Is AI coming for your Developer Job?
Вставка
- Опубліковано 9 лип 2024
- I've been thinking about AI and how it will affect software development. I've been wondering if it will replace developer jobs or if developers will use it to be more productive. I think the answer is both.
🔗 EventStoreDB
eventsto.re/codeopinion
🔔 Subscribe: / @codeopinion
💥 Join this channel to get access to a private Discord Server and any source code in my videos.
🔥 Join via Patreon
/ codeopinion
✔️ Join via UA-cam
/ @codeopinion
📝 Blog: codeopinion.com
👋 Twitter: / codeopinion
✨ LinkedIn: / dcomartin
📧 Weekly Updates: mailchi.mp/63c7a0b3ff38/codeo... - Наука та технологія
Learn more about Software Architecture & Design by subscribing to my newsletter.
🚀mailchi.mp/63c7a0b3ff38/codeopinion
People who think software engineering is just about coding, sure, they will have a hard time competing with AI in the long run.
The trouble with thinking human developers will not be replaced by AI, is that there is almost always a "right" way to design your architecture for your scenario and requirements. And there is nothing stopping AI from also learning that "right" way - or several right ways to make a choice from. If the last year of AI is anything to go by, I think we have about 7 years before we won't be programming at all, we will just be describing the application requirements, and it will write it, provision the hardware, and secure the infrastructure. It is just wishful thinking that humans will always be better at designing a system than a machine that has the ability to learn from every system ever deployed - remember the AI has access to every deployed architecture on Azure, Google, Amazon, etc. Every research paper, every book on design, everything ever put on Github, and all the developer channels on UA-cam, every public website client code, etc. Humans will never compete with that amount of knowledge as individuals.
Reality check: if AI was all that they'd be using for themselves to take over your domain instead of selling it to you.
I have yet to see a shop where jira tickets are so well defined an AI could do it. By that point its practically pseudo-code and you can offshore that for cheap anyway. In practice, this hasn't panned out.
That's my point is that software is often about communication and decisions on what to do, not actually doing it. Code is a means to get there.
That's a pretty solid point, tbh.
Had very similar experience with AI ranging from omg this suggestion is brilliant to well less satisfying results... It's also quite good in quickly explaining API's your not familiar with yet. You do always have to be careful though because it tends to have some of the details wrong and its exactly those finicky details that can make or break your app.
I only code after designing it, most of the time we do research and problem solving. Just like the guy from I robot " im sorry my answers are limited you must ask the right question"
If someone thinks that he can be replaced - he will be replaced. Usually the tickets in Jira looks like "do the shit". Not AI, humans can't understand it. And then you start process of communication, investigation, decision making and so on. When business will be able to explain what they want, maybe one day someone will be replaced. But during my 15 years career, I haven't seen such task. Also don't forget, that AI is taught using data from internet. Really good repositories are usually private.
My experience with copilot in VS has been very inconsistent as well. Sometimes it seems to know exactly what I want and it saves me a few seconds to a few minutes of typing. Other times it generates some nonsense that ends up distracting me since I have to stop and try to make sense of what it's suggesting for me. What's funny to me is it does a better job of guessing my comments than generating code, so I guess it's not a total waste.
Honestly, for me, these AI tools take away the fun part of being a developer. After spending 7 hours refining the tasks, back and forth, blah blah blah, the finally, I can spend an hour or two actually building the thing. Why would I want AI to save me a few keystrokes, when actually building something is the whole payoff.
Same thing with the image generators. Why are we giving AI the tasks that are fun?
How about we give the clients ChatGPT designed for them to write complete user stories through asking questions and interviewing the client?
I have been doing lot of experiments lately with AI. Explaining high level context to it and then taking the steps it gives to solve the problem, then further dividing steps until it reaches actual coding steps. Its not very successful but definitely progressing AND its only a beginning of AI, Its already doing great. Imagine what it can do as it gets better.
If we can give detailed context, add checks and limits, it may surely do decent job. I am personally looking forward to get it better.
I think efficient programmers can still stay for a good long time.
You missed on possible path. It is possible that we will not use code or programing per se at all(or almost, excluding embedded for now).
We will give AI tasks by prompt and AI will do the taks without using code or today computer architecure directly.
It is stars going now, ChatGPT can explain step by step how to do someting, so it also with connection to external services do it too.
I think this is indeed a much more likely path, it will take awhile since the AI must be able to do things without making mistakes. But I do think many years from now computers will fundamentally work differently, programming will just be a natural description of the system and how it should work. Maybe on a functional level at start, but it will probably get on a higher and higher level over time.
Sometimes I ask gpt4 to refactor my code. 50% of the time it indeed improves my code and the other 50% or it doesn’t cover all the paths provided by the original code or the refactor is not that good.
In my opinion AI is far from replace developers.
Man only the fact that you need to be senior to distinguise between types of code quality is the key. If someone cannot separate the false positives even if its is just 1 out of 100 its a potential bug nothing works they will have to hire me to find the issue. End of story.
Of course it is. You will only need 1/5 of the developer resource you need now.
My experience is similar to yours. One thing it does make me a lot better at is reading foreign code and spotting problems with it.
As far as replacing developers, I don't really know. It has to get better before it can start completely replacing developers. It can get better, yes. But it also got to where it is now off the back of a monumental amount of hardware. They aren't going to so trivially be able to 10x their parameter count in the models.
So while I think it's definitely possible, I think the road is going to be more hard fought than going from, for example, chatgpt3 to chatgpt4.
bro, nobody deals with hardware these days anymore. that's low level stuff. everything's in the cloud, infrastructure as code. scaling down or up is down through software. no physical button to press, no wire to connect. nadah!
@@JTMoonXX30I think Bobby was talking about the hardware required to train and run the AI models
It is great interview questions, as soon as I introduce an array or a hashmap I get pretty good answers but for the rest not really…
Even once I asked if a decorator existed in my framework, it said yes!! Ok I spent the next 10 mins looking at the source code and it does it exist… It just came up with a new one xD
EVERY single time I tried to use AI for something non-trivial, it just wasted my time, and I had to do it myself!
To be fair, it might (might) have given a clue I could have used, but all these "look at this new AI coding everything" videos are, IMHO, just crap!
The level AI works at the moment is still a bit on the unintuitive side, you have to know how to properly prompt it or supply it with comments in your code for it to actually work decently. But this will only improve over time making it easier and easier to get good results.
To be honest, not once has AI managed to generate something for me that builds or works on the first try. Not even once. And by that, I don't mean "it skips the parameter names I want it to use", but more like "this thing does not exist at all, anywhere". That's just me, though. Maybe others have different experiences.
No that's pretty much my experience. It's kind of a starting point. My assumption is eventually it will keep getting better where the produced code is closer and closer to what I need.
I see everyone in the comments basing their opinions on the current state of AI. Of course it can't currently replace us. But why do you all seem to assume that it's not going to get better? Given how fast it has been improving, I can see it being able to do everything humans can do cognitively in a few years.
AI makes a good, automated pair programmer to bounce ideas and implementation off of.
My biggest concern is the scanning of licensed code for these new language engines backing data and zero attribution to the creators of said code. Now that it is out there nothing is safe, or you could say even less so then before because you didn't have people stealing code with massive 24/7 search engines scanning everything on the internet. As well as the GIGO issue with all the crappy code out there as well.
I'm less concerned about AI taking my job than I am about my job being outsourced overseas.
I've found it to be a valuable "personal assistant". I think every developer should be spending the $20 for at least the GPT4 access on OpenAi. Our organization has begun buying subs for developers as well. The value is amazing. So, while I don't see it replacing full-stack developers, I do see it as a highly-beneficial, almost necessary, tool. If I were a business owner or CTO, I'd want people in my organization that not only aren't scared of it, but embrace it and take the initiative to explore use cases.
yes, AI will definitely replace programmers who don't think and just rely on business analysts to give them directions what to code. my wife is a ba and she tells me that she works with devs who need very detailed specs otherwise they won't code. those are the programmers who are going to be replaced because my wife will be happy to just tell the AI what she needs to get done and in matter of minutes, the app is deployed and ready for her to test. all in plain english. that's the future!
Unfortunately not, understanding what to ask is a big thing
As long as the code has to comply with safety-critical standards, I cannot rely on code generated by AI. In addition, everything that AI proposes should ultimately be reusable code - and that is better off in libraries and frameworks than copying it into programs everywhere.
"As long as the code has to comply with safety-critical standards, I cannot rely on code generated by AI"
I wouldn't be so dismissive.
If you have an executable specification of your standards, AI systems can use the feedback from them
to try different approaches, as programmers do while coding - with the difference that AI can
generate code faster and remember more precisely the past.
And since code in these domains are involved with LOTS of money, I wouldn't be surprise if
very specific AI-assistants in these domain emerge very very soon and with high quality.
Development job will be harder because there would be less time needed for coding, and more software engineering, so yes, coding would still be there, but less payed imo
Absolutely not. AI is a little smart but not smart enough to manage the multitude of things that goes on with software development. It is not simply about coding. Each system has unique requirements that cannot be trained into AI due to limited training data and lack of market to pay to develop a custom system. Will it speed up development…. Yes, will it replace developers…. Absolutely not