Employers who force their developers to use AI to code.

I was told by someone that in the US this is already a thing. Some companies buys the sub for LLM and then force their developers to use it. (not optional)

They ask them to just use prompt. after prompt, test the code. If it doesnt work, change the prompt or provide it with more information. Rinse and repeat until the LLM provides you a satisfactory code.

And their workload is adjusted based on that. Means 1 dev will be given 3-4 tickets a day to close and they are expected to close it in one day.

if they cant, developers can get scolded for coding manually. When one guy is behind his deadline the project manager will ask him "have you been coding manually? This should only take you 30 minutes using the LLM"

What do u guys think about this practice. I find it hard to believe . wouldnt that lead to crazy tech debt?

My friends in the design industry is also saying their company is practising the same thing, force designers to use AI tools to generate graphics and deploy into production for 1 client, backfired horribly eventually.

AI/LLMs are good assisted tools to complete a very specific tasks, and its never the right person/entity to paint the whole picture.

I am recently building a project that crawls foreign insurance companies website and aggregate them into a Malaysian web app. Their websites are coded in their native language (non common btw), no standards across the input. Some inputs are native words, some are English, even encoded in different ISO format (sigh...). Different Thai websites is built using different frameworks (very dinasour framework mind you).

Navigate these websites require not only specific domain knowledge but also experience in web technologies. Good luck to those who think AI tools can easily build this aggregator app.

my thought as well. I am a heavy user of LLM to code and i dont think its possible. they will do what u tell them exactly and then break other things.

LLM to me is just an auto complete. i only use it if i am lazy to type out the whole code, or i forgot the syntax. after it generates the code i will still eyeball check every line of it.

I always tell my friends: This too shall pass.

LLM is nothing but brute force of deep learning models and GPU power if it cant rationale like a human or understand causality.

Right now we are undergoing a phase where capitalists are riding the LLM opportunity boat without really understand what really is LLM. How many of them really hands-on building LLM themselves? (or have a solid deep learning background).

But it is a nice tool for code autocompletion and summarization, this we as programmer shall adopt or get replaced by younger generation very soon..

It's too early.. maybe in 10 years time would be good though. Maybe 15.

which LLM model are you using TS?

This post has been edited by ipohps3: Feb 9 2025, 04:53 PM

chatgpt , claude

try the new o3-mini high, and prompt it to code your project. it amazes me. there are YouTuber demonstrating it. can search for it.

new AI model like o3, R1 and Gemini2.0 Thinking, all fall into the category of thinking LLM with Agentic AI capabilities. can solve hard problems better than older models, things will only get better forward. adopt or get replaced by younger gen with AI skills.

heck can check my earlier posts.

https://forum.lowyat.net/index.php?showtopic=5507922&hl=

https://forum.lowyat.net/index.php?showtopic=5508234&hl=

This post has been edited by ipohps3: Feb 9 2025, 05:20 PM

already using.

very convenient asking it about design decisions for data intensive app

nice $200 per month investment

-----

This post has been edited by Mat Quasar: Feb 19 2025, 10:49 PM

The price for Pro version(whatever you call it) is $200?

Can it analyse thousands of lines of codes?

for small scale , localised code, blocks of code that can be tested can. Dont ever ask it to write integrated code.

do it too often without review and you accumulate tech debt

This post has been edited by 15cm: Feb 11 2025, 10:40 PM

AI can only further improve from here. it won't stagnant as tech progresses. just be prepared.

My company subscribed to Copilot and I've been using it frequently.

Purely rely on AI to resolve your issue is just nut, they can give very good advice in small piece of code because AI only able to view very little part of your application, they have no knowledge about the entire system architecture (I believe most of the engineer also don't), they often breaking other components, or simply don't have enough code coverage to fully understanding the issue.

With the AI capability today it still require engineer attention to seek the root cause, AI did improve the work a lot (especially on research or exploring tech stack that you are not familiar), but thinking AI could just solve all the issue magically is just nut.

This post has been edited by ragk: Feb 12 2025, 03:24 PM

exactly my experience lol

btw microsoft copilot is the worst of all LLMs. use claude instead, much better

This post has been edited by 15cm: Feb 12 2025, 05:31 PM

So far so good on my end, maybe becoz I'm using paid version? I heard behind the system copilot is using chatGPT.
And I can integrate it in VSCode/VS Studio, i can toss multiple files easily (although most of the time it went nut lol) or just highlight the code and ask my question.

This post has been edited by ragk: Feb 12 2025, 07:23 PM