Hacker News Reader

Comment by omarhaneef | Hacker News Reader

I know you’re trying to see where the puck is headed but I think there is a lot of work between where we are and where LLMs replace 4 Junior devs.

The workflows we have are not quite right for it. Coding has always been 10% coding and 90% debugging but I think the rate at which we generate the 10% will grow exponentially.

This means that the debugging has to grow. We will generate errors at an unprecedented rate.

LLMs trained on previous errors and methods won’t catch them. They’ll be more complicated and spread out over the code.

We need new tools to visualize the code and track errors. I think what it means to be a programmer will change. More testing, thinking and less klocs.

View on HN

> a lot of work between where we are and where LLMs replace 4 Junior devs

The projection from current SOTA is that coding LLMs replace 1 senior dev + 5 junior devs with 1 senior dev + LLM.

Ergo, there will be fewer junior dev positions and senior dev salaries can capture some of the additional value they deliver.

Whether that's offset by the increase in senior supply as junior -> senior conversions happen...?

The two areas that will likely expand demand to absorb those junior devs are:

(1) Integration engineering (i.e. poorly-documented interface to poorly-documented interface)

(2) Testing (because if kloc are cheap, there will be more money to validate -- and it will make people feel safer to have human-in-the-loop before sign-off / production).

Coding LLMs are always going to be better at things well-represented in their training sets. Which is to say, the most popular languages, API styles, apps as of todayish.

So, other things will be the best use of human time.

> The projection from current SOTA is that coding LLMs replace 1 senior dev + 5 junior devs with 1 senior dev + LLM.

Thanks - I needed a good laugh.

You think junior devs, as the positions exist today, are going to be sticky in the face of coding LLMs?

Smart juniors shouldn't be juniors long. I think another angle to think about this is Junior devs that can leverage AI to learn have a much shorter path to Senior than current Seniors had to take (aka the hard way). If as a Junior you know of things to look out for but don't know the implementation details, AI can help you.

For really obscure issues or bugs, I had to scroll through seemingly random StackOverflow results (or Experts Exchange, ha) looking for exact issues that I had, weeding through answers that weren't relevant, things based on older versions, etc. Now you give the context to AI tech stack, data, and your source code, explain the problem with the error message, and ask what might be happening and that shaves a few hours off the old way of trial and error. Even if the 1st answer isn't totally right, a few follow on questions will help narrow it down.

The tradeoff is that searching for answers develops people's ability to do research in general. It's an essential skill to be able to skim though material quickly and decide if it's relevant or not. it's in the same way that calculators save labor on large calculations, but giving up all of one's mathematical intuition in exchange for a better calculator would be decrease in net performance.

The hard way makes for better quality.

I can have 18% BMI with ozempic, or I can have 18% BMI running and lifting weights. "We are not the same" to quote the meme [0].

The LLM-bros can do LLM-bro stuff. The hardened Seniors that have seen/done it all are the ones you need on a crisis.

[0]: https://i.imgflip.com/5yhw7e.jpg

How else are you going to get senior+ level engineers going forward?

Yes. But why would a company today hire a junior dev? If I have limited budget, why hire a junior dev instead of poaching a mid level developer from somewhere else?

A junior dev does negative work. And with corp dev/enterprise dev the difference in price between a junior and a “senior” dev is only about $70K. The gap is even smaller with remote work even if you hire in the US. There are plenty of developers in low cost of living areas who would be glad to work for relative peanuts.

As far as where will the next generation of developers come from? That’s a collection action problem

Sometimes you want them "in 2 years from now" so you 'buy' them now. Also sometimes you want someone to do 3 out of 10 tasks of the job for a couple of years, and then get them to do another 4 out of 10, an ultimately get them to do 10/10 in 6-7 years from now.

I remember those years, I hope they are still around.

In 2 years from now, you’re going to want to promote them but your HR department is not going to give them the raise that would keep them there so they are going to find another job. They are going to watch new people come in in two years making market rates - salary compression.

The average tenure of a developer is 2-3 years. Almost no one stays at a company for 6-7 years

As a manager, the chance of you staying at a job for 6-7 years are slim and staying in the same position is even slimmer. Besides, you are going to be measured by your quarterly or at most yearly “OP1 goals” or equivalent

Ha! That's the rub: we won't. :(

In the same way that we don't get the volume of true hackers that the 70s-90s produced, because of the locked-down computing experiences most kids were introduced to from 00+.

I was going to comment that there will be a need for many IT professionals to fix problems produced by AI. Your comment painted a much clearer picture.

Massively kicking out junior devs is ignorant and short sighted. Critical IT areas are going to implode understaffed by retiring seniors and freefall quality of AI slop retrained on more AI slop.

As GenX/Millenial senior, I am not going to stay long here, no matter what money I am offered. If cornered I would rather sell organic produce, honey, wood work, be handyman to my local community than spending any more days on debugging fruits of MBA-driven AI lunacy.

I am not a super coder but I am quite conscious what I write.

I copy pasted a part of my code into Claude.ai to help me add a few lines. It completed that part and also rewritten another few lines. I asked why it did that, explanation was that I missed a edge case.

LLMs are good for writing boring parts of code and also helpful for catching bugs.

You have to be careful because they will sometimes make mistakes they do that. I’ve had Claude decide to “simplify” my code by removing a fix that took hours to chase down.

Was your fix commented? If not, a human developer refactoring your code could have done the same. I’ve seen it happen more than once.

Reply on news.ycombinator.com