Hacker News Reader

Comment by Rush2112 | Hacker News Reader

I still write code, by hand mostly. Perhaps it’s my sector or something, but I’m always quite surprised when I read things like this.

View on HN

Might really be your sector. Claude is massive in my organization, with no one writing code by hand anymore. The biggest struggle is keeping up with what is being done, reviewing and understanding.

What has your code base ended up looking like? Is there a coherent abstraction hierarchy?

I keep checking back on LLMs every couple months and while they are now great for figuring out bugs in a system I don’t know much about (say in my use of a framework I’ve never used before), it’s been super helpful. But whenever I have it write a lot of code, even in a not huge project, it mostly works but it’s like a gum and popsicle stick machine gun and I’m scared of what large org code bases will look like in a few years.

As someone who makes heavy use of AI tools to write code I also find it surprising. The idea that even the latest models of Claude and Codex are capable of writing sufficiently useful and production usable code is baffling. Unless you are operating at extremely low stakes with few meaningful constraints I don't see how anyone can seriously rely fully on AI generated code. I see the code daily, it certainly can save some time when used properly but there are very hard limits to the amount of concepts that can be affirmatively managed before AI begins to tie itself in knots. The only solution AI is capable of from that point is of course generating more code to sort out any issues. "Hand" writing actual clear, concise and unbloated code is still very much a requirement.

What type of applications are you working on?

Primarily e-commerce and finance, so interacting with a lot of APIs and pretty much everything has real money attached so if things are wrong there are pretty direct costs that can quickly become apparent. There are of course the easy to see issues where things are mispriced or fulfillment errors but there's also the trickier things where errors that cause a lack of exposure can be converted to money lost.

IME, there has definitely been a pull back in the reliance on AI in all aspects of the business, and fortunately there has mostly been caution with the AI in the related software development so not too much of a push to go all in.

I've seen some pretty drastic losses occur from AI transposing things like dates and compatible brands. Then again people have fat fingered million dollar losses before as well.

Many companies are essentially forcing the use of it via raising the bar for performance by 10x(hyperbole). Yeah it’s not great everywhere and I’m sure there firms that aren’t as pushy.

10x really isn't hyperbole at this point

Producing 10x the amount of slop isn't being 10x productive.

Certainly just my experience, but my supervisors totally and holistically believe the two are one and the same.

I don’t think any of our developers have written any code since March/April. I wonder if they even know how to anymore.

>I wonder if they even know how to anymore.

Oh they will, once they get laid off and need to start interviewing again.

Half the people I interview seem to have those AI interview helpers that just plug into the audio directly and give you realtime responses to questions that are asked. Plus they have it running on the same screen as the webcam so it’s near impossible to tell who is cheating.

Have you tried those AI leetcode platforms that proctor the candidate?

> it’s near impossible to tell who is cheating

Ask them to answer with their eyes closed?

The leetcode platforms, that we use, have the candidate on camera while they work on the take home but it’s the same problem. If they have ChatGPT on an alternate laptop it’s not recorded. Recently everyone seems to be getting 100% on every take home.

> Ask them to answer with their eyes closed?

These are senior and principal engineers that we want on our team. I fear being distrusting like this is a terrible first step and might dissuade them from joining.

you are going to be one of the most valuable software engineers in 5 yrs, dont stop doing what you do!!!

porting, tests and scaffolding code, documentation is what ai is quite good at and a human is terrible at

Reply on news.ycombinator.com