middle-of-the-road stuff 😴 😆

[@petrusboniatus]

In my opinion the python code showed is just boilerplate you should not have written 90% of the lines showed on screen.

[@CodingtheFuture-jg1he]

Thanks for your feedback @petrusboniatus! Which specific lines? And what's the functionality those lines are providing?

[@petrusboniatus]

​@@CodingtheFuture-jg1he More than give you a concrete line let me show you a similar script I have, but just takes four lines: ```bash # basher is just codellama with my personal prompt ol () { command=$(ollama run basher "{\"question\": \"$1\"}" | jq -r '.command') echo "The following command will be executed: '$command'" read -p "Do you want to proceed? (yes/no) " yn if [ "$yn" == "yes" ]; then eval $command; fi } ``` This might not have all the functionality, but does the same job without needing a python environment and can be understood with just one glance. Also as a general observation I usually find that when AI is able to write X you can import/copy from a library that does X or X is just boilerplate that you can abstract away in modern languages.

[@petrusboniatus]

@@CodingtheFuture-jg1he More than give you a concrete line let me show you a similar script I have, but just takes four lines: # basher is just codellama with my personal prompt ol () { command=$(ollama run basher "{\"question\": \"$1\"}" | jq -r '.command') echo "The following command will be executed: '$command'" read -p "Do you want to proceed? (yes/no) " yn if [ "$yn" == "yes" ]; then eval $command; fi } This might not have all the functionality, but does the same job without needing a python environment and can be understood with just one glance. Also as a general observation I usually find that when AI is able to write X you can import/copy from a library that does X or X is just boilerplate that you can abstract away in modern languages.

[@petrusboniatus]

More than give you a concrete line let me show you a similar script I have, but just takes four lines: ```bash # basher is just codellama with my personal prompt ol () { command=$(ollama run basher "{\"question\": \"$1\"}" | jq -r '.command') echo "The following command will be executed: '$command'" read -p "Do you want to proceed? (yes/no) " yn if [ "$yn" == "yes" ]; then eval $command; fi } ``` This might not have all the functionality, but does the same job without needing a python environment and can be understood with just one glance. Also as a general observation I usually find that when AI is able to write X you can import/copy from a library that does X or X is just boilerplate that you can abstract away in modern languages.

[@charliesumorok6765]

Macros are more powerful than AI. They are underused because lisps look strange.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @charliesumorok6765! I can neither agree nor disagree. I have almost no experience using macros, at least not to any meaningful level. But I realize technological history is littered with misunderstood technologies that had more promise than we ever realized.

[@kittenhero568]

If your code is generated by AI, then you dont have rights over that IP

[@CodingtheFuture-jg1he]

Thanks for your feedback @kittenhero568! Curious... what's your information source for that? Serious question, not trying to be rude - are you an IP attorney? I admit a few people have raised this concern, but I've yet to see anything solid on this front that says we should be worried.

[@kittenhero568]

@@CodingtheFuture-jg1he you can find articles by WIPO on AI and copyright. Traditionally, copyright is only granted to works created by human authors with a sufficient degree of originality. Ultimately it will depend on the degree of contribution of human authors to the work If you have a codebase that is "entirely generated by AI", it would be at least ethically if not legally dubious to claim copyright ownership over that. No, I'm not an IP attorney, and even if I was, it would still depend on jurisdiction and be decided on a case by case basis

[@tryh4rd999]

So, software will be cheaper and faster to produce. Therefore, software engineers salaries will drastically drop, no?

[@CodingtheFuture-jg1he]

Hi @tryh4rd999! I've answered some similar questions in previous comments. Try searching the comments for more details on my thoughts in this area. Short version: for software engineers who can "think like software engineers" vs coders who only learned to write code, I still think the future is bright (at least for the next decade). The major caveat here is: IF they embrace AI and master it for the things it's good at. I think those folks will likely be in high demand very soon. In my head, the jury is out for pure coders who haven't learned the art of critical thinking and true problem solving. I think those folks could be in some real trouble within the next 2 years if they can't/won't work on their problem-solving skills. Not sure about the salaries. But, I think if you're have great critical thinking skills, solid engineering chops and master using AI... I'd expect those folks will be in almost a "name your price" situation soon. That's all just my best guess really. The world is so unpredictable right now. All we can do is make our best judgement calls, then do the work to prepare.

[@OriginalRaveParty]

For the journeyman coder, it'll be eating your lunch way sooner than a Maths Olympiad FANG dev. World class coders are few and far between, and likely not that worried yet, but the mere mortals should be watching it closely. It's certainly not there yet, but it's also not unreasonable to expect that it could be within the next 5 years.

[@marvin_hansen]

True, the main responsibility of professional engineering is delivering business value. I’ve tried a lot of different coding Ai assistants since 2019 and I have seen a noticeable uptick in capabilities over the past 6 months alone. Currently, my daily driver is Codium Ai as it does the best job for generating code that requires the least fixing and it generates the best tests that usually run out of the box. Massive time saver every day. I would like to add that Ai mostly shifts left meaning the focus of my engineering becomes so much more design and architecture while leaving the bulk of implementation to Ai. It really is the emerging age of human augmentation.🎉

[@CodingtheFuture-jg1he]

Thank you so much @marvin_hansen! You have nailed it, IMHO! Your "shift left" take is spot on - just shifting our focus, not at all replacing us. Now, having said that, I do think that a lot of folks who choose to "resist AI to the bitter end" could find themselves in a tight spot in a year, maybe 2. The good news is that there's still time and the market will have its way and people adapt, whether they want to or not. Also, Claude 3.5 is so much better than any LLM before it, at least at code-related tasks. Not sure why many are saying "OK, but that's as far as we can go". I heard that same thing from people when GPT 4 Turbo was released. I haven't tried Codium, but have heard good things about it, especially when it comes to testing.

[@hackmedia7755]

I'm going to replace executives really soon by creating profit-sharing companies and employee-owned companies.

[@dasaauploads1143]

Codemonkey is more productive, codemonkey doesn't needs peers anymore...

[@TheHellishFrog]

AI is very-very resource hungry tech and thus - prohibitively expensive. All companies like OpenAI are operating at huge loss. LLM that is capable of mostly correct guessing of the right combination of symbols most of the time, is actually far more expensive to operate that human programmer. LLM does not understand a thing about the code, it makes guesses based on statistics. Sometimes it guesses right, even brilliant, sometimes dead wrong... but always at loss. Silicon based chips aren't capable to operate cheaply, at the scale needed to make those guesses economically feasible. Photonic chips have their own set of problems, and by far less common and studied. So - no hope for the AI, otherwise human programmers were replaced already, long ago replaced, because AI/NN/ML aren't that new.

[@lasselasse5215]

Well, In the early days of Windows Microsoft lost a lot of money from piracy. But these pirates were de facto ambassadors helping Microsoft grow. Easily justified as marketing cost before what was to come. When corporations get out from their conservative AI-sleep and realize the true potential, they will want to pay. It's even something nations would have strategic interest in, reducing taxes etc for LLM providers, supporting energy cost, funding citizens etc

[@TheHellishFrog]

@@lasselasse5215 But what potential? We are getting mindless guesses, no more than that. Some of them are good fit - some are not.

[@lasselasse5215]

@@TheHellishFrog This video shows the potential. From personal experience, LLM's write my code while I get to think more about the big picture, incredible productivity boost. Based on that, I easily see how disruptive this will be to the profession: Much much cheaper to staff projects. Wait and see.

[@TheHellishFrog]

@@lasselasse5215 There is nothing to wait, everyone can google up and see GPU prices for himself. Anyone who has invested into the hype would gladly support you. But the bubble will be deflated soon, as all the bubbles before, starting with Dutch tulip-mania. In reality AI rarely guesses programming code correctly - and always we have to analyse and check what AI did.

[@lasselasse5215]

​@@TheHellishFrog What language & LLM are you using? My experience is that people in general are not very skilled in writing user stories etc, making them ambiguous and filled with holes. If you do that with an LLM it's not a problem for the LLM, it will fill in the gaps and silently make assumptions. But it will become a problem for you because the more it has to assume things, the risk that it will fail increases exponentially. Also, if you're using some obscure or new framework then the LLM won't be well trained.

[@boonkiathan]

i actually don't give a heck for github copilot, or cursor/replit agents but its a big heck for many software projects that are just that regurgitations and repetitions, copypaste from stack overflows but if your work is exactly that and you don't want to move out of your software dev't comfort zone traditional software companies to tackle problems and projects (that value all the understanding of computing layers) that lie ahead then.... one probably should embrace and not beat up on AI today

[@CodingtheFuture-jg1he]

Hi @boonkiathan! Yep, even if AI could only build apps by borrowing code snippets, patterns, designs from a bunch of different repos it was trained on, there are tons (I'd say most) of valid cases for using it. I've worked in enterprise dev across many industries for the better part of 29 years and I can tell you that almost every new enterprise app is a variation of other apps that have already been created many, many times. Most of us are not really creating what most people would refer to as "novel solutions". It's more like we're borrowing proven design patterns and code that's worked in the past. "Novelty" is far less valuable in the real world than of us like the think.

[@mongi244]

20 years from now

[@m12652]

It's a band-wagon and there's a lot of people trying to jump on 🙄

[@alexforget]

AI coding is just getting started, I think AI will blast past what human are able to do shortly. I think it will take one to five year. By that point it's all jobs, including manual job replaced by robotics. What is missing at this point is a consciousness loop, a self simulation and be sure, this is coming. My concerns are way more around alignements, global control and meaning for human that all become useless. I have been developing software for 20 years and I can tell you we don't have for long anymore.

[@newtanagmukhopadhyay4716]

damn i just entered this industry just a year back, seems i need to do something else now! maybe would open up food business somehow.

[@CodingtheFuture-jg1he]

I really appreciate your feedback @alexforget! I'm not saying you're wrong about your five-year projection on AI's ability to do most cognitive tasks about as well as a human. But I'm nowhere close to saying that we human engineers are going to be replaced anytime soon. I think, as one commenter put it, it's more likely to be a "shift left" scenario, in which we can apply more of the skills that humans are better at and that require more "intuition" and breadth of real experience to make good judgement calls. And I also don't think as a society we're going to take the approach of "well, now that AI can do a bunch of stuff we had to do before, we'll just do the same stuff with less people". I say that because we've already undergone many major technology transformations that have made us WAY more productive. And we just move the bar - we expect more and more output. In actuality, today we put in more work hours than the average person did 50 years ago. But, who knows?

[@alexforget]

@@CodingtheFuture-jg1he Were I live (Canada) I think most of the advanced in productivity have been eaten up by the bureaucratic state. At the speed of acceleration we have I don't think we will have tasks left for humans (all tasks will be better done and faster by computer and machine). what will be left will be art, crafts, things we do for fun. This will completely upend our model of capitalism that is based on scarcity and productivity of individuals. massive meaning crisis. As we speak I program with AI and get it to do in 40sec what would have taken days or week to do by myself. I merely provide the desire, the need and some details about what I want. The bottleneck is already me, reading back and writing back instructions. I think we should focus on alignment and building a positive future with AI way smarter than us and without work.

[@alexforget]

@@newtanagmukhopadhyay4716 Take up architecture, move up the ladder. The coding cog are available for free and are already 100X faster than human. There will be a massive explosion of what we are able to do. I am currently coding a system with AI, it's about 10X faster as it used to be. No more googling or stack overflow.

[@snorman1911]

I use AI extensively for coding and IMO, the better you are at coding the less useful AI is, because it is really bad at solving novel problems (but great at helping speed up mundane tasks). Also, the quality seems to be plateouing. AI won't take over any time soon.