Ask HN: AI impact on future developer career | Hacker News Reader

Ask HN: AI impact on future developer career

10 pointsalismayilov36 comments

My friend (who is not in technical field) was asking me if it is good idea that his son chooses a software development career.

If he asked me this question two years ago, I could definitely say that there will be always opportunities for him. Now it looks like it is not that easy to answer this question. We see that the coding performance of LLMs are getting better and better.

If your non-technical friend asks you and what would you suggest?

Do you see in your circle that the demand for junior developers are decreasing?

View on HN

The best field to go to is electrical engeering with a focus on computer engineering (EE/ECE), and I say that as an ex Aerospace Engineer who taught himself how to code and worked at startups and big tech.

Most every single talented ECE person can easily get any software job with minimal prep. Half of the leetocde style coding questions are pointer manipulations, the other half rely on some n linked lists, both of which you get a lot of exposure to when working with low level algorithms.

Furthermore, in the scope of AI, we aren't really close to AGI, and even if we were, the power draw of compute is still quite large. There is a lot of progress to be made in making the compute more accessible to average person.

Qualified agree. Personally, I knew since I was a teenager that if I wanted to end up in software, it would be "easy" for me (as in it would take time and practice, but I could do it fully self-taught - I just love doing it too much).

So instead of majoring in CS I majored in EE and mathematics. I did a lot of hard math and physical sciences stuff most of my peers didn't, and it still buys me an unusually large amount of latitude in the kinds of work people can trust me to take on.

what do you think about studying embedded systems engineering? i got accepted for august. so far, finding a team in this job market has been challenging so i'm forced to think longer term.

All my devs who try to use ChatGPT or similar tools end up wasting more time that those who just jump into the code and try to work everything out themselves. How long till this balance changes is not known to me.

GPT-4 is able to do Leetcode with well over 90% accuracy, and all you have to do is give it a screenshot of the question. It takes another minute or so to find more optimal solutions. It's not the best, but if the entire value of a junior is solving Leetcode Hard questions, then ChatGPT is going to it faster and cheaper.

If you can phrase all programming in the style of a Stack Overflow or Leetcode problem, AI will solve it easily. But that's what's slower than jumping in. A middle ground is writing tests.

It does really badly on Swift and is benchmarked on Python. So there's still blind spots. It can't do frameworks like Ren'Py either.

> GPT-4 is able to do Leetcode with well over 90% accuracy, and all you have to do is give it a screenshot of the question.

This is true in the way that Google Search is able to solve Leetcode problems.

GPT-4 struggles with even simple real world codebases.

I did a small survey around my circle and I got response that I did not believe. The majority of developers are using LLMs at least 10-20% time. There are some developers that more than 40% of their code are written by LLMs.

It would be great if stackoverflow and/or similar companies do the developer survey in the industry.

I think 20% is easily believable. LLM may not do more complex functions or work with your data, but a lot of the busywork of coding is done almost instantly. Just not having to look through library docs or sort through StackOverflow is already worth 10%.

I don't think LLM will replace software engineers anytime soon, as a whole, but removing the need for the bottom 10% is not a stretch at all. That has massive implications for salary, imo, except for the top end.

> Do you see in your circle that the demand for junior developers are decreasing?

I expect the demand for junior developers to increase exponentially in the next years when the market adapts to the new AI tools as more than ever, every business will be a tech business.

While LLMs can appear to write code, they can not execute, maintain, or design the systems that use this code.

The actual "code" was never the hard part to begin with.

Developing software is not going away soon. LLMs only addresses a small part of what it means to develop software.

If we think about the demand and supply of the market, since developers will spend less time writing code, they will have more time to solve other parts of a project. This overall increase in productivity could eventually decrease the number of developers needed. In the past, if a project needed five developers, it could eventually be done by two developers. Hence, the market will need less developers than before.

the cost of fixing and maintaining the AI generated code will turn out to be enormous and a lot of code will need to be thrown away and re-written from scratch by experienced developers

But maybe the quality of the code will be very good in the long run. Why do you think that developer will write better code than LLMs in the long run?

why do you think LLMs will achieve actual intelligence and reasoning matching that of humans?

It was only seven years ago that Google introduced “Transformers,” and today we see a completely different level of LLMs. Do you think this development will stop? Maybe it won’t be LLMs alone that do reasoning better than humans, but five years ago, nobody expected that we would have today’s performance.

the need for less developers will actually increase the demand because it will be easier for smaller companies or business not related to tech to have a software department.

I don’t know anyone who’s looking for junior devs today. That was not the case 3 years ago.

Re LLMs: most people responding here seem to assume there will be zero progress in ML after today. I don’t understand why. It’s likely we will get gpt-5 later this year, followed by Opus 3.5 and Ultra 1.5. I expect all three to be significantly better at coding than current models. Again, all three are expected within the next 6 months. Next gen after that (gpt-6 level): 2025-2026. Again, I don’t see any reasons not to expect further improvements. At that point (2026 at the latest) it will be strange to pay humans to write code, at least in the typical way we view SWE role today.

define “better at coding” and also estimate cost of fixing generated solutions

“Better at coding” as defined by seniority levels: GPT4 - an intern; GPT5 - a junior developer; GPT6 - a senior developer.

Yes, GPT5 will do a lot of damage, as it will be the first model replacing actual human coders. But GPT6 should be able to fix most of the issues. During this transition period I expect human senior developers mostly doing code reviews.

By this logic, Photoshop 6 or 7 should have replaced designers.

It will, as soon as it becomes “powered by GPT6”.

i have seen no proof of actual reasoning or intelligence

I think this is not actually a reasonable marker to compare against in this case.

Neither of these are actually needed, all that is needed is to be able to disrupt the pipeline of jobs that interns/junior developers would normally build their initial skills on before moving on to more senior positions. Cutting these entry positions out means continually fewer replacements, and less talent (a deflationary trend, the limited number of positions acts as a filter, and its not necessarily merit based).

Without replacement, eventually older more senior developers will age out, retire, and then everyone will panic but by then it will be far too late (its already cascaded).

If it is not economical to develop or maintain a skill set, the knowledge involved becomes lost knowledge when that population dies out (as we all do given sufficient time).

A perfect corollary is the repair of electronics in the US. There might be 1 person per population of 100-200,000 that goes into business for this sector. If they get hit by a bus, its no longer locally available, and mortality happens to us all.

This is one of the main structural problems with burning bridges in blind marches towards some idealized progress.

How can you go back when all your competitors do the same thing and its no longer economic to run the business except by following the same doom loop as your competitors.

Have you seen at leas one tool that is capable to replace a junior dev? At least one please. I would love to read anything about it.

No, GPT5 is not out yet. As I said, it should be out within 6 months. Please be patient.

Do you think it will be capable to replace a junior dev?

Yes. It will go like that: “should we hire a senior dev and a junior dev? Nah, a senior dev plus GPT5 subscription should be enough to do the workload”. And it will be enough.

I think you're wrong honestly. I think the nature of Software Engineering is something else. Automation and increasing efficiency. All the tools are for this. AI is just a tool. Basically it will increase expectations for all of the levels (Juniors to Principals). There won't be less roles. It's just SWE will be required to do more.

I really hope I’m wrong. Can you imagine the massive unemployment and social chaos if I’m not? I’m rarely wrong about AI though. For example, five years ago I predicted (in a comment here on HN) Turing test will be passed within two years. People called me bat shit crazy, said “in 50 years“, “not in my lifetime”, only one said “in 10 years”. GPT3 happened a year later, and after GPT3.5 people suddenly forgot about the importance of TT as a milestone. I guess it no longer requires “AI”.

I’m a Principal level SWE, and a team lead. I have junior devs on my team. I only need a similar increase of intelligence from GPT4 as 3.5–>4.0 to replace them.

It's strange. I'm in a similar position. I need more of them with more of a tools like GPT-N-whatever just to achieve more. Maybe your market niche is a bit small that you'll get to the ceiling and no development needed. Which is ok.

Anyways I hope you're right to be honest. And let's see.

Why do you hope they are right?

Well.. Its a combination.

I would love to the industry to have a higher entrance barrier.

I would love to be able to solve more with less.

I would love industry to deliver more value with a higher speed.

Progress is inevitable. So I would love to start adapting faster and be on top of it even if it will require me to leave my current role and maybe switch to another one.

I've seen too many people here in IT doing the things they don't like or coming to the job simply because "it pays bills". I've seen 25 yoe engineers who know literally nothing because they were doing literally nothing for the last 20 years. I've seen junior developers who barely know fundamentals and I'm speaking of some really dummy things. I hope industry will naturally heal itself even if it'll make me leave or switch.

I dont think I'd suggest a computer science or engineering degree. Maybe something quicker/shorter. being a junior myself (almost 4 years as a data professional) I must say that I lost the will to learn programming deeply. For many reasons: bosses already demanding high productivity knowing that you can use LLMs for faster coding (in the python/sql realm). But ofc knowing the basics of logic and how to ask/prompt helps a lot

IMHO there is a huge misconception in what is the impact of AI. AI will not take engineering jobs. It will help engineers do more. That's it. Engineering is all about automation, performance improvement. Expectations will be "you have a wonderful tool, we expect you to do more".

Don't do it for the money.

Underrated comment of the century.

The insane lure of gob-smacking amounts of money that got thrown around the tech industry in the past decade with FAANGS unfortunately brought in a lot of people who really weren't passionate about computer science in the first place.

By way of contrast, compare your average software developer with an electrical/mechanical engineer. Just as challenging an industry (if not MORE SO in a lot of ways), and yet the pay is usually a fraction of what a software dev can make. Nobody becomes an EE for the money.

Reply on news.ycombinator.com