5 ms·

Ask HN: AI impact on future developer career

10 points by alismayilov 2y ago

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?

tobinfekkes 2y ago
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.
- alismayilov 2y ago
  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.
  - meiraleal 2y ago
    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.
  - b20000 2y ago
    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
    - alismayilov 2y ago
      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?
      - b20000 2y ago
        why do you think LLMs will achieve actual intelligence and reasoning matching that of humans?
        alismayilov 2y ago
        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.
meiraleal 2y ago
> 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.
p1esk 2y ago
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.
- b20000 2y ago
  define “better at coding” and also estimate cost of fixing generated solutions
  - p1esk 2y ago
    “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.
    - meiraleal 2y ago
      By this logic, Photoshop 6 or 7 should have replaced designers.
      - p1esk 2y ago
        It will, as soon as it becomes “powered by GPT6”.
    - b20000 2y ago
      i have seen no proof of actual reasoning or intelligence
ghoshbishakh 2y ago
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.
- alismayilov 2y ago
  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.
  - savorypiano 2y ago
    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.
- muzani 2y ago
  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.
  - WgaqPdNr7PGLGVW 2y ago
    > 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.
savorypiano 2y ago
Don't do it for the money.
- vunderba 2y ago
  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.
boredemployee 2y ago
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
ActorNightly 2y ago
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.
- neontomo 2y ago
  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.
- hiAndrewQuinn 2y ago
  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.
cultofthecow 2y ago
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".

10 points by alismayilov 2y ago

tobinfekkes 2y ago

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.

alismayilov 2y ago
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.
- meiraleal 2y ago
  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.
- b20000 2y ago
  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
  - alismayilov 2y ago
    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?
    - b20000 2y ago
      why do you think LLMs will achieve actual intelligence and reasoning matching that of humans?
      - alismayilov 2y ago
        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.

meiraleal 2y ago

> 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.

p1esk 2y ago

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.

b20000 2y ago
define “better at coding” and also estimate cost of fixing generated solutions
- p1esk 2y ago
  “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.
  - meiraleal 2y ago
    By this logic, Photoshop 6 or 7 should have replaced designers.
    - p1esk 2y ago
      It will, as soon as it becomes “powered by GPT6”.
  - b20000 2y ago
    i have seen no proof of actual reasoning or intelligence

ghoshbishakh 2y ago

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.

alismayilov 2y ago
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.
- savorypiano 2y ago
  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.
muzani 2y ago
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.
- WgaqPdNr7PGLGVW 2y ago
  > 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.

savorypiano 2y ago

Don't do it for the money.

vunderba 2y ago
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.

boredemployee 2y ago

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

ActorNightly 2y ago

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.

neontomo 2y ago
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.
hiAndrewQuinn 2y ago
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.

cultofthecow 2y ago

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".